xref: /freebsd/contrib/llvm-project/llvm/lib/Analysis/MemoryDependenceAnalysis.cpp (revision a90b9d0159070121c221b966469c3e36d912bf82)
1 //===- MemoryDependenceAnalysis.cpp - Mem Deps Implementation -------------===//
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 an analysis that determines, for a given memory
10 // operation, what preceding memory operations it depends on.  It builds on
11 // alias analysis information, and tries to provide a lazy, caching interface to
12 // a common kind of alias information query.
13 //
14 //===----------------------------------------------------------------------===//
15 
16 #include "llvm/Analysis/MemoryDependenceAnalysis.h"
17 #include "llvm/ADT/DenseMap.h"
18 #include "llvm/ADT/STLExtras.h"
19 #include "llvm/ADT/SmallPtrSet.h"
20 #include "llvm/ADT/SmallVector.h"
21 #include "llvm/ADT/Statistic.h"
22 #include "llvm/Analysis/AliasAnalysis.h"
23 #include "llvm/Analysis/AssumptionCache.h"
24 #include "llvm/Analysis/MemoryBuiltins.h"
25 #include "llvm/Analysis/MemoryLocation.h"
26 #include "llvm/Analysis/PHITransAddr.h"
27 #include "llvm/Analysis/TargetLibraryInfo.h"
28 #include "llvm/Analysis/ValueTracking.h"
29 #include "llvm/IR/BasicBlock.h"
30 #include "llvm/IR/Dominators.h"
31 #include "llvm/IR/Function.h"
32 #include "llvm/IR/InstrTypes.h"
33 #include "llvm/IR/Instruction.h"
34 #include "llvm/IR/Instructions.h"
35 #include "llvm/IR/IntrinsicInst.h"
36 #include "llvm/IR/LLVMContext.h"
37 #include "llvm/IR/Metadata.h"
38 #include "llvm/IR/Module.h"
39 #include "llvm/IR/PredIteratorCache.h"
40 #include "llvm/IR/Type.h"
41 #include "llvm/IR/Use.h"
42 #include "llvm/IR/Value.h"
43 #include "llvm/InitializePasses.h"
44 #include "llvm/Pass.h"
45 #include "llvm/Support/AtomicOrdering.h"
46 #include "llvm/Support/Casting.h"
47 #include "llvm/Support/CommandLine.h"
48 #include "llvm/Support/Compiler.h"
49 #include "llvm/Support/Debug.h"
50 #include <algorithm>
51 #include <cassert>
52 #include <iterator>
53 #include <utility>
54 
55 using namespace llvm;
56 
57 #define DEBUG_TYPE "memdep"
58 
59 STATISTIC(NumCacheNonLocal, "Number of fully cached non-local responses");
60 STATISTIC(NumCacheDirtyNonLocal, "Number of dirty cached non-local responses");
61 STATISTIC(NumUncacheNonLocal, "Number of uncached non-local responses");
62 
63 STATISTIC(NumCacheNonLocalPtr,
64           "Number of fully cached non-local ptr responses");
65 STATISTIC(NumCacheDirtyNonLocalPtr,
66           "Number of cached, but dirty, non-local ptr responses");
67 STATISTIC(NumUncacheNonLocalPtr, "Number of uncached non-local ptr responses");
68 STATISTIC(NumCacheCompleteNonLocalPtr,
69           "Number of block queries that were completely cached");
70 
71 // Limit for the number of instructions to scan in a block.
72 
73 static cl::opt<unsigned> BlockScanLimit(
74     "memdep-block-scan-limit", cl::Hidden, cl::init(100),
75     cl::desc("The number of instructions to scan in a block in memory "
76              "dependency analysis (default = 100)"));
77 
78 static cl::opt<unsigned>
79     BlockNumberLimit("memdep-block-number-limit", cl::Hidden, cl::init(200),
80                      cl::desc("The number of blocks to scan during memory "
81                               "dependency analysis (default = 200)"));
82 
83 // Limit on the number of memdep results to process.
84 static const unsigned int NumResultsLimit = 100;
85 
86 /// This is a helper function that removes Val from 'Inst's set in ReverseMap.
87 ///
88 /// If the set becomes empty, remove Inst's entry.
89 template <typename KeyTy>
90 static void
91 RemoveFromReverseMap(DenseMap<Instruction *, SmallPtrSet<KeyTy, 4>> &ReverseMap,
92                      Instruction *Inst, KeyTy Val) {
93   typename DenseMap<Instruction *, SmallPtrSet<KeyTy, 4>>::iterator InstIt =
94       ReverseMap.find(Inst);
95   assert(InstIt != ReverseMap.end() && "Reverse map out of sync?");
96   bool Found = InstIt->second.erase(Val);
97   assert(Found && "Invalid reverse map!");
98   (void)Found;
99   if (InstIt->second.empty())
100     ReverseMap.erase(InstIt);
101 }
102 
103 /// If the given instruction references a specific memory location, fill in Loc
104 /// with the details, otherwise set Loc.Ptr to null.
105 ///
106 /// Returns a ModRefInfo value describing the general behavior of the
107 /// instruction.
108 static ModRefInfo GetLocation(const Instruction *Inst, MemoryLocation &Loc,
109                               const TargetLibraryInfo &TLI) {
110   if (const LoadInst *LI = dyn_cast<LoadInst>(Inst)) {
111     if (LI->isUnordered()) {
112       Loc = MemoryLocation::get(LI);
113       return ModRefInfo::Ref;
114     }
115     if (LI->getOrdering() == AtomicOrdering::Monotonic) {
116       Loc = MemoryLocation::get(LI);
117       return ModRefInfo::ModRef;
118     }
119     Loc = MemoryLocation();
120     return ModRefInfo::ModRef;
121   }
122 
123   if (const StoreInst *SI = dyn_cast<StoreInst>(Inst)) {
124     if (SI->isUnordered()) {
125       Loc = MemoryLocation::get(SI);
126       return ModRefInfo::Mod;
127     }
128     if (SI->getOrdering() == AtomicOrdering::Monotonic) {
129       Loc = MemoryLocation::get(SI);
130       return ModRefInfo::ModRef;
131     }
132     Loc = MemoryLocation();
133     return ModRefInfo::ModRef;
134   }
135 
136   if (const VAArgInst *V = dyn_cast<VAArgInst>(Inst)) {
137     Loc = MemoryLocation::get(V);
138     return ModRefInfo::ModRef;
139   }
140 
141   if (const CallBase *CB = dyn_cast<CallBase>(Inst)) {
142     if (Value *FreedOp = getFreedOperand(CB, &TLI)) {
143       // calls to free() deallocate the entire structure
144       Loc = MemoryLocation::getAfter(FreedOp);
145       return ModRefInfo::Mod;
146     }
147   }
148 
149   if (const IntrinsicInst *II = dyn_cast<IntrinsicInst>(Inst)) {
150     switch (II->getIntrinsicID()) {
151     case Intrinsic::lifetime_start:
152     case Intrinsic::lifetime_end:
153     case Intrinsic::invariant_start:
154       Loc = MemoryLocation::getForArgument(II, 1, TLI);
155       // These intrinsics don't really modify the memory, but returning Mod
156       // will allow them to be handled conservatively.
157       return ModRefInfo::Mod;
158     case Intrinsic::invariant_end:
159       Loc = MemoryLocation::getForArgument(II, 2, TLI);
160       // These intrinsics don't really modify the memory, but returning Mod
161       // will allow them to be handled conservatively.
162       return ModRefInfo::Mod;
163     case Intrinsic::masked_load:
164       Loc = MemoryLocation::getForArgument(II, 0, TLI);
165       return ModRefInfo::Ref;
166     case Intrinsic::masked_store:
167       Loc = MemoryLocation::getForArgument(II, 1, TLI);
168       return ModRefInfo::Mod;
169     default:
170       break;
171     }
172   }
173 
174   // Otherwise, just do the coarse-grained thing that always works.
175   if (Inst->mayWriteToMemory())
176     return ModRefInfo::ModRef;
177   if (Inst->mayReadFromMemory())
178     return ModRefInfo::Ref;
179   return ModRefInfo::NoModRef;
180 }
181 
182 /// Private helper for finding the local dependencies of a call site.
183 MemDepResult MemoryDependenceResults::getCallDependencyFrom(
184     CallBase *Call, bool isReadOnlyCall, BasicBlock::iterator ScanIt,
185     BasicBlock *BB) {
186   unsigned Limit = getDefaultBlockScanLimit();
187 
188   // Walk backwards through the block, looking for dependencies.
189   while (ScanIt != BB->begin()) {
190     Instruction *Inst = &*--ScanIt;
191     // Debug intrinsics don't cause dependences and should not affect Limit
192     if (isa<DbgInfoIntrinsic>(Inst))
193       continue;
194 
195     // Limit the amount of scanning we do so we don't end up with quadratic
196     // running time on extreme testcases.
197     --Limit;
198     if (!Limit)
199       return MemDepResult::getUnknown();
200 
201     // If this inst is a memory op, get the pointer it accessed
202     MemoryLocation Loc;
203     ModRefInfo MR = GetLocation(Inst, Loc, TLI);
204     if (Loc.Ptr) {
205       // A simple instruction.
206       if (isModOrRefSet(AA.getModRefInfo(Call, Loc)))
207         return MemDepResult::getClobber(Inst);
208       continue;
209     }
210 
211     if (auto *CallB = dyn_cast<CallBase>(Inst)) {
212       // If these two calls do not interfere, look past it.
213       if (isNoModRef(AA.getModRefInfo(Call, CallB))) {
214         // If the two calls are the same, return Inst as a Def, so that
215         // Call can be found redundant and eliminated.
216         if (isReadOnlyCall && !isModSet(MR) &&
217             Call->isIdenticalToWhenDefined(CallB))
218           return MemDepResult::getDef(Inst);
219 
220         // Otherwise if the two calls don't interact (e.g. CallB is readnone)
221         // keep scanning.
222         continue;
223       } else
224         return MemDepResult::getClobber(Inst);
225     }
226 
227     // If we could not obtain a pointer for the instruction and the instruction
228     // touches memory then assume that this is a dependency.
229     if (isModOrRefSet(MR))
230       return MemDepResult::getClobber(Inst);
231   }
232 
233   // No dependence found.  If this is the entry block of the function, it is
234   // unknown, otherwise it is non-local.
235   if (BB != &BB->getParent()->getEntryBlock())
236     return MemDepResult::getNonLocal();
237   return MemDepResult::getNonFuncLocal();
238 }
239 
240 MemDepResult MemoryDependenceResults::getPointerDependencyFrom(
241     const MemoryLocation &MemLoc, bool isLoad, BasicBlock::iterator ScanIt,
242     BasicBlock *BB, Instruction *QueryInst, unsigned *Limit,
243     BatchAAResults &BatchAA) {
244   MemDepResult InvariantGroupDependency = MemDepResult::getUnknown();
245   if (QueryInst != nullptr) {
246     if (auto *LI = dyn_cast<LoadInst>(QueryInst)) {
247       InvariantGroupDependency = getInvariantGroupPointerDependency(LI, BB);
248 
249       if (InvariantGroupDependency.isDef())
250         return InvariantGroupDependency;
251     }
252   }
253   MemDepResult SimpleDep = getSimplePointerDependencyFrom(
254       MemLoc, isLoad, ScanIt, BB, QueryInst, Limit, BatchAA);
255   if (SimpleDep.isDef())
256     return SimpleDep;
257   // Non-local invariant group dependency indicates there is non local Def
258   // (it only returns nonLocal if it finds nonLocal def), which is better than
259   // local clobber and everything else.
260   if (InvariantGroupDependency.isNonLocal())
261     return InvariantGroupDependency;
262 
263   assert(InvariantGroupDependency.isUnknown() &&
264          "InvariantGroupDependency should be only unknown at this point");
265   return SimpleDep;
266 }
267 
268 MemDepResult MemoryDependenceResults::getPointerDependencyFrom(
269     const MemoryLocation &MemLoc, bool isLoad, BasicBlock::iterator ScanIt,
270     BasicBlock *BB, Instruction *QueryInst, unsigned *Limit) {
271   BatchAAResults BatchAA(AA, &EII);
272   return getPointerDependencyFrom(MemLoc, isLoad, ScanIt, BB, QueryInst, Limit,
273                                   BatchAA);
274 }
275 
276 MemDepResult
277 MemoryDependenceResults::getInvariantGroupPointerDependency(LoadInst *LI,
278                                                             BasicBlock *BB) {
279 
280   if (!LI->hasMetadata(LLVMContext::MD_invariant_group))
281     return MemDepResult::getUnknown();
282 
283   // Take the ptr operand after all casts and geps 0. This way we can search
284   // cast graph down only.
285   Value *LoadOperand = LI->getPointerOperand()->stripPointerCasts();
286 
287   // It's is not safe to walk the use list of global value, because function
288   // passes aren't allowed to look outside their functions.
289   // FIXME: this could be fixed by filtering instructions from outside
290   // of current function.
291   if (isa<GlobalValue>(LoadOperand))
292     return MemDepResult::getUnknown();
293 
294   // Queue to process all pointers that are equivalent to load operand.
295   SmallVector<const Value *, 8> LoadOperandsQueue;
296   LoadOperandsQueue.push_back(LoadOperand);
297 
298   Instruction *ClosestDependency = nullptr;
299   // Order of instructions in uses list is unpredictible. In order to always
300   // get the same result, we will look for the closest dominance.
301   auto GetClosestDependency = [this](Instruction *Best, Instruction *Other) {
302     assert(Other && "Must call it with not null instruction");
303     if (Best == nullptr || DT.dominates(Best, Other))
304       return Other;
305     return Best;
306   };
307 
308   // FIXME: This loop is O(N^2) because dominates can be O(n) and in worst case
309   // we will see all the instructions. This should be fixed in MSSA.
310   while (!LoadOperandsQueue.empty()) {
311     const Value *Ptr = LoadOperandsQueue.pop_back_val();
312     assert(Ptr && !isa<GlobalValue>(Ptr) &&
313            "Null or GlobalValue should not be inserted");
314 
315     for (const Use &Us : Ptr->uses()) {
316       auto *U = dyn_cast<Instruction>(Us.getUser());
317       if (!U || U == LI || !DT.dominates(U, LI))
318         continue;
319 
320       // Bitcast or gep with zeros are using Ptr. Add to queue to check it's
321       // users.      U = bitcast Ptr
322       if (isa<BitCastInst>(U)) {
323         LoadOperandsQueue.push_back(U);
324         continue;
325       }
326       // Gep with zeros is equivalent to bitcast.
327       // FIXME: we are not sure if some bitcast should be canonicalized to gep 0
328       // or gep 0 to bitcast because of SROA, so there are 2 forms. When
329       // typeless pointers will be ready then both cases will be gone
330       // (and this BFS also won't be needed).
331       if (auto *GEP = dyn_cast<GetElementPtrInst>(U))
332         if (GEP->hasAllZeroIndices()) {
333           LoadOperandsQueue.push_back(U);
334           continue;
335         }
336 
337       // If we hit load/store with the same invariant.group metadata (and the
338       // same pointer operand) we can assume that value pointed by pointer
339       // operand didn't change.
340       if ((isa<LoadInst>(U) ||
341            (isa<StoreInst>(U) &&
342             cast<StoreInst>(U)->getPointerOperand() == Ptr)) &&
343           U->hasMetadata(LLVMContext::MD_invariant_group))
344         ClosestDependency = GetClosestDependency(ClosestDependency, U);
345     }
346   }
347 
348   if (!ClosestDependency)
349     return MemDepResult::getUnknown();
350   if (ClosestDependency->getParent() == BB)
351     return MemDepResult::getDef(ClosestDependency);
352   // Def(U) can't be returned here because it is non-local. If local
353   // dependency won't be found then return nonLocal counting that the
354   // user will call getNonLocalPointerDependency, which will return cached
355   // result.
356   NonLocalDefsCache.try_emplace(
357       LI, NonLocalDepResult(ClosestDependency->getParent(),
358                             MemDepResult::getDef(ClosestDependency), nullptr));
359   ReverseNonLocalDefsCache[ClosestDependency].insert(LI);
360   return MemDepResult::getNonLocal();
361 }
362 
363 // Check if SI that may alias with MemLoc can be safely skipped. This is
364 // possible in case if SI can only must alias or no alias with MemLoc (no
365 // partial overlapping possible) and it writes the same value that MemLoc
366 // contains now (it was loaded before this store and was not modified in
367 // between).
368 static bool canSkipClobberingStore(const StoreInst *SI,
369                                    const MemoryLocation &MemLoc,
370                                    Align MemLocAlign, BatchAAResults &BatchAA,
371                                    unsigned ScanLimit) {
372   if (!MemLoc.Size.hasValue())
373     return false;
374   if (MemoryLocation::get(SI).Size != MemLoc.Size)
375     return false;
376   if (MemLoc.Size.isScalable())
377     return false;
378   if (std::min(MemLocAlign, SI->getAlign()).value() <
379       MemLoc.Size.getValue().getKnownMinValue())
380     return false;
381 
382   auto *LI = dyn_cast<LoadInst>(SI->getValueOperand());
383   if (!LI || LI->getParent() != SI->getParent())
384     return false;
385   if (BatchAA.alias(MemoryLocation::get(LI), MemLoc) != AliasResult::MustAlias)
386     return false;
387   unsigned NumVisitedInsts = 0;
388   for (const Instruction *I = LI; I != SI; I = I->getNextNonDebugInstruction())
389     if (++NumVisitedInsts > ScanLimit ||
390         isModSet(BatchAA.getModRefInfo(I, MemLoc)))
391       return false;
392 
393   return true;
394 }
395 
396 MemDepResult MemoryDependenceResults::getSimplePointerDependencyFrom(
397     const MemoryLocation &MemLoc, bool isLoad, BasicBlock::iterator ScanIt,
398     BasicBlock *BB, Instruction *QueryInst, unsigned *Limit,
399     BatchAAResults &BatchAA) {
400   bool isInvariantLoad = false;
401   Align MemLocAlign =
402       MemLoc.Ptr->getPointerAlignment(BB->getModule()->getDataLayout());
403 
404   unsigned DefaultLimit = getDefaultBlockScanLimit();
405   if (!Limit)
406     Limit = &DefaultLimit;
407 
408   // We must be careful with atomic accesses, as they may allow another thread
409   //   to touch this location, clobbering it. We are conservative: if the
410   //   QueryInst is not a simple (non-atomic) memory access, we automatically
411   //   return getClobber.
412   // If it is simple, we know based on the results of
413   // "Compiler testing via a theory of sound optimisations in the C11/C++11
414   //   memory model" in PLDI 2013, that a non-atomic location can only be
415   //   clobbered between a pair of a release and an acquire action, with no
416   //   access to the location in between.
417   // Here is an example for giving the general intuition behind this rule.
418   // In the following code:
419   //   store x 0;
420   //   release action; [1]
421   //   acquire action; [4]
422   //   %val = load x;
423   // It is unsafe to replace %val by 0 because another thread may be running:
424   //   acquire action; [2]
425   //   store x 42;
426   //   release action; [3]
427   // with synchronization from 1 to 2 and from 3 to 4, resulting in %val
428   // being 42. A key property of this program however is that if either
429   // 1 or 4 were missing, there would be a race between the store of 42
430   // either the store of 0 or the load (making the whole program racy).
431   // The paper mentioned above shows that the same property is respected
432   // by every program that can detect any optimization of that kind: either
433   // it is racy (undefined) or there is a release followed by an acquire
434   // between the pair of accesses under consideration.
435 
436   // If the load is invariant, we "know" that it doesn't alias *any* write. We
437   // do want to respect mustalias results since defs are useful for value
438   // forwarding, but any mayalias write can be assumed to be noalias.
439   // Arguably, this logic should be pushed inside AliasAnalysis itself.
440   if (isLoad && QueryInst)
441     if (LoadInst *LI = dyn_cast<LoadInst>(QueryInst)) {
442       if (LI->hasMetadata(LLVMContext::MD_invariant_load))
443         isInvariantLoad = true;
444       MemLocAlign = LI->getAlign();
445     }
446 
447   // True for volatile instruction.
448   // For Load/Store return true if atomic ordering is stronger than AO,
449   // for other instruction just true if it can read or write to memory.
450   auto isComplexForReordering = [](Instruction * I, AtomicOrdering AO)->bool {
451     if (I->isVolatile())
452       return true;
453     if (auto *LI = dyn_cast<LoadInst>(I))
454       return isStrongerThan(LI->getOrdering(), AO);
455     if (auto *SI = dyn_cast<StoreInst>(I))
456       return isStrongerThan(SI->getOrdering(), AO);
457     return I->mayReadOrWriteMemory();
458   };
459 
460   // Walk backwards through the basic block, looking for dependencies.
461   while (ScanIt != BB->begin()) {
462     Instruction *Inst = &*--ScanIt;
463 
464     if (IntrinsicInst *II = dyn_cast<IntrinsicInst>(Inst))
465       // Debug intrinsics don't (and can't) cause dependencies.
466       if (isa<DbgInfoIntrinsic>(II))
467         continue;
468 
469     // Limit the amount of scanning we do so we don't end up with quadratic
470     // running time on extreme testcases.
471     --*Limit;
472     if (!*Limit)
473       return MemDepResult::getUnknown();
474 
475     if (IntrinsicInst *II = dyn_cast<IntrinsicInst>(Inst)) {
476       // If we reach a lifetime begin or end marker, then the query ends here
477       // because the value is undefined.
478       Intrinsic::ID ID = II->getIntrinsicID();
479       switch (ID) {
480       case Intrinsic::lifetime_start: {
481         // FIXME: This only considers queries directly on the invariant-tagged
482         // pointer, not on query pointers that are indexed off of them.  It'd
483         // be nice to handle that at some point (the right approach is to use
484         // GetPointerBaseWithConstantOffset).
485         MemoryLocation ArgLoc = MemoryLocation::getAfter(II->getArgOperand(1));
486         if (BatchAA.isMustAlias(ArgLoc, MemLoc))
487           return MemDepResult::getDef(II);
488         continue;
489       }
490       case Intrinsic::masked_load:
491       case Intrinsic::masked_store: {
492         MemoryLocation Loc;
493         /*ModRefInfo MR =*/ GetLocation(II, Loc, TLI);
494         AliasResult R = BatchAA.alias(Loc, MemLoc);
495         if (R == AliasResult::NoAlias)
496           continue;
497         if (R == AliasResult::MustAlias)
498           return MemDepResult::getDef(II);
499         if (ID == Intrinsic::masked_load)
500           continue;
501         return MemDepResult::getClobber(II);
502       }
503       }
504     }
505 
506     // Values depend on loads if the pointers are must aliased.  This means
507     // that a load depends on another must aliased load from the same value.
508     // One exception is atomic loads: a value can depend on an atomic load that
509     // it does not alias with when this atomic load indicates that another
510     // thread may be accessing the location.
511     if (LoadInst *LI = dyn_cast<LoadInst>(Inst)) {
512       // While volatile access cannot be eliminated, they do not have to clobber
513       // non-aliasing locations, as normal accesses, for example, can be safely
514       // reordered with volatile accesses.
515       if (LI->isVolatile()) {
516         if (!QueryInst)
517           // Original QueryInst *may* be volatile
518           return MemDepResult::getClobber(LI);
519         if (QueryInst->isVolatile())
520           // Ordering required if QueryInst is itself volatile
521           return MemDepResult::getClobber(LI);
522         // Otherwise, volatile doesn't imply any special ordering
523       }
524 
525       // Atomic loads have complications involved.
526       // A Monotonic (or higher) load is OK if the query inst is itself not
527       // atomic.
528       // FIXME: This is overly conservative.
529       if (LI->isAtomic() && isStrongerThanUnordered(LI->getOrdering())) {
530         if (!QueryInst ||
531             isComplexForReordering(QueryInst, AtomicOrdering::NotAtomic))
532           return MemDepResult::getClobber(LI);
533         if (LI->getOrdering() != AtomicOrdering::Monotonic)
534           return MemDepResult::getClobber(LI);
535       }
536 
537       MemoryLocation LoadLoc = MemoryLocation::get(LI);
538 
539       // If we found a pointer, check if it could be the same as our pointer.
540       AliasResult R = BatchAA.alias(LoadLoc, MemLoc);
541 
542       if (R == AliasResult::NoAlias)
543         continue;
544 
545       if (isLoad) {
546         // Must aliased loads are defs of each other.
547         if (R == AliasResult::MustAlias)
548           return MemDepResult::getDef(Inst);
549 
550         // If we have a partial alias, then return this as a clobber for the
551         // client to handle.
552         if (R == AliasResult::PartialAlias && R.hasOffset()) {
553           ClobberOffsets[LI] = R.getOffset();
554           return MemDepResult::getClobber(Inst);
555         }
556 
557         // Random may-alias loads don't depend on each other without a
558         // dependence.
559         continue;
560       }
561 
562       // Stores don't alias loads from read-only memory.
563       if (!isModSet(BatchAA.getModRefInfoMask(LoadLoc)))
564         continue;
565 
566       // Stores depend on may/must aliased loads.
567       return MemDepResult::getDef(Inst);
568     }
569 
570     if (StoreInst *SI = dyn_cast<StoreInst>(Inst)) {
571       // Atomic stores have complications involved.
572       // A Monotonic store is OK if the query inst is itself not atomic.
573       // FIXME: This is overly conservative.
574       if (!SI->isUnordered() && SI->isAtomic()) {
575         if (!QueryInst ||
576             isComplexForReordering(QueryInst, AtomicOrdering::Unordered))
577           return MemDepResult::getClobber(SI);
578         // Ok, if we are here the guard above guarantee us that
579         // QueryInst is a non-atomic or unordered load/store.
580         // SI is atomic with monotonic or release semantic (seq_cst for store
581         // is actually a release semantic plus total order over other seq_cst
582         // instructions, as soon as QueryInst is not seq_cst we can consider it
583         // as simple release semantic).
584         // Monotonic and Release semantic allows re-ordering before store
585         // so we are safe to go further and check the aliasing. It will prohibit
586         // re-ordering in case locations are may or must alias.
587       }
588 
589       // While volatile access cannot be eliminated, they do not have to clobber
590       // non-aliasing locations, as normal accesses can for example be reordered
591       // with volatile accesses.
592       if (SI->isVolatile())
593         if (!QueryInst || QueryInst->isVolatile())
594           return MemDepResult::getClobber(SI);
595 
596       // If alias analysis can tell that this store is guaranteed to not modify
597       // the query pointer, ignore it.  Use getModRefInfo to handle cases where
598       // the query pointer points to constant memory etc.
599       if (!isModOrRefSet(BatchAA.getModRefInfo(SI, MemLoc)))
600         continue;
601 
602       // Ok, this store might clobber the query pointer.  Check to see if it is
603       // a must alias: in this case, we want to return this as a def.
604       // FIXME: Use ModRefInfo::Must bit from getModRefInfo call above.
605       MemoryLocation StoreLoc = MemoryLocation::get(SI);
606 
607       // If we found a pointer, check if it could be the same as our pointer.
608       AliasResult R = BatchAA.alias(StoreLoc, MemLoc);
609 
610       if (R == AliasResult::NoAlias)
611         continue;
612       if (R == AliasResult::MustAlias)
613         return MemDepResult::getDef(Inst);
614       if (isInvariantLoad)
615         continue;
616       if (canSkipClobberingStore(SI, MemLoc, MemLocAlign, BatchAA, *Limit))
617         continue;
618       return MemDepResult::getClobber(Inst);
619     }
620 
621     // If this is an allocation, and if we know that the accessed pointer is to
622     // the allocation, return Def.  This means that there is no dependence and
623     // the access can be optimized based on that.  For example, a load could
624     // turn into undef.  Note that we can bypass the allocation itself when
625     // looking for a clobber in many cases; that's an alias property and is
626     // handled by BasicAA.
627     if (isa<AllocaInst>(Inst) || isNoAliasCall(Inst)) {
628       const Value *AccessPtr = getUnderlyingObject(MemLoc.Ptr);
629       if (AccessPtr == Inst || BatchAA.isMustAlias(Inst, AccessPtr))
630         return MemDepResult::getDef(Inst);
631     }
632 
633     // If we found a select instruction for MemLoc pointer, return it as Def
634     // dependency.
635     if (isa<SelectInst>(Inst) && MemLoc.Ptr == Inst)
636       return MemDepResult::getDef(Inst);
637 
638     if (isInvariantLoad)
639       continue;
640 
641     // A release fence requires that all stores complete before it, but does
642     // not prevent the reordering of following loads or stores 'before' the
643     // fence.  As a result, we look past it when finding a dependency for
644     // loads.  DSE uses this to find preceding stores to delete and thus we
645     // can't bypass the fence if the query instruction is a store.
646     if (FenceInst *FI = dyn_cast<FenceInst>(Inst))
647       if (isLoad && FI->getOrdering() == AtomicOrdering::Release)
648         continue;
649 
650     // See if this instruction (e.g. a call or vaarg) mod/ref's the pointer.
651     switch (BatchAA.getModRefInfo(Inst, MemLoc)) {
652     case ModRefInfo::NoModRef:
653       // If the call has no effect on the queried pointer, just ignore it.
654       continue;
655     case ModRefInfo::Mod:
656       return MemDepResult::getClobber(Inst);
657     case ModRefInfo::Ref:
658       // If the call is known to never store to the pointer, and if this is a
659       // load query, we can safely ignore it (scan past it).
660       if (isLoad)
661         continue;
662       [[fallthrough]];
663     default:
664       // Otherwise, there is a potential dependence.  Return a clobber.
665       return MemDepResult::getClobber(Inst);
666     }
667   }
668 
669   // No dependence found.  If this is the entry block of the function, it is
670   // unknown, otherwise it is non-local.
671   if (BB != &BB->getParent()->getEntryBlock())
672     return MemDepResult::getNonLocal();
673   return MemDepResult::getNonFuncLocal();
674 }
675 
676 MemDepResult MemoryDependenceResults::getDependency(Instruction *QueryInst) {
677   ClobberOffsets.clear();
678   Instruction *ScanPos = QueryInst;
679 
680   // Check for a cached result
681   MemDepResult &LocalCache = LocalDeps[QueryInst];
682 
683   // If the cached entry is non-dirty, just return it.  Note that this depends
684   // on MemDepResult's default constructing to 'dirty'.
685   if (!LocalCache.isDirty())
686     return LocalCache;
687 
688   // Otherwise, if we have a dirty entry, we know we can start the scan at that
689   // instruction, which may save us some work.
690   if (Instruction *Inst = LocalCache.getInst()) {
691     ScanPos = Inst;
692 
693     RemoveFromReverseMap(ReverseLocalDeps, Inst, QueryInst);
694   }
695 
696   BasicBlock *QueryParent = QueryInst->getParent();
697 
698   // Do the scan.
699   if (BasicBlock::iterator(QueryInst) == QueryParent->begin()) {
700     // No dependence found. If this is the entry block of the function, it is
701     // unknown, otherwise it is non-local.
702     if (QueryParent != &QueryParent->getParent()->getEntryBlock())
703       LocalCache = MemDepResult::getNonLocal();
704     else
705       LocalCache = MemDepResult::getNonFuncLocal();
706   } else {
707     MemoryLocation MemLoc;
708     ModRefInfo MR = GetLocation(QueryInst, MemLoc, TLI);
709     if (MemLoc.Ptr) {
710       // If we can do a pointer scan, make it happen.
711       bool isLoad = !isModSet(MR);
712       if (auto *II = dyn_cast<IntrinsicInst>(QueryInst))
713         isLoad |= II->getIntrinsicID() == Intrinsic::lifetime_start;
714 
715       LocalCache =
716           getPointerDependencyFrom(MemLoc, isLoad, ScanPos->getIterator(),
717                                    QueryParent, QueryInst, nullptr);
718     } else if (auto *QueryCall = dyn_cast<CallBase>(QueryInst)) {
719       bool isReadOnly = AA.onlyReadsMemory(QueryCall);
720       LocalCache = getCallDependencyFrom(QueryCall, isReadOnly,
721                                          ScanPos->getIterator(), QueryParent);
722     } else
723       // Non-memory instruction.
724       LocalCache = MemDepResult::getUnknown();
725   }
726 
727   // Remember the result!
728   if (Instruction *I = LocalCache.getInst())
729     ReverseLocalDeps[I].insert(QueryInst);
730 
731   return LocalCache;
732 }
733 
734 #ifndef NDEBUG
735 /// This method is used when -debug is specified to verify that cache arrays
736 /// are properly kept sorted.
737 static void AssertSorted(MemoryDependenceResults::NonLocalDepInfo &Cache,
738                          int Count = -1) {
739   if (Count == -1)
740     Count = Cache.size();
741   assert(std::is_sorted(Cache.begin(), Cache.begin() + Count) &&
742          "Cache isn't sorted!");
743 }
744 #endif
745 
746 const MemoryDependenceResults::NonLocalDepInfo &
747 MemoryDependenceResults::getNonLocalCallDependency(CallBase *QueryCall) {
748   assert(getDependency(QueryCall).isNonLocal() &&
749          "getNonLocalCallDependency should only be used on calls with "
750          "non-local deps!");
751   PerInstNLInfo &CacheP = NonLocalDepsMap[QueryCall];
752   NonLocalDepInfo &Cache = CacheP.first;
753 
754   // This is the set of blocks that need to be recomputed.  In the cached case,
755   // this can happen due to instructions being deleted etc. In the uncached
756   // case, this starts out as the set of predecessors we care about.
757   SmallVector<BasicBlock *, 32> DirtyBlocks;
758 
759   if (!Cache.empty()) {
760     // Okay, we have a cache entry.  If we know it is not dirty, just return it
761     // with no computation.
762     if (!CacheP.second) {
763       ++NumCacheNonLocal;
764       return Cache;
765     }
766 
767     // If we already have a partially computed set of results, scan them to
768     // determine what is dirty, seeding our initial DirtyBlocks worklist.
769     for (auto &Entry : Cache)
770       if (Entry.getResult().isDirty())
771         DirtyBlocks.push_back(Entry.getBB());
772 
773     // Sort the cache so that we can do fast binary search lookups below.
774     llvm::sort(Cache);
775 
776     ++NumCacheDirtyNonLocal;
777   } else {
778     // Seed DirtyBlocks with each of the preds of QueryInst's block.
779     BasicBlock *QueryBB = QueryCall->getParent();
780     append_range(DirtyBlocks, PredCache.get(QueryBB));
781     ++NumUncacheNonLocal;
782   }
783 
784   // isReadonlyCall - If this is a read-only call, we can be more aggressive.
785   bool isReadonlyCall = AA.onlyReadsMemory(QueryCall);
786 
787   SmallPtrSet<BasicBlock *, 32> Visited;
788 
789   unsigned NumSortedEntries = Cache.size();
790   LLVM_DEBUG(AssertSorted(Cache));
791 
792   // Iterate while we still have blocks to update.
793   while (!DirtyBlocks.empty()) {
794     BasicBlock *DirtyBB = DirtyBlocks.pop_back_val();
795 
796     // Already processed this block?
797     if (!Visited.insert(DirtyBB).second)
798       continue;
799 
800     // Do a binary search to see if we already have an entry for this block in
801     // the cache set.  If so, find it.
802     LLVM_DEBUG(AssertSorted(Cache, NumSortedEntries));
803     NonLocalDepInfo::iterator Entry =
804         std::upper_bound(Cache.begin(), Cache.begin() + NumSortedEntries,
805                          NonLocalDepEntry(DirtyBB));
806     if (Entry != Cache.begin() && std::prev(Entry)->getBB() == DirtyBB)
807       --Entry;
808 
809     NonLocalDepEntry *ExistingResult = nullptr;
810     if (Entry != Cache.begin() + NumSortedEntries &&
811         Entry->getBB() == DirtyBB) {
812       // If we already have an entry, and if it isn't already dirty, the block
813       // is done.
814       if (!Entry->getResult().isDirty())
815         continue;
816 
817       // Otherwise, remember this slot so we can update the value.
818       ExistingResult = &*Entry;
819     }
820 
821     // If the dirty entry has a pointer, start scanning from it so we don't have
822     // to rescan the entire block.
823     BasicBlock::iterator ScanPos = DirtyBB->end();
824     if (ExistingResult) {
825       if (Instruction *Inst = ExistingResult->getResult().getInst()) {
826         ScanPos = Inst->getIterator();
827         // We're removing QueryInst's use of Inst.
828         RemoveFromReverseMap<Instruction *>(ReverseNonLocalDeps, Inst,
829                                             QueryCall);
830       }
831     }
832 
833     // Find out if this block has a local dependency for QueryInst.
834     MemDepResult Dep;
835 
836     if (ScanPos != DirtyBB->begin()) {
837       Dep = getCallDependencyFrom(QueryCall, isReadonlyCall, ScanPos, DirtyBB);
838     } else if (DirtyBB != &DirtyBB->getParent()->getEntryBlock()) {
839       // No dependence found.  If this is the entry block of the function, it is
840       // a clobber, otherwise it is unknown.
841       Dep = MemDepResult::getNonLocal();
842     } else {
843       Dep = MemDepResult::getNonFuncLocal();
844     }
845 
846     // If we had a dirty entry for the block, update it.  Otherwise, just add
847     // a new entry.
848     if (ExistingResult)
849       ExistingResult->setResult(Dep);
850     else
851       Cache.push_back(NonLocalDepEntry(DirtyBB, Dep));
852 
853     // If the block has a dependency (i.e. it isn't completely transparent to
854     // the value), remember the association!
855     if (!Dep.isNonLocal()) {
856       // Keep the ReverseNonLocalDeps map up to date so we can efficiently
857       // update this when we remove instructions.
858       if (Instruction *Inst = Dep.getInst())
859         ReverseNonLocalDeps[Inst].insert(QueryCall);
860     } else {
861 
862       // If the block *is* completely transparent to the load, we need to check
863       // the predecessors of this block.  Add them to our worklist.
864       append_range(DirtyBlocks, PredCache.get(DirtyBB));
865     }
866   }
867 
868   return Cache;
869 }
870 
871 void MemoryDependenceResults::getNonLocalPointerDependency(
872     Instruction *QueryInst, SmallVectorImpl<NonLocalDepResult> &Result) {
873   const MemoryLocation Loc = MemoryLocation::get(QueryInst);
874   bool isLoad = isa<LoadInst>(QueryInst);
875   BasicBlock *FromBB = QueryInst->getParent();
876   assert(FromBB);
877 
878   assert(Loc.Ptr->getType()->isPointerTy() &&
879          "Can't get pointer deps of a non-pointer!");
880   Result.clear();
881   {
882     // Check if there is cached Def with invariant.group.
883     auto NonLocalDefIt = NonLocalDefsCache.find(QueryInst);
884     if (NonLocalDefIt != NonLocalDefsCache.end()) {
885       Result.push_back(NonLocalDefIt->second);
886       ReverseNonLocalDefsCache[NonLocalDefIt->second.getResult().getInst()]
887           .erase(QueryInst);
888       NonLocalDefsCache.erase(NonLocalDefIt);
889       return;
890     }
891   }
892   // This routine does not expect to deal with volatile instructions.
893   // Doing so would require piping through the QueryInst all the way through.
894   // TODO: volatiles can't be elided, but they can be reordered with other
895   // non-volatile accesses.
896 
897   // We currently give up on any instruction which is ordered, but we do handle
898   // atomic instructions which are unordered.
899   // TODO: Handle ordered instructions
900   auto isOrdered = [](Instruction *Inst) {
901     if (LoadInst *LI = dyn_cast<LoadInst>(Inst)) {
902       return !LI->isUnordered();
903     } else if (StoreInst *SI = dyn_cast<StoreInst>(Inst)) {
904       return !SI->isUnordered();
905     }
906     return false;
907   };
908   if (QueryInst->isVolatile() || isOrdered(QueryInst)) {
909     Result.push_back(NonLocalDepResult(FromBB, MemDepResult::getUnknown(),
910                                        const_cast<Value *>(Loc.Ptr)));
911     return;
912   }
913   const DataLayout &DL = FromBB->getModule()->getDataLayout();
914   PHITransAddr Address(const_cast<Value *>(Loc.Ptr), DL, &AC);
915 
916   // This is the set of blocks we've inspected, and the pointer we consider in
917   // each block.  Because of critical edges, we currently bail out if querying
918   // a block with multiple different pointers.  This can happen during PHI
919   // translation.
920   DenseMap<BasicBlock *, Value *> Visited;
921   if (getNonLocalPointerDepFromBB(QueryInst, Address, Loc, isLoad, FromBB,
922                                    Result, Visited, true))
923     return;
924   Result.clear();
925   Result.push_back(NonLocalDepResult(FromBB, MemDepResult::getUnknown(),
926                                      const_cast<Value *>(Loc.Ptr)));
927 }
928 
929 /// Compute the memdep value for BB with Pointer/PointeeSize using either
930 /// cached information in Cache or by doing a lookup (which may use dirty cache
931 /// info if available).
932 ///
933 /// If we do a lookup, add the result to the cache.
934 MemDepResult MemoryDependenceResults::getNonLocalInfoForBlock(
935     Instruction *QueryInst, const MemoryLocation &Loc, bool isLoad,
936     BasicBlock *BB, NonLocalDepInfo *Cache, unsigned NumSortedEntries,
937     BatchAAResults &BatchAA) {
938 
939   bool isInvariantLoad = false;
940 
941   if (LoadInst *LI = dyn_cast_or_null<LoadInst>(QueryInst))
942     isInvariantLoad = LI->getMetadata(LLVMContext::MD_invariant_load);
943 
944   // Do a binary search to see if we already have an entry for this block in
945   // the cache set.  If so, find it.
946   NonLocalDepInfo::iterator Entry = std::upper_bound(
947       Cache->begin(), Cache->begin() + NumSortedEntries, NonLocalDepEntry(BB));
948   if (Entry != Cache->begin() && (Entry - 1)->getBB() == BB)
949     --Entry;
950 
951   NonLocalDepEntry *ExistingResult = nullptr;
952   if (Entry != Cache->begin() + NumSortedEntries && Entry->getBB() == BB)
953     ExistingResult = &*Entry;
954 
955   // Use cached result for invariant load only if there is no dependency for non
956   // invariant load. In this case invariant load can not have any dependency as
957   // well.
958   if (ExistingResult && isInvariantLoad &&
959       !ExistingResult->getResult().isNonFuncLocal())
960     ExistingResult = nullptr;
961 
962   // If we have a cached entry, and it is non-dirty, use it as the value for
963   // this dependency.
964   if (ExistingResult && !ExistingResult->getResult().isDirty()) {
965     ++NumCacheNonLocalPtr;
966     return ExistingResult->getResult();
967   }
968 
969   // Otherwise, we have to scan for the value.  If we have a dirty cache
970   // entry, start scanning from its position, otherwise we scan from the end
971   // of the block.
972   BasicBlock::iterator ScanPos = BB->end();
973   if (ExistingResult && ExistingResult->getResult().getInst()) {
974     assert(ExistingResult->getResult().getInst()->getParent() == BB &&
975            "Instruction invalidated?");
976     ++NumCacheDirtyNonLocalPtr;
977     ScanPos = ExistingResult->getResult().getInst()->getIterator();
978 
979     // Eliminating the dirty entry from 'Cache', so update the reverse info.
980     ValueIsLoadPair CacheKey(Loc.Ptr, isLoad);
981     RemoveFromReverseMap(ReverseNonLocalPtrDeps, &*ScanPos, CacheKey);
982   } else {
983     ++NumUncacheNonLocalPtr;
984   }
985 
986   // Scan the block for the dependency.
987   MemDepResult Dep = getPointerDependencyFrom(Loc, isLoad, ScanPos, BB,
988                                               QueryInst, nullptr, BatchAA);
989 
990   // Don't cache results for invariant load.
991   if (isInvariantLoad)
992     return Dep;
993 
994   // If we had a dirty entry for the block, update it.  Otherwise, just add
995   // a new entry.
996   if (ExistingResult)
997     ExistingResult->setResult(Dep);
998   else
999     Cache->push_back(NonLocalDepEntry(BB, Dep));
1000 
1001   // If the block has a dependency (i.e. it isn't completely transparent to
1002   // the value), remember the reverse association because we just added it
1003   // to Cache!
1004   if (!Dep.isLocal())
1005     return Dep;
1006 
1007   // Keep the ReverseNonLocalPtrDeps map up to date so we can efficiently
1008   // update MemDep when we remove instructions.
1009   Instruction *Inst = Dep.getInst();
1010   assert(Inst && "Didn't depend on anything?");
1011   ValueIsLoadPair CacheKey(Loc.Ptr, isLoad);
1012   ReverseNonLocalPtrDeps[Inst].insert(CacheKey);
1013   return Dep;
1014 }
1015 
1016 /// Sort the NonLocalDepInfo cache, given a certain number of elements in the
1017 /// array that are already properly ordered.
1018 ///
1019 /// This is optimized for the case when only a few entries are added.
1020 static void
1021 SortNonLocalDepInfoCache(MemoryDependenceResults::NonLocalDepInfo &Cache,
1022                          unsigned NumSortedEntries) {
1023   switch (Cache.size() - NumSortedEntries) {
1024   case 0:
1025     // done, no new entries.
1026     break;
1027   case 2: {
1028     // Two new entries, insert the last one into place.
1029     NonLocalDepEntry Val = Cache.back();
1030     Cache.pop_back();
1031     MemoryDependenceResults::NonLocalDepInfo::iterator Entry =
1032         std::upper_bound(Cache.begin(), Cache.end() - 1, Val);
1033     Cache.insert(Entry, Val);
1034     [[fallthrough]];
1035   }
1036   case 1:
1037     // One new entry, Just insert the new value at the appropriate position.
1038     if (Cache.size() != 1) {
1039       NonLocalDepEntry Val = Cache.back();
1040       Cache.pop_back();
1041       MemoryDependenceResults::NonLocalDepInfo::iterator Entry =
1042           llvm::upper_bound(Cache, Val);
1043       Cache.insert(Entry, Val);
1044     }
1045     break;
1046   default:
1047     // Added many values, do a full scale sort.
1048     llvm::sort(Cache);
1049     break;
1050   }
1051 }
1052 
1053 /// Perform a dependency query based on pointer/pointeesize starting at the end
1054 /// of StartBB.
1055 ///
1056 /// Add any clobber/def results to the results vector and keep track of which
1057 /// blocks are visited in 'Visited'.
1058 ///
1059 /// This has special behavior for the first block queries (when SkipFirstBlock
1060 /// is true).  In this special case, it ignores the contents of the specified
1061 /// block and starts returning dependence info for its predecessors.
1062 ///
1063 /// This function returns true on success, or false to indicate that it could
1064 /// not compute dependence information for some reason.  This should be treated
1065 /// as a clobber dependence on the first instruction in the predecessor block.
1066 bool MemoryDependenceResults::getNonLocalPointerDepFromBB(
1067     Instruction *QueryInst, const PHITransAddr &Pointer,
1068     const MemoryLocation &Loc, bool isLoad, BasicBlock *StartBB,
1069     SmallVectorImpl<NonLocalDepResult> &Result,
1070     DenseMap<BasicBlock *, Value *> &Visited, bool SkipFirstBlock,
1071     bool IsIncomplete) {
1072   // Look up the cached info for Pointer.
1073   ValueIsLoadPair CacheKey(Pointer.getAddr(), isLoad);
1074 
1075   // Set up a temporary NLPI value. If the map doesn't yet have an entry for
1076   // CacheKey, this value will be inserted as the associated value. Otherwise,
1077   // it'll be ignored, and we'll have to check to see if the cached size and
1078   // aa tags are consistent with the current query.
1079   NonLocalPointerInfo InitialNLPI;
1080   InitialNLPI.Size = Loc.Size;
1081   InitialNLPI.AATags = Loc.AATags;
1082 
1083   bool isInvariantLoad = false;
1084   if (LoadInst *LI = dyn_cast_or_null<LoadInst>(QueryInst))
1085     isInvariantLoad = LI->getMetadata(LLVMContext::MD_invariant_load);
1086 
1087   // Get the NLPI for CacheKey, inserting one into the map if it doesn't
1088   // already have one.
1089   std::pair<CachedNonLocalPointerInfo::iterator, bool> Pair =
1090       NonLocalPointerDeps.insert(std::make_pair(CacheKey, InitialNLPI));
1091   NonLocalPointerInfo *CacheInfo = &Pair.first->second;
1092 
1093   // If we already have a cache entry for this CacheKey, we may need to do some
1094   // work to reconcile the cache entry and the current query.
1095   // Invariant loads don't participate in caching. Thus no need to reconcile.
1096   if (!isInvariantLoad && !Pair.second) {
1097     if (CacheInfo->Size != Loc.Size) {
1098       bool ThrowOutEverything;
1099       if (CacheInfo->Size.hasValue() && Loc.Size.hasValue()) {
1100         // FIXME: We may be able to do better in the face of results with mixed
1101         // precision. We don't appear to get them in practice, though, so just
1102         // be conservative.
1103         ThrowOutEverything =
1104             CacheInfo->Size.isPrecise() != Loc.Size.isPrecise() ||
1105             !TypeSize::isKnownGE(CacheInfo->Size.getValue(),
1106                                  Loc.Size.getValue());
1107       } else {
1108         // For our purposes, unknown size > all others.
1109         ThrowOutEverything = !Loc.Size.hasValue();
1110       }
1111 
1112       if (ThrowOutEverything) {
1113         // The query's Size is greater than the cached one. Throw out the
1114         // cached data and proceed with the query at the greater size.
1115         CacheInfo->Pair = BBSkipFirstBlockPair();
1116         CacheInfo->Size = Loc.Size;
1117         for (auto &Entry : CacheInfo->NonLocalDeps)
1118           if (Instruction *Inst = Entry.getResult().getInst())
1119             RemoveFromReverseMap(ReverseNonLocalPtrDeps, Inst, CacheKey);
1120         CacheInfo->NonLocalDeps.clear();
1121         // The cache is cleared (in the above line) so we will have lost
1122         // information about blocks we have already visited. We therefore must
1123         // assume that the cache information is incomplete.
1124         IsIncomplete = true;
1125       } else {
1126         // This query's Size is less than the cached one. Conservatively restart
1127         // the query using the greater size.
1128         return getNonLocalPointerDepFromBB(
1129             QueryInst, Pointer, Loc.getWithNewSize(CacheInfo->Size), isLoad,
1130             StartBB, Result, Visited, SkipFirstBlock, IsIncomplete);
1131       }
1132     }
1133 
1134     // If the query's AATags are inconsistent with the cached one,
1135     // conservatively throw out the cached data and restart the query with
1136     // no tag if needed.
1137     if (CacheInfo->AATags != Loc.AATags) {
1138       if (CacheInfo->AATags) {
1139         CacheInfo->Pair = BBSkipFirstBlockPair();
1140         CacheInfo->AATags = AAMDNodes();
1141         for (auto &Entry : CacheInfo->NonLocalDeps)
1142           if (Instruction *Inst = Entry.getResult().getInst())
1143             RemoveFromReverseMap(ReverseNonLocalPtrDeps, Inst, CacheKey);
1144         CacheInfo->NonLocalDeps.clear();
1145         // The cache is cleared (in the above line) so we will have lost
1146         // information about blocks we have already visited. We therefore must
1147         // assume that the cache information is incomplete.
1148         IsIncomplete = true;
1149       }
1150       if (Loc.AATags)
1151         return getNonLocalPointerDepFromBB(
1152             QueryInst, Pointer, Loc.getWithoutAATags(), isLoad, StartBB, Result,
1153             Visited, SkipFirstBlock, IsIncomplete);
1154     }
1155   }
1156 
1157   NonLocalDepInfo *Cache = &CacheInfo->NonLocalDeps;
1158 
1159   // If we have valid cached information for exactly the block we are
1160   // investigating, just return it with no recomputation.
1161   // Don't use cached information for invariant loads since it is valid for
1162   // non-invariant loads only.
1163   if (!IsIncomplete && !isInvariantLoad &&
1164       CacheInfo->Pair == BBSkipFirstBlockPair(StartBB, SkipFirstBlock)) {
1165     // We have a fully cached result for this query then we can just return the
1166     // cached results and populate the visited set.  However, we have to verify
1167     // that we don't already have conflicting results for these blocks.  Check
1168     // to ensure that if a block in the results set is in the visited set that
1169     // it was for the same pointer query.
1170     if (!Visited.empty()) {
1171       for (auto &Entry : *Cache) {
1172         DenseMap<BasicBlock *, Value *>::iterator VI =
1173             Visited.find(Entry.getBB());
1174         if (VI == Visited.end() || VI->second == Pointer.getAddr())
1175           continue;
1176 
1177         // We have a pointer mismatch in a block.  Just return false, saying
1178         // that something was clobbered in this result.  We could also do a
1179         // non-fully cached query, but there is little point in doing this.
1180         return false;
1181       }
1182     }
1183 
1184     Value *Addr = Pointer.getAddr();
1185     for (auto &Entry : *Cache) {
1186       Visited.insert(std::make_pair(Entry.getBB(), Addr));
1187       if (Entry.getResult().isNonLocal()) {
1188         continue;
1189       }
1190 
1191       if (DT.isReachableFromEntry(Entry.getBB())) {
1192         Result.push_back(
1193             NonLocalDepResult(Entry.getBB(), Entry.getResult(), Addr));
1194       }
1195     }
1196     ++NumCacheCompleteNonLocalPtr;
1197     return true;
1198   }
1199 
1200   // Otherwise, either this is a new block, a block with an invalid cache
1201   // pointer or one that we're about to invalidate by putting more info into
1202   // it than its valid cache info.  If empty and not explicitly indicated as
1203   // incomplete, the result will be valid cache info, otherwise it isn't.
1204   //
1205   // Invariant loads don't affect cache in any way thus no need to update
1206   // CacheInfo as well.
1207   if (!isInvariantLoad) {
1208     if (!IsIncomplete && Cache->empty())
1209       CacheInfo->Pair = BBSkipFirstBlockPair(StartBB, SkipFirstBlock);
1210     else
1211       CacheInfo->Pair = BBSkipFirstBlockPair();
1212   }
1213 
1214   SmallVector<BasicBlock *, 32> Worklist;
1215   Worklist.push_back(StartBB);
1216 
1217   // PredList used inside loop.
1218   SmallVector<std::pair<BasicBlock *, PHITransAddr>, 16> PredList;
1219 
1220   // Keep track of the entries that we know are sorted.  Previously cached
1221   // entries will all be sorted.  The entries we add we only sort on demand (we
1222   // don't insert every element into its sorted position).  We know that we
1223   // won't get any reuse from currently inserted values, because we don't
1224   // revisit blocks after we insert info for them.
1225   unsigned NumSortedEntries = Cache->size();
1226   unsigned WorklistEntries = BlockNumberLimit;
1227   bool GotWorklistLimit = false;
1228   LLVM_DEBUG(AssertSorted(*Cache));
1229 
1230   BatchAAResults BatchAA(AA, &EII);
1231   while (!Worklist.empty()) {
1232     BasicBlock *BB = Worklist.pop_back_val();
1233 
1234     // If we do process a large number of blocks it becomes very expensive and
1235     // likely it isn't worth worrying about
1236     if (Result.size() > NumResultsLimit) {
1237       // Sort it now (if needed) so that recursive invocations of
1238       // getNonLocalPointerDepFromBB and other routines that could reuse the
1239       // cache value will only see properly sorted cache arrays.
1240       if (Cache && NumSortedEntries != Cache->size()) {
1241         SortNonLocalDepInfoCache(*Cache, NumSortedEntries);
1242       }
1243       // Since we bail out, the "Cache" set won't contain all of the
1244       // results for the query.  This is ok (we can still use it to accelerate
1245       // specific block queries) but we can't do the fastpath "return all
1246       // results from the set".  Clear out the indicator for this.
1247       CacheInfo->Pair = BBSkipFirstBlockPair();
1248       return false;
1249     }
1250 
1251     // Skip the first block if we have it.
1252     if (!SkipFirstBlock) {
1253       // Analyze the dependency of *Pointer in FromBB.  See if we already have
1254       // been here.
1255       assert(Visited.count(BB) && "Should check 'visited' before adding to WL");
1256 
1257       // Get the dependency info for Pointer in BB.  If we have cached
1258       // information, we will use it, otherwise we compute it.
1259       LLVM_DEBUG(AssertSorted(*Cache, NumSortedEntries));
1260       MemDepResult Dep = getNonLocalInfoForBlock(
1261           QueryInst, Loc, isLoad, BB, Cache, NumSortedEntries, BatchAA);
1262 
1263       // If we got a Def or Clobber, add this to the list of results.
1264       if (!Dep.isNonLocal()) {
1265         if (DT.isReachableFromEntry(BB)) {
1266           Result.push_back(NonLocalDepResult(BB, Dep, Pointer.getAddr()));
1267           continue;
1268         }
1269       }
1270     }
1271 
1272     // If 'Pointer' is an instruction defined in this block, then we need to do
1273     // phi translation to change it into a value live in the predecessor block.
1274     // If not, we just add the predecessors to the worklist and scan them with
1275     // the same Pointer.
1276     if (!Pointer.needsPHITranslationFromBlock(BB)) {
1277       SkipFirstBlock = false;
1278       SmallVector<BasicBlock *, 16> NewBlocks;
1279       for (BasicBlock *Pred : PredCache.get(BB)) {
1280         // Verify that we haven't looked at this block yet.
1281         std::pair<DenseMap<BasicBlock *, Value *>::iterator, bool> InsertRes =
1282             Visited.insert(std::make_pair(Pred, Pointer.getAddr()));
1283         if (InsertRes.second) {
1284           // First time we've looked at *PI.
1285           NewBlocks.push_back(Pred);
1286           continue;
1287         }
1288 
1289         // If we have seen this block before, but it was with a different
1290         // pointer then we have a phi translation failure and we have to treat
1291         // this as a clobber.
1292         if (InsertRes.first->second != Pointer.getAddr()) {
1293           // Make sure to clean up the Visited map before continuing on to
1294           // PredTranslationFailure.
1295           for (auto *NewBlock : NewBlocks)
1296             Visited.erase(NewBlock);
1297           goto PredTranslationFailure;
1298         }
1299       }
1300       if (NewBlocks.size() > WorklistEntries) {
1301         // Make sure to clean up the Visited map before continuing on to
1302         // PredTranslationFailure.
1303         for (auto *NewBlock : NewBlocks)
1304           Visited.erase(NewBlock);
1305         GotWorklistLimit = true;
1306         goto PredTranslationFailure;
1307       }
1308       WorklistEntries -= NewBlocks.size();
1309       Worklist.append(NewBlocks.begin(), NewBlocks.end());
1310       continue;
1311     }
1312 
1313     // We do need to do phi translation, if we know ahead of time we can't phi
1314     // translate this value, don't even try.
1315     if (!Pointer.isPotentiallyPHITranslatable())
1316       goto PredTranslationFailure;
1317 
1318     // We may have added values to the cache list before this PHI translation.
1319     // If so, we haven't done anything to ensure that the cache remains sorted.
1320     // Sort it now (if needed) so that recursive invocations of
1321     // getNonLocalPointerDepFromBB and other routines that could reuse the cache
1322     // value will only see properly sorted cache arrays.
1323     if (Cache && NumSortedEntries != Cache->size()) {
1324       SortNonLocalDepInfoCache(*Cache, NumSortedEntries);
1325       NumSortedEntries = Cache->size();
1326     }
1327     Cache = nullptr;
1328 
1329     PredList.clear();
1330     for (BasicBlock *Pred : PredCache.get(BB)) {
1331       PredList.push_back(std::make_pair(Pred, Pointer));
1332 
1333       // Get the PHI translated pointer in this predecessor.  This can fail if
1334       // not translatable, in which case the getAddr() returns null.
1335       PHITransAddr &PredPointer = PredList.back().second;
1336       Value *PredPtrVal =
1337           PredPointer.translateValue(BB, Pred, &DT, /*MustDominate=*/false);
1338 
1339       // Check to see if we have already visited this pred block with another
1340       // pointer.  If so, we can't do this lookup.  This failure can occur
1341       // with PHI translation when a critical edge exists and the PHI node in
1342       // the successor translates to a pointer value different than the
1343       // pointer the block was first analyzed with.
1344       std::pair<DenseMap<BasicBlock *, Value *>::iterator, bool> InsertRes =
1345           Visited.insert(std::make_pair(Pred, PredPtrVal));
1346 
1347       if (!InsertRes.second) {
1348         // We found the pred; take it off the list of preds to visit.
1349         PredList.pop_back();
1350 
1351         // If the predecessor was visited with PredPtr, then we already did
1352         // the analysis and can ignore it.
1353         if (InsertRes.first->second == PredPtrVal)
1354           continue;
1355 
1356         // Otherwise, the block was previously analyzed with a different
1357         // pointer.  We can't represent the result of this case, so we just
1358         // treat this as a phi translation failure.
1359 
1360         // Make sure to clean up the Visited map before continuing on to
1361         // PredTranslationFailure.
1362         for (const auto &Pred : PredList)
1363           Visited.erase(Pred.first);
1364 
1365         goto PredTranslationFailure;
1366       }
1367     }
1368 
1369     // Actually process results here; this need to be a separate loop to avoid
1370     // calling getNonLocalPointerDepFromBB for blocks we don't want to return
1371     // any results for.  (getNonLocalPointerDepFromBB will modify our
1372     // datastructures in ways the code after the PredTranslationFailure label
1373     // doesn't expect.)
1374     for (auto &I : PredList) {
1375       BasicBlock *Pred = I.first;
1376       PHITransAddr &PredPointer = I.second;
1377       Value *PredPtrVal = PredPointer.getAddr();
1378 
1379       bool CanTranslate = true;
1380       // If PHI translation was unable to find an available pointer in this
1381       // predecessor, then we have to assume that the pointer is clobbered in
1382       // that predecessor.  We can still do PRE of the load, which would insert
1383       // a computation of the pointer in this predecessor.
1384       if (!PredPtrVal)
1385         CanTranslate = false;
1386 
1387       // FIXME: it is entirely possible that PHI translating will end up with
1388       // the same value.  Consider PHI translating something like:
1389       // X = phi [x, bb1], [y, bb2].  PHI translating for bb1 doesn't *need*
1390       // to recurse here, pedantically speaking.
1391 
1392       // If getNonLocalPointerDepFromBB fails here, that means the cached
1393       // result conflicted with the Visited list; we have to conservatively
1394       // assume it is unknown, but this also does not block PRE of the load.
1395       if (!CanTranslate ||
1396           !getNonLocalPointerDepFromBB(QueryInst, PredPointer,
1397                                       Loc.getWithNewPtr(PredPtrVal), isLoad,
1398                                       Pred, Result, Visited)) {
1399         // Add the entry to the Result list.
1400         NonLocalDepResult Entry(Pred, MemDepResult::getUnknown(), PredPtrVal);
1401         Result.push_back(Entry);
1402 
1403         // Since we had a phi translation failure, the cache for CacheKey won't
1404         // include all of the entries that we need to immediately satisfy future
1405         // queries.  Mark this in NonLocalPointerDeps by setting the
1406         // BBSkipFirstBlockPair pointer to null.  This requires reuse of the
1407         // cached value to do more work but not miss the phi trans failure.
1408         NonLocalPointerInfo &NLPI = NonLocalPointerDeps[CacheKey];
1409         NLPI.Pair = BBSkipFirstBlockPair();
1410         continue;
1411       }
1412     }
1413 
1414     // Refresh the CacheInfo/Cache pointer so that it isn't invalidated.
1415     CacheInfo = &NonLocalPointerDeps[CacheKey];
1416     Cache = &CacheInfo->NonLocalDeps;
1417     NumSortedEntries = Cache->size();
1418 
1419     // Since we did phi translation, the "Cache" set won't contain all of the
1420     // results for the query.  This is ok (we can still use it to accelerate
1421     // specific block queries) but we can't do the fastpath "return all
1422     // results from the set"  Clear out the indicator for this.
1423     CacheInfo->Pair = BBSkipFirstBlockPair();
1424     SkipFirstBlock = false;
1425     continue;
1426 
1427   PredTranslationFailure:
1428     // The following code is "failure"; we can't produce a sane translation
1429     // for the given block.  It assumes that we haven't modified any of
1430     // our datastructures while processing the current block.
1431 
1432     if (!Cache) {
1433       // Refresh the CacheInfo/Cache pointer if it got invalidated.
1434       CacheInfo = &NonLocalPointerDeps[CacheKey];
1435       Cache = &CacheInfo->NonLocalDeps;
1436       NumSortedEntries = Cache->size();
1437     }
1438 
1439     // Since we failed phi translation, the "Cache" set won't contain all of the
1440     // results for the query.  This is ok (we can still use it to accelerate
1441     // specific block queries) but we can't do the fastpath "return all
1442     // results from the set".  Clear out the indicator for this.
1443     CacheInfo->Pair = BBSkipFirstBlockPair();
1444 
1445     // If *nothing* works, mark the pointer as unknown.
1446     //
1447     // If this is the magic first block, return this as a clobber of the whole
1448     // incoming value.  Since we can't phi translate to one of the predecessors,
1449     // we have to bail out.
1450     if (SkipFirstBlock)
1451       return false;
1452 
1453     // Results of invariant loads are not cached thus no need to update cached
1454     // information.
1455     if (!isInvariantLoad) {
1456       for (NonLocalDepEntry &I : llvm::reverse(*Cache)) {
1457         if (I.getBB() != BB)
1458           continue;
1459 
1460         assert((GotWorklistLimit || I.getResult().isNonLocal() ||
1461                 !DT.isReachableFromEntry(BB)) &&
1462                "Should only be here with transparent block");
1463 
1464         I.setResult(MemDepResult::getUnknown());
1465 
1466 
1467         break;
1468       }
1469     }
1470     (void)GotWorklistLimit;
1471     // Go ahead and report unknown dependence.
1472     Result.push_back(
1473         NonLocalDepResult(BB, MemDepResult::getUnknown(), Pointer.getAddr()));
1474   }
1475 
1476   // Okay, we're done now.  If we added new values to the cache, re-sort it.
1477   SortNonLocalDepInfoCache(*Cache, NumSortedEntries);
1478   LLVM_DEBUG(AssertSorted(*Cache));
1479   return true;
1480 }
1481 
1482 /// If P exists in CachedNonLocalPointerInfo or NonLocalDefsCache, remove it.
1483 void MemoryDependenceResults::removeCachedNonLocalPointerDependencies(
1484     ValueIsLoadPair P) {
1485 
1486   // Most of the time this cache is empty.
1487   if (!NonLocalDefsCache.empty()) {
1488     auto it = NonLocalDefsCache.find(P.getPointer());
1489     if (it != NonLocalDefsCache.end()) {
1490       RemoveFromReverseMap(ReverseNonLocalDefsCache,
1491                            it->second.getResult().getInst(), P.getPointer());
1492       NonLocalDefsCache.erase(it);
1493     }
1494 
1495     if (auto *I = dyn_cast<Instruction>(P.getPointer())) {
1496       auto toRemoveIt = ReverseNonLocalDefsCache.find(I);
1497       if (toRemoveIt != ReverseNonLocalDefsCache.end()) {
1498         for (const auto *entry : toRemoveIt->second)
1499           NonLocalDefsCache.erase(entry);
1500         ReverseNonLocalDefsCache.erase(toRemoveIt);
1501       }
1502     }
1503   }
1504 
1505   CachedNonLocalPointerInfo::iterator It = NonLocalPointerDeps.find(P);
1506   if (It == NonLocalPointerDeps.end())
1507     return;
1508 
1509   // Remove all of the entries in the BB->val map.  This involves removing
1510   // instructions from the reverse map.
1511   NonLocalDepInfo &PInfo = It->second.NonLocalDeps;
1512 
1513   for (const NonLocalDepEntry &DE : PInfo) {
1514     Instruction *Target = DE.getResult().getInst();
1515     if (!Target)
1516       continue; // Ignore non-local dep results.
1517     assert(Target->getParent() == DE.getBB());
1518 
1519     // Eliminating the dirty entry from 'Cache', so update the reverse info.
1520     RemoveFromReverseMap(ReverseNonLocalPtrDeps, Target, P);
1521   }
1522 
1523   // Remove P from NonLocalPointerDeps (which deletes NonLocalDepInfo).
1524   NonLocalPointerDeps.erase(It);
1525 }
1526 
1527 void MemoryDependenceResults::invalidateCachedPointerInfo(Value *Ptr) {
1528   // If Ptr isn't really a pointer, just ignore it.
1529   if (!Ptr->getType()->isPointerTy())
1530     return;
1531   // Flush store info for the pointer.
1532   removeCachedNonLocalPointerDependencies(ValueIsLoadPair(Ptr, false));
1533   // Flush load info for the pointer.
1534   removeCachedNonLocalPointerDependencies(ValueIsLoadPair(Ptr, true));
1535 }
1536 
1537 void MemoryDependenceResults::invalidateCachedPredecessors() {
1538   PredCache.clear();
1539 }
1540 
1541 void MemoryDependenceResults::removeInstruction(Instruction *RemInst) {
1542   EII.removeInstruction(RemInst);
1543 
1544   // Walk through the Non-local dependencies, removing this one as the value
1545   // for any cached queries.
1546   NonLocalDepMapType::iterator NLDI = NonLocalDepsMap.find(RemInst);
1547   if (NLDI != NonLocalDepsMap.end()) {
1548     NonLocalDepInfo &BlockMap = NLDI->second.first;
1549     for (auto &Entry : BlockMap)
1550       if (Instruction *Inst = Entry.getResult().getInst())
1551         RemoveFromReverseMap(ReverseNonLocalDeps, Inst, RemInst);
1552     NonLocalDepsMap.erase(NLDI);
1553   }
1554 
1555   // If we have a cached local dependence query for this instruction, remove it.
1556   LocalDepMapType::iterator LocalDepEntry = LocalDeps.find(RemInst);
1557   if (LocalDepEntry != LocalDeps.end()) {
1558     // Remove us from DepInst's reverse set now that the local dep info is gone.
1559     if (Instruction *Inst = LocalDepEntry->second.getInst())
1560       RemoveFromReverseMap(ReverseLocalDeps, Inst, RemInst);
1561 
1562     // Remove this local dependency info.
1563     LocalDeps.erase(LocalDepEntry);
1564   }
1565 
1566   // If we have any cached dependencies on this instruction, remove
1567   // them.
1568 
1569   // If the instruction is a pointer, remove it from both the load info and the
1570   // store info.
1571   if (RemInst->getType()->isPointerTy()) {
1572     removeCachedNonLocalPointerDependencies(ValueIsLoadPair(RemInst, false));
1573     removeCachedNonLocalPointerDependencies(ValueIsLoadPair(RemInst, true));
1574   } else {
1575     // Otherwise, if the instructions is in the map directly, it must be a load.
1576     // Remove it.
1577     auto toRemoveIt = NonLocalDefsCache.find(RemInst);
1578     if (toRemoveIt != NonLocalDefsCache.end()) {
1579       assert(isa<LoadInst>(RemInst) &&
1580              "only load instructions should be added directly");
1581       const Instruction *DepV = toRemoveIt->second.getResult().getInst();
1582       ReverseNonLocalDefsCache.find(DepV)->second.erase(RemInst);
1583       NonLocalDefsCache.erase(toRemoveIt);
1584     }
1585   }
1586 
1587   // Loop over all of the things that depend on the instruction we're removing.
1588   SmallVector<std::pair<Instruction *, Instruction *>, 8> ReverseDepsToAdd;
1589 
1590   // If we find RemInst as a clobber or Def in any of the maps for other values,
1591   // we need to replace its entry with a dirty version of the instruction after
1592   // it.  If RemInst is a terminator, we use a null dirty value.
1593   //
1594   // Using a dirty version of the instruction after RemInst saves having to scan
1595   // the entire block to get to this point.
1596   MemDepResult NewDirtyVal;
1597   if (!RemInst->isTerminator())
1598     NewDirtyVal = MemDepResult::getDirty(&*++RemInst->getIterator());
1599 
1600   ReverseDepMapType::iterator ReverseDepIt = ReverseLocalDeps.find(RemInst);
1601   if (ReverseDepIt != ReverseLocalDeps.end()) {
1602     // RemInst can't be the terminator if it has local stuff depending on it.
1603     assert(!ReverseDepIt->second.empty() && !RemInst->isTerminator() &&
1604            "Nothing can locally depend on a terminator");
1605 
1606     for (Instruction *InstDependingOnRemInst : ReverseDepIt->second) {
1607       assert(InstDependingOnRemInst != RemInst &&
1608              "Already removed our local dep info");
1609 
1610       LocalDeps[InstDependingOnRemInst] = NewDirtyVal;
1611 
1612       // Make sure to remember that new things depend on NewDepInst.
1613       assert(NewDirtyVal.getInst() &&
1614              "There is no way something else can have "
1615              "a local dep on this if it is a terminator!");
1616       ReverseDepsToAdd.push_back(
1617           std::make_pair(NewDirtyVal.getInst(), InstDependingOnRemInst));
1618     }
1619 
1620     ReverseLocalDeps.erase(ReverseDepIt);
1621 
1622     // Add new reverse deps after scanning the set, to avoid invalidating the
1623     // 'ReverseDeps' reference.
1624     while (!ReverseDepsToAdd.empty()) {
1625       ReverseLocalDeps[ReverseDepsToAdd.back().first].insert(
1626           ReverseDepsToAdd.back().second);
1627       ReverseDepsToAdd.pop_back();
1628     }
1629   }
1630 
1631   ReverseDepIt = ReverseNonLocalDeps.find(RemInst);
1632   if (ReverseDepIt != ReverseNonLocalDeps.end()) {
1633     for (Instruction *I : ReverseDepIt->second) {
1634       assert(I != RemInst && "Already removed NonLocalDep info for RemInst");
1635 
1636       PerInstNLInfo &INLD = NonLocalDepsMap[I];
1637       // The information is now dirty!
1638       INLD.second = true;
1639 
1640       for (auto &Entry : INLD.first) {
1641         if (Entry.getResult().getInst() != RemInst)
1642           continue;
1643 
1644         // Convert to a dirty entry for the subsequent instruction.
1645         Entry.setResult(NewDirtyVal);
1646 
1647         if (Instruction *NextI = NewDirtyVal.getInst())
1648           ReverseDepsToAdd.push_back(std::make_pair(NextI, I));
1649       }
1650     }
1651 
1652     ReverseNonLocalDeps.erase(ReverseDepIt);
1653 
1654     // Add new reverse deps after scanning the set, to avoid invalidating 'Set'
1655     while (!ReverseDepsToAdd.empty()) {
1656       ReverseNonLocalDeps[ReverseDepsToAdd.back().first].insert(
1657           ReverseDepsToAdd.back().second);
1658       ReverseDepsToAdd.pop_back();
1659     }
1660   }
1661 
1662   // If the instruction is in ReverseNonLocalPtrDeps then it appears as a
1663   // value in the NonLocalPointerDeps info.
1664   ReverseNonLocalPtrDepTy::iterator ReversePtrDepIt =
1665       ReverseNonLocalPtrDeps.find(RemInst);
1666   if (ReversePtrDepIt != ReverseNonLocalPtrDeps.end()) {
1667     SmallVector<std::pair<Instruction *, ValueIsLoadPair>, 8>
1668         ReversePtrDepsToAdd;
1669 
1670     for (ValueIsLoadPair P : ReversePtrDepIt->second) {
1671       assert(P.getPointer() != RemInst &&
1672              "Already removed NonLocalPointerDeps info for RemInst");
1673 
1674       NonLocalDepInfo &NLPDI = NonLocalPointerDeps[P].NonLocalDeps;
1675 
1676       // The cache is not valid for any specific block anymore.
1677       NonLocalPointerDeps[P].Pair = BBSkipFirstBlockPair();
1678 
1679       // Update any entries for RemInst to use the instruction after it.
1680       for (auto &Entry : NLPDI) {
1681         if (Entry.getResult().getInst() != RemInst)
1682           continue;
1683 
1684         // Convert to a dirty entry for the subsequent instruction.
1685         Entry.setResult(NewDirtyVal);
1686 
1687         if (Instruction *NewDirtyInst = NewDirtyVal.getInst())
1688           ReversePtrDepsToAdd.push_back(std::make_pair(NewDirtyInst, P));
1689       }
1690 
1691       // Re-sort the NonLocalDepInfo.  Changing the dirty entry to its
1692       // subsequent value may invalidate the sortedness.
1693       llvm::sort(NLPDI);
1694     }
1695 
1696     ReverseNonLocalPtrDeps.erase(ReversePtrDepIt);
1697 
1698     while (!ReversePtrDepsToAdd.empty()) {
1699       ReverseNonLocalPtrDeps[ReversePtrDepsToAdd.back().first].insert(
1700           ReversePtrDepsToAdd.back().second);
1701       ReversePtrDepsToAdd.pop_back();
1702     }
1703   }
1704 
1705   assert(!NonLocalDepsMap.count(RemInst) && "RemInst got reinserted?");
1706   LLVM_DEBUG(verifyRemoved(RemInst));
1707 }
1708 
1709 /// Verify that the specified instruction does not occur in our internal data
1710 /// structures.
1711 ///
1712 /// This function verifies by asserting in debug builds.
1713 void MemoryDependenceResults::verifyRemoved(Instruction *D) const {
1714 #ifndef NDEBUG
1715   for (const auto &DepKV : LocalDeps) {
1716     assert(DepKV.first != D && "Inst occurs in data structures");
1717     assert(DepKV.second.getInst() != D && "Inst occurs in data structures");
1718   }
1719 
1720   for (const auto &DepKV : NonLocalPointerDeps) {
1721     assert(DepKV.first.getPointer() != D && "Inst occurs in NLPD map key");
1722     for (const auto &Entry : DepKV.second.NonLocalDeps)
1723       assert(Entry.getResult().getInst() != D && "Inst occurs as NLPD value");
1724   }
1725 
1726   for (const auto &DepKV : NonLocalDepsMap) {
1727     assert(DepKV.first != D && "Inst occurs in data structures");
1728     const PerInstNLInfo &INLD = DepKV.second;
1729     for (const auto &Entry : INLD.first)
1730       assert(Entry.getResult().getInst() != D &&
1731              "Inst occurs in data structures");
1732   }
1733 
1734   for (const auto &DepKV : ReverseLocalDeps) {
1735     assert(DepKV.first != D && "Inst occurs in data structures");
1736     for (Instruction *Inst : DepKV.second)
1737       assert(Inst != D && "Inst occurs in data structures");
1738   }
1739 
1740   for (const auto &DepKV : ReverseNonLocalDeps) {
1741     assert(DepKV.first != D && "Inst occurs in data structures");
1742     for (Instruction *Inst : DepKV.second)
1743       assert(Inst != D && "Inst occurs in data structures");
1744   }
1745 
1746   for (const auto &DepKV : ReverseNonLocalPtrDeps) {
1747     assert(DepKV.first != D && "Inst occurs in rev NLPD map");
1748 
1749     for (ValueIsLoadPair P : DepKV.second)
1750       assert(P != ValueIsLoadPair(D, false) && P != ValueIsLoadPair(D, true) &&
1751              "Inst occurs in ReverseNonLocalPtrDeps map");
1752   }
1753 #endif
1754 }
1755 
1756 AnalysisKey MemoryDependenceAnalysis::Key;
1757 
1758 MemoryDependenceAnalysis::MemoryDependenceAnalysis()
1759     : DefaultBlockScanLimit(BlockScanLimit) {}
1760 
1761 MemoryDependenceResults
1762 MemoryDependenceAnalysis::run(Function &F, FunctionAnalysisManager &AM) {
1763   auto &AA = AM.getResult<AAManager>(F);
1764   auto &AC = AM.getResult<AssumptionAnalysis>(F);
1765   auto &TLI = AM.getResult<TargetLibraryAnalysis>(F);
1766   auto &DT = AM.getResult<DominatorTreeAnalysis>(F);
1767   return MemoryDependenceResults(AA, AC, TLI, DT, DefaultBlockScanLimit);
1768 }
1769 
1770 char MemoryDependenceWrapperPass::ID = 0;
1771 
1772 INITIALIZE_PASS_BEGIN(MemoryDependenceWrapperPass, "memdep",
1773                       "Memory Dependence Analysis", false, true)
1774 INITIALIZE_PASS_DEPENDENCY(AssumptionCacheTracker)
1775 INITIALIZE_PASS_DEPENDENCY(AAResultsWrapperPass)
1776 INITIALIZE_PASS_DEPENDENCY(DominatorTreeWrapperPass)
1777 INITIALIZE_PASS_DEPENDENCY(TargetLibraryInfoWrapperPass)
1778 INITIALIZE_PASS_END(MemoryDependenceWrapperPass, "memdep",
1779                     "Memory Dependence Analysis", false, true)
1780 
1781 MemoryDependenceWrapperPass::MemoryDependenceWrapperPass() : FunctionPass(ID) {
1782   initializeMemoryDependenceWrapperPassPass(*PassRegistry::getPassRegistry());
1783 }
1784 
1785 MemoryDependenceWrapperPass::~MemoryDependenceWrapperPass() = default;
1786 
1787 void MemoryDependenceWrapperPass::releaseMemory() {
1788   MemDep.reset();
1789 }
1790 
1791 void MemoryDependenceWrapperPass::getAnalysisUsage(AnalysisUsage &AU) const {
1792   AU.setPreservesAll();
1793   AU.addRequired<AssumptionCacheTracker>();
1794   AU.addRequired<DominatorTreeWrapperPass>();
1795   AU.addRequiredTransitive<AAResultsWrapperPass>();
1796   AU.addRequiredTransitive<TargetLibraryInfoWrapperPass>();
1797 }
1798 
1799 bool MemoryDependenceResults::invalidate(Function &F, const PreservedAnalyses &PA,
1800                                FunctionAnalysisManager::Invalidator &Inv) {
1801   // Check whether our analysis is preserved.
1802   auto PAC = PA.getChecker<MemoryDependenceAnalysis>();
1803   if (!PAC.preserved() && !PAC.preservedSet<AllAnalysesOn<Function>>())
1804     // If not, give up now.
1805     return true;
1806 
1807   // Check whether the analyses we depend on became invalid for any reason.
1808   if (Inv.invalidate<AAManager>(F, PA) ||
1809       Inv.invalidate<AssumptionAnalysis>(F, PA) ||
1810       Inv.invalidate<DominatorTreeAnalysis>(F, PA))
1811     return true;
1812 
1813   // Otherwise this analysis result remains valid.
1814   return false;
1815 }
1816 
1817 unsigned MemoryDependenceResults::getDefaultBlockScanLimit() const {
1818   return DefaultBlockScanLimit;
1819 }
1820 
1821 bool MemoryDependenceWrapperPass::runOnFunction(Function &F) {
1822   auto &AA = getAnalysis<AAResultsWrapperPass>().getAAResults();
1823   auto &AC = getAnalysis<AssumptionCacheTracker>().getAssumptionCache(F);
1824   auto &TLI = getAnalysis<TargetLibraryInfoWrapperPass>().getTLI(F);
1825   auto &DT = getAnalysis<DominatorTreeWrapperPass>().getDomTree();
1826   MemDep.emplace(AA, AC, TLI, DT, BlockScanLimit);
1827   return false;
1828 }
1829