1 //===- MemoryDependenceAnalysis.cpp - Mem Deps Implementation -------------===//
2 //
3 // The LLVM Compiler Infrastructure
4 //
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
7 //
8 //===----------------------------------------------------------------------===//
9 //
10 // This file implements an analysis that determines, for a given memory
11 // operation, what preceding memory operations it depends on. It builds on
12 // alias analysis information, and tries to provide a lazy, caching interface to
13 // a common kind of alias information query.
14 //
15 //===----------------------------------------------------------------------===//
16
17 #include "llvm/Analysis/MemoryDependenceAnalysis.h"
18 #include "llvm/ADT/STLExtras.h"
19 #include "llvm/ADT/Statistic.h"
20 #include "llvm/Analysis/AliasAnalysis.h"
21 #include "llvm/Analysis/AssumptionCache.h"
22 #include "llvm/Analysis/InstructionSimplify.h"
23 #include "llvm/Analysis/MemoryBuiltins.h"
24 #include "llvm/Analysis/PHITransAddr.h"
25 #include "llvm/Analysis/OrderedBasicBlock.h"
26 #include "llvm/Analysis/ValueTracking.h"
27 #include "llvm/Analysis/TargetLibraryInfo.h"
28 #include "llvm/IR/DataLayout.h"
29 #include "llvm/IR/Dominators.h"
30 #include "llvm/IR/Function.h"
31 #include "llvm/IR/Instructions.h"
32 #include "llvm/IR/IntrinsicInst.h"
33 #include "llvm/IR/LLVMContext.h"
34 #include "llvm/IR/PredIteratorCache.h"
35 #include "llvm/Support/Debug.h"
36 using namespace llvm;
37
38 #define DEBUG_TYPE "memdep"
39
40 STATISTIC(NumCacheNonLocal, "Number of fully cached non-local responses");
41 STATISTIC(NumCacheDirtyNonLocal, "Number of dirty cached non-local responses");
42 STATISTIC(NumUncacheNonLocal, "Number of uncached non-local responses");
43
44 STATISTIC(NumCacheNonLocalPtr,
45 "Number of fully cached non-local ptr responses");
46 STATISTIC(NumCacheDirtyNonLocalPtr,
47 "Number of cached, but dirty, non-local ptr responses");
48 STATISTIC(NumUncacheNonLocalPtr, "Number of uncached non-local ptr responses");
49 STATISTIC(NumCacheCompleteNonLocalPtr,
50 "Number of block queries that were completely cached");
51
52 // Limit for the number of instructions to scan in a block.
53
54 static cl::opt<unsigned> BlockScanLimit(
55 "memdep-block-scan-limit", cl::Hidden, cl::init(100),
56 cl::desc("The number of instructions to scan in a block in memory "
57 "dependency analysis (default = 100)"));
58
59 static cl::opt<unsigned>
60 BlockNumberLimit("memdep-block-number-limit", cl::Hidden, cl::init(1000),
61 cl::desc("The number of blocks to scan during memory "
62 "dependency analysis (default = 1000)"));
63
64 // Limit on the number of memdep results to process.
65 static const unsigned int NumResultsLimit = 100;
66
67 /// This is a helper function that removes Val from 'Inst's set in ReverseMap.
68 ///
69 /// If the set becomes empty, remove Inst's entry.
70 template <typename KeyTy>
71 static void
RemoveFromReverseMap(DenseMap<Instruction *,SmallPtrSet<KeyTy,4>> & ReverseMap,Instruction * Inst,KeyTy Val)72 RemoveFromReverseMap(DenseMap<Instruction *, SmallPtrSet<KeyTy, 4>> &ReverseMap,
73 Instruction *Inst, KeyTy Val) {
74 typename DenseMap<Instruction *, SmallPtrSet<KeyTy, 4>>::iterator InstIt =
75 ReverseMap.find(Inst);
76 assert(InstIt != ReverseMap.end() && "Reverse map out of sync?");
77 bool Found = InstIt->second.erase(Val);
78 assert(Found && "Invalid reverse map!");
79 (void)Found;
80 if (InstIt->second.empty())
81 ReverseMap.erase(InstIt);
82 }
83
84 /// If the given instruction references a specific memory location, fill in Loc
85 /// with the details, otherwise set Loc.Ptr to null.
86 ///
87 /// Returns a ModRefInfo value describing the general behavior of the
88 /// instruction.
GetLocation(const Instruction * Inst,MemoryLocation & Loc,const TargetLibraryInfo & TLI)89 static ModRefInfo GetLocation(const Instruction *Inst, MemoryLocation &Loc,
90 const TargetLibraryInfo &TLI) {
91 if (const LoadInst *LI = dyn_cast<LoadInst>(Inst)) {
92 if (LI->isUnordered()) {
93 Loc = MemoryLocation::get(LI);
94 return MRI_Ref;
95 }
96 if (LI->getOrdering() == AtomicOrdering::Monotonic) {
97 Loc = MemoryLocation::get(LI);
98 return MRI_ModRef;
99 }
100 Loc = MemoryLocation();
101 return MRI_ModRef;
102 }
103
104 if (const StoreInst *SI = dyn_cast<StoreInst>(Inst)) {
105 if (SI->isUnordered()) {
106 Loc = MemoryLocation::get(SI);
107 return MRI_Mod;
108 }
109 if (SI->getOrdering() == AtomicOrdering::Monotonic) {
110 Loc = MemoryLocation::get(SI);
111 return MRI_ModRef;
112 }
113 Loc = MemoryLocation();
114 return MRI_ModRef;
115 }
116
117 if (const VAArgInst *V = dyn_cast<VAArgInst>(Inst)) {
118 Loc = MemoryLocation::get(V);
119 return MRI_ModRef;
120 }
121
122 if (const CallInst *CI = isFreeCall(Inst, &TLI)) {
123 // calls to free() deallocate the entire structure
124 Loc = MemoryLocation(CI->getArgOperand(0));
125 return MRI_Mod;
126 }
127
128 if (const IntrinsicInst *II = dyn_cast<IntrinsicInst>(Inst)) {
129 AAMDNodes AAInfo;
130
131 switch (II->getIntrinsicID()) {
132 case Intrinsic::lifetime_start:
133 case Intrinsic::lifetime_end:
134 case Intrinsic::invariant_start:
135 II->getAAMetadata(AAInfo);
136 Loc = MemoryLocation(
137 II->getArgOperand(1),
138 cast<ConstantInt>(II->getArgOperand(0))->getZExtValue(), AAInfo);
139 // These intrinsics don't really modify the memory, but returning Mod
140 // will allow them to be handled conservatively.
141 return MRI_Mod;
142 case Intrinsic::invariant_end:
143 II->getAAMetadata(AAInfo);
144 Loc = MemoryLocation(
145 II->getArgOperand(2),
146 cast<ConstantInt>(II->getArgOperand(1))->getZExtValue(), AAInfo);
147 // These intrinsics don't really modify the memory, but returning Mod
148 // will allow them to be handled conservatively.
149 return MRI_Mod;
150 default:
151 break;
152 }
153 }
154
155 // Otherwise, just do the coarse-grained thing that always works.
156 if (Inst->mayWriteToMemory())
157 return MRI_ModRef;
158 if (Inst->mayReadFromMemory())
159 return MRI_Ref;
160 return MRI_NoModRef;
161 }
162
163 /// Private helper for finding the local dependencies of a call site.
getCallSiteDependencyFrom(CallSite CS,bool isReadOnlyCall,BasicBlock::iterator ScanIt,BasicBlock * BB)164 MemDepResult MemoryDependenceResults::getCallSiteDependencyFrom(
165 CallSite CS, bool isReadOnlyCall, BasicBlock::iterator ScanIt,
166 BasicBlock *BB) {
167 unsigned Limit = BlockScanLimit;
168
169 // Walk backwards through the block, looking for dependencies
170 while (ScanIt != BB->begin()) {
171 // Limit the amount of scanning we do so we don't end up with quadratic
172 // running time on extreme testcases.
173 --Limit;
174 if (!Limit)
175 return MemDepResult::getUnknown();
176
177 Instruction *Inst = &*--ScanIt;
178
179 // If this inst is a memory op, get the pointer it accessed
180 MemoryLocation Loc;
181 ModRefInfo MR = GetLocation(Inst, Loc, TLI);
182 if (Loc.Ptr) {
183 // A simple instruction.
184 if (AA.getModRefInfo(CS, Loc) != MRI_NoModRef)
185 return MemDepResult::getClobber(Inst);
186 continue;
187 }
188
189 if (auto InstCS = CallSite(Inst)) {
190 // Debug intrinsics don't cause dependences.
191 if (isa<DbgInfoIntrinsic>(Inst))
192 continue;
193 // If these two calls do not interfere, look past it.
194 switch (AA.getModRefInfo(CS, InstCS)) {
195 case MRI_NoModRef:
196 // If the two calls are the same, return InstCS as a Def, so that
197 // CS can be found redundant and eliminated.
198 if (isReadOnlyCall && !(MR & MRI_Mod) &&
199 CS.getInstruction()->isIdenticalToWhenDefined(Inst))
200 return MemDepResult::getDef(Inst);
201
202 // Otherwise if the two calls don't interact (e.g. InstCS is readnone)
203 // keep scanning.
204 continue;
205 default:
206 return MemDepResult::getClobber(Inst);
207 }
208 }
209
210 // If we could not obtain a pointer for the instruction and the instruction
211 // touches memory then assume that this is a dependency.
212 if (MR != MRI_NoModRef)
213 return MemDepResult::getClobber(Inst);
214 }
215
216 // No dependence found. If this is the entry block of the function, it is
217 // unknown, otherwise it is non-local.
218 if (BB != &BB->getParent()->getEntryBlock())
219 return MemDepResult::getNonLocal();
220 return MemDepResult::getNonFuncLocal();
221 }
222
223 /// Return true if LI is a load that would fully overlap MemLoc if done as
224 /// a wider legal integer load.
225 ///
226 /// MemLocBase, MemLocOffset are lazily computed here the first time the
227 /// base/offs of memloc is needed.
isLoadLoadClobberIfExtendedToFullWidth(const MemoryLocation & MemLoc,const Value * & MemLocBase,int64_t & MemLocOffs,const LoadInst * LI)228 static bool isLoadLoadClobberIfExtendedToFullWidth(const MemoryLocation &MemLoc,
229 const Value *&MemLocBase,
230 int64_t &MemLocOffs,
231 const LoadInst *LI) {
232 const DataLayout &DL = LI->getModule()->getDataLayout();
233
234 // If we haven't already computed the base/offset of MemLoc, do so now.
235 if (!MemLocBase)
236 MemLocBase = GetPointerBaseWithConstantOffset(MemLoc.Ptr, MemLocOffs, DL);
237
238 unsigned Size = MemoryDependenceResults::getLoadLoadClobberFullWidthSize(
239 MemLocBase, MemLocOffs, MemLoc.Size, LI);
240 return Size != 0;
241 }
242
getLoadLoadClobberFullWidthSize(const Value * MemLocBase,int64_t MemLocOffs,unsigned MemLocSize,const LoadInst * LI)243 unsigned MemoryDependenceResults::getLoadLoadClobberFullWidthSize(
244 const Value *MemLocBase, int64_t MemLocOffs, unsigned MemLocSize,
245 const LoadInst *LI) {
246 // We can only extend simple integer loads.
247 if (!isa<IntegerType>(LI->getType()) || !LI->isSimple())
248 return 0;
249
250 // Load widening is hostile to ThreadSanitizer: it may cause false positives
251 // or make the reports more cryptic (access sizes are wrong).
252 if (LI->getParent()->getParent()->hasFnAttribute(Attribute::SanitizeThread))
253 return 0;
254
255 const DataLayout &DL = LI->getModule()->getDataLayout();
256
257 // Get the base of this load.
258 int64_t LIOffs = 0;
259 const Value *LIBase =
260 GetPointerBaseWithConstantOffset(LI->getPointerOperand(), LIOffs, DL);
261
262 // If the two pointers are not based on the same pointer, we can't tell that
263 // they are related.
264 if (LIBase != MemLocBase)
265 return 0;
266
267 // Okay, the two values are based on the same pointer, but returned as
268 // no-alias. This happens when we have things like two byte loads at "P+1"
269 // and "P+3". Check to see if increasing the size of the "LI" load up to its
270 // alignment (or the largest native integer type) will allow us to load all
271 // the bits required by MemLoc.
272
273 // If MemLoc is before LI, then no widening of LI will help us out.
274 if (MemLocOffs < LIOffs)
275 return 0;
276
277 // Get the alignment of the load in bytes. We assume that it is safe to load
278 // any legal integer up to this size without a problem. For example, if we're
279 // looking at an i8 load on x86-32 that is known 1024 byte aligned, we can
280 // widen it up to an i32 load. If it is known 2-byte aligned, we can widen it
281 // to i16.
282 unsigned LoadAlign = LI->getAlignment();
283
284 int64_t MemLocEnd = MemLocOffs + MemLocSize;
285
286 // If no amount of rounding up will let MemLoc fit into LI, then bail out.
287 if (LIOffs + LoadAlign < MemLocEnd)
288 return 0;
289
290 // This is the size of the load to try. Start with the next larger power of
291 // two.
292 unsigned NewLoadByteSize = LI->getType()->getPrimitiveSizeInBits() / 8U;
293 NewLoadByteSize = NextPowerOf2(NewLoadByteSize);
294
295 while (1) {
296 // If this load size is bigger than our known alignment or would not fit
297 // into a native integer register, then we fail.
298 if (NewLoadByteSize > LoadAlign ||
299 !DL.fitsInLegalInteger(NewLoadByteSize * 8))
300 return 0;
301
302 if (LIOffs + NewLoadByteSize > MemLocEnd &&
303 LI->getParent()->getParent()->hasFnAttribute(
304 Attribute::SanitizeAddress))
305 // We will be reading past the location accessed by the original program.
306 // While this is safe in a regular build, Address Safety analysis tools
307 // may start reporting false warnings. So, don't do widening.
308 return 0;
309
310 // If a load of this width would include all of MemLoc, then we succeed.
311 if (LIOffs + NewLoadByteSize >= MemLocEnd)
312 return NewLoadByteSize;
313
314 NewLoadByteSize <<= 1;
315 }
316 }
317
isVolatile(Instruction * Inst)318 static bool isVolatile(Instruction *Inst) {
319 if (LoadInst *LI = dyn_cast<LoadInst>(Inst))
320 return LI->isVolatile();
321 else if (StoreInst *SI = dyn_cast<StoreInst>(Inst))
322 return SI->isVolatile();
323 else if (AtomicCmpXchgInst *AI = dyn_cast<AtomicCmpXchgInst>(Inst))
324 return AI->isVolatile();
325 return false;
326 }
327
getPointerDependencyFrom(const MemoryLocation & MemLoc,bool isLoad,BasicBlock::iterator ScanIt,BasicBlock * BB,Instruction * QueryInst)328 MemDepResult MemoryDependenceResults::getPointerDependencyFrom(
329 const MemoryLocation &MemLoc, bool isLoad, BasicBlock::iterator ScanIt,
330 BasicBlock *BB, Instruction *QueryInst) {
331
332 if (QueryInst != nullptr) {
333 if (auto *LI = dyn_cast<LoadInst>(QueryInst)) {
334 MemDepResult invariantGroupDependency =
335 getInvariantGroupPointerDependency(LI, BB);
336
337 if (invariantGroupDependency.isDef())
338 return invariantGroupDependency;
339 }
340 }
341 return getSimplePointerDependencyFrom(MemLoc, isLoad, ScanIt, BB, QueryInst);
342 }
343
344 MemDepResult
getInvariantGroupPointerDependency(LoadInst * LI,BasicBlock * BB)345 MemoryDependenceResults::getInvariantGroupPointerDependency(LoadInst *LI,
346 BasicBlock *BB) {
347 Value *LoadOperand = LI->getPointerOperand();
348 // It's is not safe to walk the use list of global value, because function
349 // passes aren't allowed to look outside their functions.
350 if (isa<GlobalValue>(LoadOperand))
351 return MemDepResult::getUnknown();
352
353 auto *InvariantGroupMD = LI->getMetadata(LLVMContext::MD_invariant_group);
354 if (!InvariantGroupMD)
355 return MemDepResult::getUnknown();
356
357 MemDepResult Result = MemDepResult::getUnknown();
358 llvm::SmallSet<Value *, 14> Seen;
359 // Queue to process all pointers that are equivalent to load operand.
360 llvm::SmallVector<Value *, 8> LoadOperandsQueue;
361 LoadOperandsQueue.push_back(LoadOperand);
362 while (!LoadOperandsQueue.empty()) {
363 Value *Ptr = LoadOperandsQueue.pop_back_val();
364 if (isa<GlobalValue>(Ptr))
365 continue;
366
367 if (auto *BCI = dyn_cast<BitCastInst>(Ptr)) {
368 if (Seen.insert(BCI->getOperand(0)).second) {
369 LoadOperandsQueue.push_back(BCI->getOperand(0));
370 }
371 }
372
373 for (Use &Us : Ptr->uses()) {
374 auto *U = dyn_cast<Instruction>(Us.getUser());
375 if (!U || U == LI || !DT.dominates(U, LI))
376 continue;
377
378 if (auto *BCI = dyn_cast<BitCastInst>(U)) {
379 if (Seen.insert(BCI).second) {
380 LoadOperandsQueue.push_back(BCI);
381 }
382 continue;
383 }
384 // If we hit load/store with the same invariant.group metadata (and the
385 // same pointer operand) we can assume that value pointed by pointer
386 // operand didn't change.
387 if ((isa<LoadInst>(U) || isa<StoreInst>(U)) && U->getParent() == BB &&
388 U->getMetadata(LLVMContext::MD_invariant_group) == InvariantGroupMD)
389 return MemDepResult::getDef(U);
390 }
391 }
392 return Result;
393 }
394
getSimplePointerDependencyFrom(const MemoryLocation & MemLoc,bool isLoad,BasicBlock::iterator ScanIt,BasicBlock * BB,Instruction * QueryInst)395 MemDepResult MemoryDependenceResults::getSimplePointerDependencyFrom(
396 const MemoryLocation &MemLoc, bool isLoad, BasicBlock::iterator ScanIt,
397 BasicBlock *BB, Instruction *QueryInst) {
398
399 const Value *MemLocBase = nullptr;
400 int64_t MemLocOffset = 0;
401 unsigned Limit = BlockScanLimit;
402 bool isInvariantLoad = false;
403
404 // We must be careful with atomic accesses, as they may allow another thread
405 // to touch this location, clobbering it. We are conservative: if the
406 // QueryInst is not a simple (non-atomic) memory access, we automatically
407 // return getClobber.
408 // If it is simple, we know based on the results of
409 // "Compiler testing via a theory of sound optimisations in the C11/C++11
410 // memory model" in PLDI 2013, that a non-atomic location can only be
411 // clobbered between a pair of a release and an acquire action, with no
412 // access to the location in between.
413 // Here is an example for giving the general intuition behind this rule.
414 // In the following code:
415 // store x 0;
416 // release action; [1]
417 // acquire action; [4]
418 // %val = load x;
419 // It is unsafe to replace %val by 0 because another thread may be running:
420 // acquire action; [2]
421 // store x 42;
422 // release action; [3]
423 // with synchronization from 1 to 2 and from 3 to 4, resulting in %val
424 // being 42. A key property of this program however is that if either
425 // 1 or 4 were missing, there would be a race between the store of 42
426 // either the store of 0 or the load (making the whole program racy).
427 // The paper mentioned above shows that the same property is respected
428 // by every program that can detect any optimization of that kind: either
429 // it is racy (undefined) or there is a release followed by an acquire
430 // between the pair of accesses under consideration.
431
432 // If the load is invariant, we "know" that it doesn't alias *any* write. We
433 // do want to respect mustalias results since defs are useful for value
434 // forwarding, but any mayalias write can be assumed to be noalias.
435 // Arguably, this logic should be pushed inside AliasAnalysis itself.
436 if (isLoad && QueryInst) {
437 LoadInst *LI = dyn_cast<LoadInst>(QueryInst);
438 if (LI && LI->getMetadata(LLVMContext::MD_invariant_load) != nullptr)
439 isInvariantLoad = true;
440 }
441
442 const DataLayout &DL = BB->getModule()->getDataLayout();
443
444 // Create a numbered basic block to lazily compute and cache instruction
445 // positions inside a BB. This is used to provide fast queries for relative
446 // position between two instructions in a BB and can be used by
447 // AliasAnalysis::callCapturesBefore.
448 OrderedBasicBlock OBB(BB);
449
450 // Return "true" if and only if the instruction I is either a non-simple
451 // load or a non-simple store.
452 auto isNonSimpleLoadOrStore = [](Instruction *I) -> bool {
453 if (auto *LI = dyn_cast<LoadInst>(I))
454 return !LI->isSimple();
455 if (auto *SI = dyn_cast<StoreInst>(I))
456 return !SI->isSimple();
457 return false;
458 };
459
460 // Return "true" if I is not a load and not a store, but it does access
461 // memory.
462 auto isOtherMemAccess = [](Instruction *I) -> bool {
463 return !isa<LoadInst>(I) && !isa<StoreInst>(I) && I->mayReadOrWriteMemory();
464 };
465
466 // Walk backwards through the basic block, looking for dependencies.
467 while (ScanIt != BB->begin()) {
468 Instruction *Inst = &*--ScanIt;
469
470 if (IntrinsicInst *II = dyn_cast<IntrinsicInst>(Inst))
471 // Debug intrinsics don't (and can't) cause dependencies.
472 if (isa<DbgInfoIntrinsic>(II))
473 continue;
474
475 // Limit the amount of scanning we do so we don't end up with quadratic
476 // running time on extreme testcases.
477 --Limit;
478 if (!Limit)
479 return MemDepResult::getUnknown();
480
481 if (IntrinsicInst *II = dyn_cast<IntrinsicInst>(Inst)) {
482 // If we reach a lifetime begin or end marker, then the query ends here
483 // because the value is undefined.
484 if (II->getIntrinsicID() == Intrinsic::lifetime_start) {
485 // FIXME: This only considers queries directly on the invariant-tagged
486 // pointer, not on query pointers that are indexed off of them. It'd
487 // be nice to handle that at some point (the right approach is to use
488 // GetPointerBaseWithConstantOffset).
489 if (AA.isMustAlias(MemoryLocation(II->getArgOperand(1)), MemLoc))
490 return MemDepResult::getDef(II);
491 continue;
492 }
493 }
494
495 // Values depend on loads if the pointers are must aliased. This means
496 // that a load depends on another must aliased load from the same value.
497 // One exception is atomic loads: a value can depend on an atomic load that
498 // it does not alias with when this atomic load indicates that another
499 // thread may be accessing the location.
500 if (LoadInst *LI = dyn_cast<LoadInst>(Inst)) {
501
502 // While volatile access cannot be eliminated, they do not have to clobber
503 // non-aliasing locations, as normal accesses, for example, can be safely
504 // reordered with volatile accesses.
505 if (LI->isVolatile()) {
506 if (!QueryInst)
507 // Original QueryInst *may* be volatile
508 return MemDepResult::getClobber(LI);
509 if (isVolatile(QueryInst))
510 // Ordering required if QueryInst is itself volatile
511 return MemDepResult::getClobber(LI);
512 // Otherwise, volatile doesn't imply any special ordering
513 }
514
515 // Atomic loads have complications involved.
516 // A Monotonic (or higher) load is OK if the query inst is itself not
517 // atomic.
518 // FIXME: This is overly conservative.
519 if (LI->isAtomic() && isStrongerThanUnordered(LI->getOrdering())) {
520 if (!QueryInst || isNonSimpleLoadOrStore(QueryInst) ||
521 isOtherMemAccess(QueryInst))
522 return MemDepResult::getClobber(LI);
523 if (LI->getOrdering() != AtomicOrdering::Monotonic)
524 return MemDepResult::getClobber(LI);
525 }
526
527 MemoryLocation LoadLoc = MemoryLocation::get(LI);
528
529 // If we found a pointer, check if it could be the same as our pointer.
530 AliasResult R = AA.alias(LoadLoc, MemLoc);
531
532 if (isLoad) {
533 if (R == NoAlias) {
534 // If this is an over-aligned integer load (for example,
535 // "load i8* %P, align 4") see if it would obviously overlap with the
536 // queried location if widened to a larger load (e.g. if the queried
537 // location is 1 byte at P+1). If so, return it as a load/load
538 // clobber result, allowing the client to decide to widen the load if
539 // it wants to.
540 if (IntegerType *ITy = dyn_cast<IntegerType>(LI->getType())) {
541 if (LI->getAlignment() * 8 > ITy->getPrimitiveSizeInBits() &&
542 isLoadLoadClobberIfExtendedToFullWidth(MemLoc, MemLocBase,
543 MemLocOffset, LI))
544 return MemDepResult::getClobber(Inst);
545 }
546 continue;
547 }
548
549 // Must aliased loads are defs of each other.
550 if (R == MustAlias)
551 return MemDepResult::getDef(Inst);
552
553 #if 0 // FIXME: Temporarily disabled. GVN is cleverly rewriting loads
554 // in terms of clobbering loads, but since it does this by looking
555 // at the clobbering load directly, it doesn't know about any
556 // phi translation that may have happened along the way.
557
558 // If we have a partial alias, then return this as a clobber for the
559 // client to handle.
560 if (R == PartialAlias)
561 return MemDepResult::getClobber(Inst);
562 #endif
563
564 // Random may-alias loads don't depend on each other without a
565 // dependence.
566 continue;
567 }
568
569 // Stores don't depend on other no-aliased accesses.
570 if (R == NoAlias)
571 continue;
572
573 // Stores don't alias loads from read-only memory.
574 if (AA.pointsToConstantMemory(LoadLoc))
575 continue;
576
577 // Stores depend on may/must aliased loads.
578 return MemDepResult::getDef(Inst);
579 }
580
581 if (StoreInst *SI = dyn_cast<StoreInst>(Inst)) {
582 // Atomic stores have complications involved.
583 // A Monotonic store is OK if the query inst is itself not atomic.
584 // FIXME: This is overly conservative.
585 if (!SI->isUnordered() && SI->isAtomic()) {
586 if (!QueryInst || isNonSimpleLoadOrStore(QueryInst) ||
587 isOtherMemAccess(QueryInst))
588 return MemDepResult::getClobber(SI);
589 if (SI->getOrdering() != AtomicOrdering::Monotonic)
590 return MemDepResult::getClobber(SI);
591 }
592
593 // FIXME: this is overly conservative.
594 // While volatile access cannot be eliminated, they do not have to clobber
595 // non-aliasing locations, as normal accesses can for example be reordered
596 // with volatile accesses.
597 if (SI->isVolatile())
598 if (!QueryInst || isNonSimpleLoadOrStore(QueryInst) ||
599 isOtherMemAccess(QueryInst))
600 return MemDepResult::getClobber(SI);
601
602 // If alias analysis can tell that this store is guaranteed to not modify
603 // the query pointer, ignore it. Use getModRefInfo to handle cases where
604 // the query pointer points to constant memory etc.
605 if (AA.getModRefInfo(SI, MemLoc) == MRI_NoModRef)
606 continue;
607
608 // Ok, this store might clobber the query pointer. Check to see if it is
609 // a must alias: in this case, we want to return this as a def.
610 MemoryLocation StoreLoc = MemoryLocation::get(SI);
611
612 // If we found a pointer, check if it could be the same as our pointer.
613 AliasResult R = AA.alias(StoreLoc, MemLoc);
614
615 if (R == NoAlias)
616 continue;
617 if (R == MustAlias)
618 return MemDepResult::getDef(Inst);
619 if (isInvariantLoad)
620 continue;
621 return MemDepResult::getClobber(Inst);
622 }
623
624 // If this is an allocation, and if we know that the accessed pointer is to
625 // the allocation, return Def. This means that there is no dependence and
626 // the access can be optimized based on that. For example, a load could
627 // turn into undef. Note that we can bypass the allocation itself when
628 // looking for a clobber in many cases; that's an alias property and is
629 // handled by BasicAA.
630 if (isa<AllocaInst>(Inst) || isNoAliasFn(Inst, &TLI)) {
631 const Value *AccessPtr = GetUnderlyingObject(MemLoc.Ptr, DL);
632 if (AccessPtr == Inst || AA.isMustAlias(Inst, AccessPtr))
633 return MemDepResult::getDef(Inst);
634 }
635
636 if (isInvariantLoad)
637 continue;
638
639 // A release fence requires that all stores complete before it, but does
640 // not prevent the reordering of following loads or stores 'before' the
641 // fence. As a result, we look past it when finding a dependency for
642 // loads. DSE uses this to find preceeding stores to delete and thus we
643 // can't bypass the fence if the query instruction is a store.
644 if (FenceInst *FI = dyn_cast<FenceInst>(Inst))
645 if (isLoad && FI->getOrdering() == AtomicOrdering::Release)
646 continue;
647
648 // See if this instruction (e.g. a call or vaarg) mod/ref's the pointer.
649 ModRefInfo MR = AA.getModRefInfo(Inst, MemLoc);
650 // If necessary, perform additional analysis.
651 if (MR == MRI_ModRef)
652 MR = AA.callCapturesBefore(Inst, MemLoc, &DT, &OBB);
653 switch (MR) {
654 case MRI_NoModRef:
655 // If the call has no effect on the queried pointer, just ignore it.
656 continue;
657 case MRI_Mod:
658 return MemDepResult::getClobber(Inst);
659 case MRI_Ref:
660 // If the call is known to never store to the pointer, and if this is a
661 // load query, we can safely ignore it (scan past it).
662 if (isLoad)
663 continue;
664 default:
665 // Otherwise, there is a potential dependence. Return a clobber.
666 return MemDepResult::getClobber(Inst);
667 }
668 }
669
670 // No dependence found. If this is the entry block of the function, it is
671 // unknown, otherwise it is non-local.
672 if (BB != &BB->getParent()->getEntryBlock())
673 return MemDepResult::getNonLocal();
674 return MemDepResult::getNonFuncLocal();
675 }
676
getDependency(Instruction * QueryInst)677 MemDepResult MemoryDependenceResults::getDependency(Instruction *QueryInst) {
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 = !(MR & MRI_Mod);
712 if (IntrinsicInst *II = dyn_cast<IntrinsicInst>(QueryInst))
713 isLoad |= II->getIntrinsicID() == Intrinsic::lifetime_start;
714
715 LocalCache = getPointerDependencyFrom(
716 MemLoc, isLoad, ScanPos->getIterator(), QueryParent, QueryInst);
717 } else if (isa<CallInst>(QueryInst) || isa<InvokeInst>(QueryInst)) {
718 CallSite QueryCS(QueryInst);
719 bool isReadOnly = AA.onlyReadsMemory(QueryCS);
720 LocalCache = getCallSiteDependencyFrom(
721 QueryCS, isReadOnly, 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.
AssertSorted(MemoryDependenceResults::NonLocalDepInfo & Cache,int Count=-1)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 &
getNonLocalCallDependency(CallSite QueryCS)747 MemoryDependenceResults::getNonLocalCallDependency(CallSite QueryCS) {
748 assert(getDependency(QueryCS.getInstruction()).isNonLocal() &&
749 "getNonLocalCallDependency should only be used on calls with "
750 "non-local deps!");
751 PerInstNLInfo &CacheP = NonLocalDeps[QueryCS.getInstruction()];
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 std::sort(Cache.begin(), Cache.end());
775
776 ++NumCacheDirtyNonLocal;
777 // cerr << "CACHED CASE: " << DirtyBlocks.size() << " dirty: "
778 // << Cache.size() << " cached: " << *QueryInst;
779 } else {
780 // Seed DirtyBlocks with each of the preds of QueryInst's block.
781 BasicBlock *QueryBB = QueryCS.getInstruction()->getParent();
782 for (BasicBlock *Pred : PredCache.get(QueryBB))
783 DirtyBlocks.push_back(Pred);
784 ++NumUncacheNonLocal;
785 }
786
787 // isReadonlyCall - If this is a read-only call, we can be more aggressive.
788 bool isReadonlyCall = AA.onlyReadsMemory(QueryCS);
789
790 SmallPtrSet<BasicBlock *, 32> Visited;
791
792 unsigned NumSortedEntries = Cache.size();
793 DEBUG(AssertSorted(Cache));
794
795 // Iterate while we still have blocks to update.
796 while (!DirtyBlocks.empty()) {
797 BasicBlock *DirtyBB = DirtyBlocks.back();
798 DirtyBlocks.pop_back();
799
800 // Already processed this block?
801 if (!Visited.insert(DirtyBB).second)
802 continue;
803
804 // Do a binary search to see if we already have an entry for this block in
805 // the cache set. If so, find it.
806 DEBUG(AssertSorted(Cache, NumSortedEntries));
807 NonLocalDepInfo::iterator Entry =
808 std::upper_bound(Cache.begin(), Cache.begin() + NumSortedEntries,
809 NonLocalDepEntry(DirtyBB));
810 if (Entry != Cache.begin() && std::prev(Entry)->getBB() == DirtyBB)
811 --Entry;
812
813 NonLocalDepEntry *ExistingResult = nullptr;
814 if (Entry != Cache.begin() + NumSortedEntries &&
815 Entry->getBB() == DirtyBB) {
816 // If we already have an entry, and if it isn't already dirty, the block
817 // is done.
818 if (!Entry->getResult().isDirty())
819 continue;
820
821 // Otherwise, remember this slot so we can update the value.
822 ExistingResult = &*Entry;
823 }
824
825 // If the dirty entry has a pointer, start scanning from it so we don't have
826 // to rescan the entire block.
827 BasicBlock::iterator ScanPos = DirtyBB->end();
828 if (ExistingResult) {
829 if (Instruction *Inst = ExistingResult->getResult().getInst()) {
830 ScanPos = Inst->getIterator();
831 // We're removing QueryInst's use of Inst.
832 RemoveFromReverseMap(ReverseNonLocalDeps, Inst,
833 QueryCS.getInstruction());
834 }
835 }
836
837 // Find out if this block has a local dependency for QueryInst.
838 MemDepResult Dep;
839
840 if (ScanPos != DirtyBB->begin()) {
841 Dep =
842 getCallSiteDependencyFrom(QueryCS, isReadonlyCall, ScanPos, DirtyBB);
843 } else if (DirtyBB != &DirtyBB->getParent()->getEntryBlock()) {
844 // No dependence found. If this is the entry block of the function, it is
845 // a clobber, otherwise it is unknown.
846 Dep = MemDepResult::getNonLocal();
847 } else {
848 Dep = MemDepResult::getNonFuncLocal();
849 }
850
851 // If we had a dirty entry for the block, update it. Otherwise, just add
852 // a new entry.
853 if (ExistingResult)
854 ExistingResult->setResult(Dep);
855 else
856 Cache.push_back(NonLocalDepEntry(DirtyBB, Dep));
857
858 // If the block has a dependency (i.e. it isn't completely transparent to
859 // the value), remember the association!
860 if (!Dep.isNonLocal()) {
861 // Keep the ReverseNonLocalDeps map up to date so we can efficiently
862 // update this when we remove instructions.
863 if (Instruction *Inst = Dep.getInst())
864 ReverseNonLocalDeps[Inst].insert(QueryCS.getInstruction());
865 } else {
866
867 // If the block *is* completely transparent to the load, we need to check
868 // the predecessors of this block. Add them to our worklist.
869 for (BasicBlock *Pred : PredCache.get(DirtyBB))
870 DirtyBlocks.push_back(Pred);
871 }
872 }
873
874 return Cache;
875 }
876
getNonLocalPointerDependency(Instruction * QueryInst,SmallVectorImpl<NonLocalDepResult> & Result)877 void MemoryDependenceResults::getNonLocalPointerDependency(
878 Instruction *QueryInst, SmallVectorImpl<NonLocalDepResult> &Result) {
879 const MemoryLocation Loc = MemoryLocation::get(QueryInst);
880 bool isLoad = isa<LoadInst>(QueryInst);
881 BasicBlock *FromBB = QueryInst->getParent();
882 assert(FromBB);
883
884 assert(Loc.Ptr->getType()->isPointerTy() &&
885 "Can't get pointer deps of a non-pointer!");
886 Result.clear();
887
888 // This routine does not expect to deal with volatile instructions.
889 // Doing so would require piping through the QueryInst all the way through.
890 // TODO: volatiles can't be elided, but they can be reordered with other
891 // non-volatile accesses.
892
893 // We currently give up on any instruction which is ordered, but we do handle
894 // atomic instructions which are unordered.
895 // TODO: Handle ordered instructions
896 auto isOrdered = [](Instruction *Inst) {
897 if (LoadInst *LI = dyn_cast<LoadInst>(Inst)) {
898 return !LI->isUnordered();
899 } else if (StoreInst *SI = dyn_cast<StoreInst>(Inst)) {
900 return !SI->isUnordered();
901 }
902 return false;
903 };
904 if (isVolatile(QueryInst) || isOrdered(QueryInst)) {
905 Result.push_back(NonLocalDepResult(FromBB, MemDepResult::getUnknown(),
906 const_cast<Value *>(Loc.Ptr)));
907 return;
908 }
909 const DataLayout &DL = FromBB->getModule()->getDataLayout();
910 PHITransAddr Address(const_cast<Value *>(Loc.Ptr), DL, &AC);
911
912 // This is the set of blocks we've inspected, and the pointer we consider in
913 // each block. Because of critical edges, we currently bail out if querying
914 // a block with multiple different pointers. This can happen during PHI
915 // translation.
916 DenseMap<BasicBlock *, Value *> Visited;
917 if (getNonLocalPointerDepFromBB(QueryInst, Address, Loc, isLoad, FromBB,
918 Result, Visited, true))
919 return;
920 Result.clear();
921 Result.push_back(NonLocalDepResult(FromBB, MemDepResult::getUnknown(),
922 const_cast<Value *>(Loc.Ptr)));
923 }
924
925 /// Compute the memdep value for BB with Pointer/PointeeSize using either
926 /// cached information in Cache or by doing a lookup (which may use dirty cache
927 /// info if available).
928 ///
929 /// If we do a lookup, add the result to the cache.
GetNonLocalInfoForBlock(Instruction * QueryInst,const MemoryLocation & Loc,bool isLoad,BasicBlock * BB,NonLocalDepInfo * Cache,unsigned NumSortedEntries)930 MemDepResult MemoryDependenceResults::GetNonLocalInfoForBlock(
931 Instruction *QueryInst, const MemoryLocation &Loc, bool isLoad,
932 BasicBlock *BB, NonLocalDepInfo *Cache, unsigned NumSortedEntries) {
933
934 // Do a binary search to see if we already have an entry for this block in
935 // the cache set. If so, find it.
936 NonLocalDepInfo::iterator Entry = std::upper_bound(
937 Cache->begin(), Cache->begin() + NumSortedEntries, NonLocalDepEntry(BB));
938 if (Entry != Cache->begin() && (Entry - 1)->getBB() == BB)
939 --Entry;
940
941 NonLocalDepEntry *ExistingResult = nullptr;
942 if (Entry != Cache->begin() + NumSortedEntries && Entry->getBB() == BB)
943 ExistingResult = &*Entry;
944
945 // If we have a cached entry, and it is non-dirty, use it as the value for
946 // this dependency.
947 if (ExistingResult && !ExistingResult->getResult().isDirty()) {
948 ++NumCacheNonLocalPtr;
949 return ExistingResult->getResult();
950 }
951
952 // Otherwise, we have to scan for the value. If we have a dirty cache
953 // entry, start scanning from its position, otherwise we scan from the end
954 // of the block.
955 BasicBlock::iterator ScanPos = BB->end();
956 if (ExistingResult && ExistingResult->getResult().getInst()) {
957 assert(ExistingResult->getResult().getInst()->getParent() == BB &&
958 "Instruction invalidated?");
959 ++NumCacheDirtyNonLocalPtr;
960 ScanPos = ExistingResult->getResult().getInst()->getIterator();
961
962 // Eliminating the dirty entry from 'Cache', so update the reverse info.
963 ValueIsLoadPair CacheKey(Loc.Ptr, isLoad);
964 RemoveFromReverseMap(ReverseNonLocalPtrDeps, &*ScanPos, CacheKey);
965 } else {
966 ++NumUncacheNonLocalPtr;
967 }
968
969 // Scan the block for the dependency.
970 MemDepResult Dep =
971 getPointerDependencyFrom(Loc, isLoad, ScanPos, BB, QueryInst);
972
973 // If we had a dirty entry for the block, update it. Otherwise, just add
974 // a new entry.
975 if (ExistingResult)
976 ExistingResult->setResult(Dep);
977 else
978 Cache->push_back(NonLocalDepEntry(BB, Dep));
979
980 // If the block has a dependency (i.e. it isn't completely transparent to
981 // the value), remember the reverse association because we just added it
982 // to Cache!
983 if (!Dep.isDef() && !Dep.isClobber())
984 return Dep;
985
986 // Keep the ReverseNonLocalPtrDeps map up to date so we can efficiently
987 // update MemDep when we remove instructions.
988 Instruction *Inst = Dep.getInst();
989 assert(Inst && "Didn't depend on anything?");
990 ValueIsLoadPair CacheKey(Loc.Ptr, isLoad);
991 ReverseNonLocalPtrDeps[Inst].insert(CacheKey);
992 return Dep;
993 }
994
995 /// Sort the NonLocalDepInfo cache, given a certain number of elements in the
996 /// array that are already properly ordered.
997 ///
998 /// This is optimized for the case when only a few entries are added.
999 static void
SortNonLocalDepInfoCache(MemoryDependenceResults::NonLocalDepInfo & Cache,unsigned NumSortedEntries)1000 SortNonLocalDepInfoCache(MemoryDependenceResults::NonLocalDepInfo &Cache,
1001 unsigned NumSortedEntries) {
1002 switch (Cache.size() - NumSortedEntries) {
1003 case 0:
1004 // done, no new entries.
1005 break;
1006 case 2: {
1007 // Two new entries, insert the last one into place.
1008 NonLocalDepEntry Val = Cache.back();
1009 Cache.pop_back();
1010 MemoryDependenceResults::NonLocalDepInfo::iterator Entry =
1011 std::upper_bound(Cache.begin(), Cache.end() - 1, Val);
1012 Cache.insert(Entry, Val);
1013 // FALL THROUGH.
1014 }
1015 case 1:
1016 // One new entry, Just insert the new value at the appropriate position.
1017 if (Cache.size() != 1) {
1018 NonLocalDepEntry Val = Cache.back();
1019 Cache.pop_back();
1020 MemoryDependenceResults::NonLocalDepInfo::iterator Entry =
1021 std::upper_bound(Cache.begin(), Cache.end(), Val);
1022 Cache.insert(Entry, Val);
1023 }
1024 break;
1025 default:
1026 // Added many values, do a full scale sort.
1027 std::sort(Cache.begin(), Cache.end());
1028 break;
1029 }
1030 }
1031
1032 /// Perform a dependency query based on pointer/pointeesize starting at the end
1033 /// of StartBB.
1034 ///
1035 /// Add any clobber/def results to the results vector and keep track of which
1036 /// blocks are visited in 'Visited'.
1037 ///
1038 /// This has special behavior for the first block queries (when SkipFirstBlock
1039 /// is true). In this special case, it ignores the contents of the specified
1040 /// block and starts returning dependence info for its predecessors.
1041 ///
1042 /// This function returns true on success, or false to indicate that it could
1043 /// not compute dependence information for some reason. This should be treated
1044 /// as a clobber dependence on the first instruction in the predecessor block.
getNonLocalPointerDepFromBB(Instruction * QueryInst,const PHITransAddr & Pointer,const MemoryLocation & Loc,bool isLoad,BasicBlock * StartBB,SmallVectorImpl<NonLocalDepResult> & Result,DenseMap<BasicBlock *,Value * > & Visited,bool SkipFirstBlock)1045 bool MemoryDependenceResults::getNonLocalPointerDepFromBB(
1046 Instruction *QueryInst, const PHITransAddr &Pointer,
1047 const MemoryLocation &Loc, bool isLoad, BasicBlock *StartBB,
1048 SmallVectorImpl<NonLocalDepResult> &Result,
1049 DenseMap<BasicBlock *, Value *> &Visited, bool SkipFirstBlock) {
1050 // Look up the cached info for Pointer.
1051 ValueIsLoadPair CacheKey(Pointer.getAddr(), isLoad);
1052
1053 // Set up a temporary NLPI value. If the map doesn't yet have an entry for
1054 // CacheKey, this value will be inserted as the associated value. Otherwise,
1055 // it'll be ignored, and we'll have to check to see if the cached size and
1056 // aa tags are consistent with the current query.
1057 NonLocalPointerInfo InitialNLPI;
1058 InitialNLPI.Size = Loc.Size;
1059 InitialNLPI.AATags = Loc.AATags;
1060
1061 // Get the NLPI for CacheKey, inserting one into the map if it doesn't
1062 // already have one.
1063 std::pair<CachedNonLocalPointerInfo::iterator, bool> Pair =
1064 NonLocalPointerDeps.insert(std::make_pair(CacheKey, InitialNLPI));
1065 NonLocalPointerInfo *CacheInfo = &Pair.first->second;
1066
1067 // If we already have a cache entry for this CacheKey, we may need to do some
1068 // work to reconcile the cache entry and the current query.
1069 if (!Pair.second) {
1070 if (CacheInfo->Size < Loc.Size) {
1071 // The query's Size is greater than the cached one. Throw out the
1072 // cached data and proceed with the query at the greater size.
1073 CacheInfo->Pair = BBSkipFirstBlockPair();
1074 CacheInfo->Size = Loc.Size;
1075 for (auto &Entry : CacheInfo->NonLocalDeps)
1076 if (Instruction *Inst = Entry.getResult().getInst())
1077 RemoveFromReverseMap(ReverseNonLocalPtrDeps, Inst, CacheKey);
1078 CacheInfo->NonLocalDeps.clear();
1079 } else if (CacheInfo->Size > Loc.Size) {
1080 // This query's Size is less than the cached one. Conservatively restart
1081 // the query using the greater size.
1082 return getNonLocalPointerDepFromBB(
1083 QueryInst, Pointer, Loc.getWithNewSize(CacheInfo->Size), isLoad,
1084 StartBB, Result, Visited, SkipFirstBlock);
1085 }
1086
1087 // If the query's AATags are inconsistent with the cached one,
1088 // conservatively throw out the cached data and restart the query with
1089 // no tag if needed.
1090 if (CacheInfo->AATags != Loc.AATags) {
1091 if (CacheInfo->AATags) {
1092 CacheInfo->Pair = BBSkipFirstBlockPair();
1093 CacheInfo->AATags = AAMDNodes();
1094 for (auto &Entry : CacheInfo->NonLocalDeps)
1095 if (Instruction *Inst = Entry.getResult().getInst())
1096 RemoveFromReverseMap(ReverseNonLocalPtrDeps, Inst, CacheKey);
1097 CacheInfo->NonLocalDeps.clear();
1098 }
1099 if (Loc.AATags)
1100 return getNonLocalPointerDepFromBB(
1101 QueryInst, Pointer, Loc.getWithoutAATags(), isLoad, StartBB, Result,
1102 Visited, SkipFirstBlock);
1103 }
1104 }
1105
1106 NonLocalDepInfo *Cache = &CacheInfo->NonLocalDeps;
1107
1108 // If we have valid cached information for exactly the block we are
1109 // investigating, just return it with no recomputation.
1110 if (CacheInfo->Pair == BBSkipFirstBlockPair(StartBB, SkipFirstBlock)) {
1111 // We have a fully cached result for this query then we can just return the
1112 // cached results and populate the visited set. However, we have to verify
1113 // that we don't already have conflicting results for these blocks. Check
1114 // to ensure that if a block in the results set is in the visited set that
1115 // it was for the same pointer query.
1116 if (!Visited.empty()) {
1117 for (auto &Entry : *Cache) {
1118 DenseMap<BasicBlock *, Value *>::iterator VI =
1119 Visited.find(Entry.getBB());
1120 if (VI == Visited.end() || VI->second == Pointer.getAddr())
1121 continue;
1122
1123 // We have a pointer mismatch in a block. Just return false, saying
1124 // that something was clobbered in this result. We could also do a
1125 // non-fully cached query, but there is little point in doing this.
1126 return false;
1127 }
1128 }
1129
1130 Value *Addr = Pointer.getAddr();
1131 for (auto &Entry : *Cache) {
1132 Visited.insert(std::make_pair(Entry.getBB(), Addr));
1133 if (Entry.getResult().isNonLocal()) {
1134 continue;
1135 }
1136
1137 if (DT.isReachableFromEntry(Entry.getBB())) {
1138 Result.push_back(
1139 NonLocalDepResult(Entry.getBB(), Entry.getResult(), Addr));
1140 }
1141 }
1142 ++NumCacheCompleteNonLocalPtr;
1143 return true;
1144 }
1145
1146 // Otherwise, either this is a new block, a block with an invalid cache
1147 // pointer or one that we're about to invalidate by putting more info into it
1148 // than its valid cache info. If empty, the result will be valid cache info,
1149 // otherwise it isn't.
1150 if (Cache->empty())
1151 CacheInfo->Pair = BBSkipFirstBlockPair(StartBB, SkipFirstBlock);
1152 else
1153 CacheInfo->Pair = BBSkipFirstBlockPair();
1154
1155 SmallVector<BasicBlock *, 32> Worklist;
1156 Worklist.push_back(StartBB);
1157
1158 // PredList used inside loop.
1159 SmallVector<std::pair<BasicBlock *, PHITransAddr>, 16> PredList;
1160
1161 // Keep track of the entries that we know are sorted. Previously cached
1162 // entries will all be sorted. The entries we add we only sort on demand (we
1163 // don't insert every element into its sorted position). We know that we
1164 // won't get any reuse from currently inserted values, because we don't
1165 // revisit blocks after we insert info for them.
1166 unsigned NumSortedEntries = Cache->size();
1167 unsigned WorklistEntries = BlockNumberLimit;
1168 bool GotWorklistLimit = false;
1169 DEBUG(AssertSorted(*Cache));
1170
1171 while (!Worklist.empty()) {
1172 BasicBlock *BB = Worklist.pop_back_val();
1173
1174 // If we do process a large number of blocks it becomes very expensive and
1175 // likely it isn't worth worrying about
1176 if (Result.size() > NumResultsLimit) {
1177 Worklist.clear();
1178 // Sort it now (if needed) so that recursive invocations of
1179 // getNonLocalPointerDepFromBB and other routines that could reuse the
1180 // cache value will only see properly sorted cache arrays.
1181 if (Cache && NumSortedEntries != Cache->size()) {
1182 SortNonLocalDepInfoCache(*Cache, NumSortedEntries);
1183 }
1184 // Since we bail out, the "Cache" set won't contain all of the
1185 // results for the query. This is ok (we can still use it to accelerate
1186 // specific block queries) but we can't do the fastpath "return all
1187 // results from the set". Clear out the indicator for this.
1188 CacheInfo->Pair = BBSkipFirstBlockPair();
1189 return false;
1190 }
1191
1192 // Skip the first block if we have it.
1193 if (!SkipFirstBlock) {
1194 // Analyze the dependency of *Pointer in FromBB. See if we already have
1195 // been here.
1196 assert(Visited.count(BB) && "Should check 'visited' before adding to WL");
1197
1198 // Get the dependency info for Pointer in BB. If we have cached
1199 // information, we will use it, otherwise we compute it.
1200 DEBUG(AssertSorted(*Cache, NumSortedEntries));
1201 MemDepResult Dep = GetNonLocalInfoForBlock(QueryInst, Loc, isLoad, BB,
1202 Cache, NumSortedEntries);
1203
1204 // If we got a Def or Clobber, add this to the list of results.
1205 if (!Dep.isNonLocal()) {
1206 if (DT.isReachableFromEntry(BB)) {
1207 Result.push_back(NonLocalDepResult(BB, Dep, Pointer.getAddr()));
1208 continue;
1209 }
1210 }
1211 }
1212
1213 // If 'Pointer' is an instruction defined in this block, then we need to do
1214 // phi translation to change it into a value live in the predecessor block.
1215 // If not, we just add the predecessors to the worklist and scan them with
1216 // the same Pointer.
1217 if (!Pointer.NeedsPHITranslationFromBlock(BB)) {
1218 SkipFirstBlock = false;
1219 SmallVector<BasicBlock *, 16> NewBlocks;
1220 for (BasicBlock *Pred : PredCache.get(BB)) {
1221 // Verify that we haven't looked at this block yet.
1222 std::pair<DenseMap<BasicBlock *, Value *>::iterator, bool> InsertRes =
1223 Visited.insert(std::make_pair(Pred, Pointer.getAddr()));
1224 if (InsertRes.second) {
1225 // First time we've looked at *PI.
1226 NewBlocks.push_back(Pred);
1227 continue;
1228 }
1229
1230 // If we have seen this block before, but it was with a different
1231 // pointer then we have a phi translation failure and we have to treat
1232 // this as a clobber.
1233 if (InsertRes.first->second != Pointer.getAddr()) {
1234 // Make sure to clean up the Visited map before continuing on to
1235 // PredTranslationFailure.
1236 for (unsigned i = 0; i < NewBlocks.size(); i++)
1237 Visited.erase(NewBlocks[i]);
1238 goto PredTranslationFailure;
1239 }
1240 }
1241 if (NewBlocks.size() > WorklistEntries) {
1242 // Make sure to clean up the Visited map before continuing on to
1243 // PredTranslationFailure.
1244 for (unsigned i = 0; i < NewBlocks.size(); i++)
1245 Visited.erase(NewBlocks[i]);
1246 GotWorklistLimit = true;
1247 goto PredTranslationFailure;
1248 }
1249 WorklistEntries -= NewBlocks.size();
1250 Worklist.append(NewBlocks.begin(), NewBlocks.end());
1251 continue;
1252 }
1253
1254 // We do need to do phi translation, if we know ahead of time we can't phi
1255 // translate this value, don't even try.
1256 if (!Pointer.IsPotentiallyPHITranslatable())
1257 goto PredTranslationFailure;
1258
1259 // We may have added values to the cache list before this PHI translation.
1260 // If so, we haven't done anything to ensure that the cache remains sorted.
1261 // Sort it now (if needed) so that recursive invocations of
1262 // getNonLocalPointerDepFromBB and other routines that could reuse the cache
1263 // value will only see properly sorted cache arrays.
1264 if (Cache && NumSortedEntries != Cache->size()) {
1265 SortNonLocalDepInfoCache(*Cache, NumSortedEntries);
1266 NumSortedEntries = Cache->size();
1267 }
1268 Cache = nullptr;
1269
1270 PredList.clear();
1271 for (BasicBlock *Pred : PredCache.get(BB)) {
1272 PredList.push_back(std::make_pair(Pred, Pointer));
1273
1274 // Get the PHI translated pointer in this predecessor. This can fail if
1275 // not translatable, in which case the getAddr() returns null.
1276 PHITransAddr &PredPointer = PredList.back().second;
1277 PredPointer.PHITranslateValue(BB, Pred, &DT, /*MustDominate=*/false);
1278 Value *PredPtrVal = PredPointer.getAddr();
1279
1280 // Check to see if we have already visited this pred block with another
1281 // pointer. If so, we can't do this lookup. This failure can occur
1282 // with PHI translation when a critical edge exists and the PHI node in
1283 // the successor translates to a pointer value different than the
1284 // pointer the block was first analyzed with.
1285 std::pair<DenseMap<BasicBlock *, Value *>::iterator, bool> InsertRes =
1286 Visited.insert(std::make_pair(Pred, PredPtrVal));
1287
1288 if (!InsertRes.second) {
1289 // We found the pred; take it off the list of preds to visit.
1290 PredList.pop_back();
1291
1292 // If the predecessor was visited with PredPtr, then we already did
1293 // the analysis and can ignore it.
1294 if (InsertRes.first->second == PredPtrVal)
1295 continue;
1296
1297 // Otherwise, the block was previously analyzed with a different
1298 // pointer. We can't represent the result of this case, so we just
1299 // treat this as a phi translation failure.
1300
1301 // Make sure to clean up the Visited map before continuing on to
1302 // PredTranslationFailure.
1303 for (unsigned i = 0, n = PredList.size(); i < n; ++i)
1304 Visited.erase(PredList[i].first);
1305
1306 goto PredTranslationFailure;
1307 }
1308 }
1309
1310 // Actually process results here; this need to be a separate loop to avoid
1311 // calling getNonLocalPointerDepFromBB for blocks we don't want to return
1312 // any results for. (getNonLocalPointerDepFromBB will modify our
1313 // datastructures in ways the code after the PredTranslationFailure label
1314 // doesn't expect.)
1315 for (unsigned i = 0, n = PredList.size(); i < n; ++i) {
1316 BasicBlock *Pred = PredList[i].first;
1317 PHITransAddr &PredPointer = PredList[i].second;
1318 Value *PredPtrVal = PredPointer.getAddr();
1319
1320 bool CanTranslate = true;
1321 // If PHI translation was unable to find an available pointer in this
1322 // predecessor, then we have to assume that the pointer is clobbered in
1323 // that predecessor. We can still do PRE of the load, which would insert
1324 // a computation of the pointer in this predecessor.
1325 if (!PredPtrVal)
1326 CanTranslate = false;
1327
1328 // FIXME: it is entirely possible that PHI translating will end up with
1329 // the same value. Consider PHI translating something like:
1330 // X = phi [x, bb1], [y, bb2]. PHI translating for bb1 doesn't *need*
1331 // to recurse here, pedantically speaking.
1332
1333 // If getNonLocalPointerDepFromBB fails here, that means the cached
1334 // result conflicted with the Visited list; we have to conservatively
1335 // assume it is unknown, but this also does not block PRE of the load.
1336 if (!CanTranslate ||
1337 !getNonLocalPointerDepFromBB(QueryInst, PredPointer,
1338 Loc.getWithNewPtr(PredPtrVal), isLoad,
1339 Pred, Result, Visited)) {
1340 // Add the entry to the Result list.
1341 NonLocalDepResult Entry(Pred, MemDepResult::getUnknown(), PredPtrVal);
1342 Result.push_back(Entry);
1343
1344 // Since we had a phi translation failure, the cache for CacheKey won't
1345 // include all of the entries that we need to immediately satisfy future
1346 // queries. Mark this in NonLocalPointerDeps by setting the
1347 // BBSkipFirstBlockPair pointer to null. This requires reuse of the
1348 // cached value to do more work but not miss the phi trans failure.
1349 NonLocalPointerInfo &NLPI = NonLocalPointerDeps[CacheKey];
1350 NLPI.Pair = BBSkipFirstBlockPair();
1351 continue;
1352 }
1353 }
1354
1355 // Refresh the CacheInfo/Cache pointer so that it isn't invalidated.
1356 CacheInfo = &NonLocalPointerDeps[CacheKey];
1357 Cache = &CacheInfo->NonLocalDeps;
1358 NumSortedEntries = Cache->size();
1359
1360 // Since we did phi translation, the "Cache" set won't contain all of the
1361 // results for the query. This is ok (we can still use it to accelerate
1362 // specific block queries) but we can't do the fastpath "return all
1363 // results from the set" Clear out the indicator for this.
1364 CacheInfo->Pair = BBSkipFirstBlockPair();
1365 SkipFirstBlock = false;
1366 continue;
1367
1368 PredTranslationFailure:
1369 // The following code is "failure"; we can't produce a sane translation
1370 // for the given block. It assumes that we haven't modified any of
1371 // our datastructures while processing the current block.
1372
1373 if (!Cache) {
1374 // Refresh the CacheInfo/Cache pointer if it got invalidated.
1375 CacheInfo = &NonLocalPointerDeps[CacheKey];
1376 Cache = &CacheInfo->NonLocalDeps;
1377 NumSortedEntries = Cache->size();
1378 }
1379
1380 // Since we failed phi translation, the "Cache" set won't contain all of the
1381 // results for the query. This is ok (we can still use it to accelerate
1382 // specific block queries) but we can't do the fastpath "return all
1383 // results from the set". Clear out the indicator for this.
1384 CacheInfo->Pair = BBSkipFirstBlockPair();
1385
1386 // If *nothing* works, mark the pointer as unknown.
1387 //
1388 // If this is the magic first block, return this as a clobber of the whole
1389 // incoming value. Since we can't phi translate to one of the predecessors,
1390 // we have to bail out.
1391 if (SkipFirstBlock)
1392 return false;
1393
1394 bool foundBlock = false;
1395 for (NonLocalDepEntry &I : llvm::reverse(*Cache)) {
1396 if (I.getBB() != BB)
1397 continue;
1398
1399 assert((GotWorklistLimit || I.getResult().isNonLocal() ||
1400 !DT.isReachableFromEntry(BB)) &&
1401 "Should only be here with transparent block");
1402 foundBlock = true;
1403 I.setResult(MemDepResult::getUnknown());
1404 Result.push_back(
1405 NonLocalDepResult(I.getBB(), I.getResult(), Pointer.getAddr()));
1406 break;
1407 }
1408 (void)foundBlock; (void)GotWorklistLimit;
1409 assert((foundBlock || GotWorklistLimit) && "Current block not in cache?");
1410 }
1411
1412 // Okay, we're done now. If we added new values to the cache, re-sort it.
1413 SortNonLocalDepInfoCache(*Cache, NumSortedEntries);
1414 DEBUG(AssertSorted(*Cache));
1415 return true;
1416 }
1417
1418 /// If P exists in CachedNonLocalPointerInfo, remove it.
RemoveCachedNonLocalPointerDependencies(ValueIsLoadPair P)1419 void MemoryDependenceResults::RemoveCachedNonLocalPointerDependencies(
1420 ValueIsLoadPair P) {
1421 CachedNonLocalPointerInfo::iterator It = NonLocalPointerDeps.find(P);
1422 if (It == NonLocalPointerDeps.end())
1423 return;
1424
1425 // Remove all of the entries in the BB->val map. This involves removing
1426 // instructions from the reverse map.
1427 NonLocalDepInfo &PInfo = It->second.NonLocalDeps;
1428
1429 for (unsigned i = 0, e = PInfo.size(); i != e; ++i) {
1430 Instruction *Target = PInfo[i].getResult().getInst();
1431 if (!Target)
1432 continue; // Ignore non-local dep results.
1433 assert(Target->getParent() == PInfo[i].getBB());
1434
1435 // Eliminating the dirty entry from 'Cache', so update the reverse info.
1436 RemoveFromReverseMap(ReverseNonLocalPtrDeps, Target, P);
1437 }
1438
1439 // Remove P from NonLocalPointerDeps (which deletes NonLocalDepInfo).
1440 NonLocalPointerDeps.erase(It);
1441 }
1442
invalidateCachedPointerInfo(Value * Ptr)1443 void MemoryDependenceResults::invalidateCachedPointerInfo(Value *Ptr) {
1444 // If Ptr isn't really a pointer, just ignore it.
1445 if (!Ptr->getType()->isPointerTy())
1446 return;
1447 // Flush store info for the pointer.
1448 RemoveCachedNonLocalPointerDependencies(ValueIsLoadPair(Ptr, false));
1449 // Flush load info for the pointer.
1450 RemoveCachedNonLocalPointerDependencies(ValueIsLoadPair(Ptr, true));
1451 }
1452
invalidateCachedPredecessors()1453 void MemoryDependenceResults::invalidateCachedPredecessors() {
1454 PredCache.clear();
1455 }
1456
removeInstruction(Instruction * RemInst)1457 void MemoryDependenceResults::removeInstruction(Instruction *RemInst) {
1458 // Walk through the Non-local dependencies, removing this one as the value
1459 // for any cached queries.
1460 NonLocalDepMapType::iterator NLDI = NonLocalDeps.find(RemInst);
1461 if (NLDI != NonLocalDeps.end()) {
1462 NonLocalDepInfo &BlockMap = NLDI->second.first;
1463 for (auto &Entry : BlockMap)
1464 if (Instruction *Inst = Entry.getResult().getInst())
1465 RemoveFromReverseMap(ReverseNonLocalDeps, Inst, RemInst);
1466 NonLocalDeps.erase(NLDI);
1467 }
1468
1469 // If we have a cached local dependence query for this instruction, remove it.
1470 //
1471 LocalDepMapType::iterator LocalDepEntry = LocalDeps.find(RemInst);
1472 if (LocalDepEntry != LocalDeps.end()) {
1473 // Remove us from DepInst's reverse set now that the local dep info is gone.
1474 if (Instruction *Inst = LocalDepEntry->second.getInst())
1475 RemoveFromReverseMap(ReverseLocalDeps, Inst, RemInst);
1476
1477 // Remove this local dependency info.
1478 LocalDeps.erase(LocalDepEntry);
1479 }
1480
1481 // If we have any cached pointer dependencies on this instruction, remove
1482 // them. If the instruction has non-pointer type, then it can't be a pointer
1483 // base.
1484
1485 // Remove it from both the load info and the store info. The instruction
1486 // can't be in either of these maps if it is non-pointer.
1487 if (RemInst->getType()->isPointerTy()) {
1488 RemoveCachedNonLocalPointerDependencies(ValueIsLoadPair(RemInst, false));
1489 RemoveCachedNonLocalPointerDependencies(ValueIsLoadPair(RemInst, true));
1490 }
1491
1492 // Loop over all of the things that depend on the instruction we're removing.
1493 //
1494 SmallVector<std::pair<Instruction *, Instruction *>, 8> ReverseDepsToAdd;
1495
1496 // If we find RemInst as a clobber or Def in any of the maps for other values,
1497 // we need to replace its entry with a dirty version of the instruction after
1498 // it. If RemInst is a terminator, we use a null dirty value.
1499 //
1500 // Using a dirty version of the instruction after RemInst saves having to scan
1501 // the entire block to get to this point.
1502 MemDepResult NewDirtyVal;
1503 if (!RemInst->isTerminator())
1504 NewDirtyVal = MemDepResult::getDirty(&*++RemInst->getIterator());
1505
1506 ReverseDepMapType::iterator ReverseDepIt = ReverseLocalDeps.find(RemInst);
1507 if (ReverseDepIt != ReverseLocalDeps.end()) {
1508 // RemInst can't be the terminator if it has local stuff depending on it.
1509 assert(!ReverseDepIt->second.empty() && !isa<TerminatorInst>(RemInst) &&
1510 "Nothing can locally depend on a terminator");
1511
1512 for (Instruction *InstDependingOnRemInst : ReverseDepIt->second) {
1513 assert(InstDependingOnRemInst != RemInst &&
1514 "Already removed our local dep info");
1515
1516 LocalDeps[InstDependingOnRemInst] = NewDirtyVal;
1517
1518 // Make sure to remember that new things depend on NewDepInst.
1519 assert(NewDirtyVal.getInst() &&
1520 "There is no way something else can have "
1521 "a local dep on this if it is a terminator!");
1522 ReverseDepsToAdd.push_back(
1523 std::make_pair(NewDirtyVal.getInst(), InstDependingOnRemInst));
1524 }
1525
1526 ReverseLocalDeps.erase(ReverseDepIt);
1527
1528 // Add new reverse deps after scanning the set, to avoid invalidating the
1529 // 'ReverseDeps' reference.
1530 while (!ReverseDepsToAdd.empty()) {
1531 ReverseLocalDeps[ReverseDepsToAdd.back().first].insert(
1532 ReverseDepsToAdd.back().second);
1533 ReverseDepsToAdd.pop_back();
1534 }
1535 }
1536
1537 ReverseDepIt = ReverseNonLocalDeps.find(RemInst);
1538 if (ReverseDepIt != ReverseNonLocalDeps.end()) {
1539 for (Instruction *I : ReverseDepIt->second) {
1540 assert(I != RemInst && "Already removed NonLocalDep info for RemInst");
1541
1542 PerInstNLInfo &INLD = NonLocalDeps[I];
1543 // The information is now dirty!
1544 INLD.second = true;
1545
1546 for (auto &Entry : INLD.first) {
1547 if (Entry.getResult().getInst() != RemInst)
1548 continue;
1549
1550 // Convert to a dirty entry for the subsequent instruction.
1551 Entry.setResult(NewDirtyVal);
1552
1553 if (Instruction *NextI = NewDirtyVal.getInst())
1554 ReverseDepsToAdd.push_back(std::make_pair(NextI, I));
1555 }
1556 }
1557
1558 ReverseNonLocalDeps.erase(ReverseDepIt);
1559
1560 // Add new reverse deps after scanning the set, to avoid invalidating 'Set'
1561 while (!ReverseDepsToAdd.empty()) {
1562 ReverseNonLocalDeps[ReverseDepsToAdd.back().first].insert(
1563 ReverseDepsToAdd.back().second);
1564 ReverseDepsToAdd.pop_back();
1565 }
1566 }
1567
1568 // If the instruction is in ReverseNonLocalPtrDeps then it appears as a
1569 // value in the NonLocalPointerDeps info.
1570 ReverseNonLocalPtrDepTy::iterator ReversePtrDepIt =
1571 ReverseNonLocalPtrDeps.find(RemInst);
1572 if (ReversePtrDepIt != ReverseNonLocalPtrDeps.end()) {
1573 SmallVector<std::pair<Instruction *, ValueIsLoadPair>, 8>
1574 ReversePtrDepsToAdd;
1575
1576 for (ValueIsLoadPair P : ReversePtrDepIt->second) {
1577 assert(P.getPointer() != RemInst &&
1578 "Already removed NonLocalPointerDeps info for RemInst");
1579
1580 NonLocalDepInfo &NLPDI = NonLocalPointerDeps[P].NonLocalDeps;
1581
1582 // The cache is not valid for any specific block anymore.
1583 NonLocalPointerDeps[P].Pair = BBSkipFirstBlockPair();
1584
1585 // Update any entries for RemInst to use the instruction after it.
1586 for (auto &Entry : NLPDI) {
1587 if (Entry.getResult().getInst() != RemInst)
1588 continue;
1589
1590 // Convert to a dirty entry for the subsequent instruction.
1591 Entry.setResult(NewDirtyVal);
1592
1593 if (Instruction *NewDirtyInst = NewDirtyVal.getInst())
1594 ReversePtrDepsToAdd.push_back(std::make_pair(NewDirtyInst, P));
1595 }
1596
1597 // Re-sort the NonLocalDepInfo. Changing the dirty entry to its
1598 // subsequent value may invalidate the sortedness.
1599 std::sort(NLPDI.begin(), NLPDI.end());
1600 }
1601
1602 ReverseNonLocalPtrDeps.erase(ReversePtrDepIt);
1603
1604 while (!ReversePtrDepsToAdd.empty()) {
1605 ReverseNonLocalPtrDeps[ReversePtrDepsToAdd.back().first].insert(
1606 ReversePtrDepsToAdd.back().second);
1607 ReversePtrDepsToAdd.pop_back();
1608 }
1609 }
1610
1611 assert(!NonLocalDeps.count(RemInst) && "RemInst got reinserted?");
1612 DEBUG(verifyRemoved(RemInst));
1613 }
1614
1615 /// Verify that the specified instruction does not occur in our internal data
1616 /// structures.
1617 ///
1618 /// This function verifies by asserting in debug builds.
verifyRemoved(Instruction * D) const1619 void MemoryDependenceResults::verifyRemoved(Instruction *D) const {
1620 #ifndef NDEBUG
1621 for (const auto &DepKV : LocalDeps) {
1622 assert(DepKV.first != D && "Inst occurs in data structures");
1623 assert(DepKV.second.getInst() != D && "Inst occurs in data structures");
1624 }
1625
1626 for (const auto &DepKV : NonLocalPointerDeps) {
1627 assert(DepKV.first.getPointer() != D && "Inst occurs in NLPD map key");
1628 for (const auto &Entry : DepKV.second.NonLocalDeps)
1629 assert(Entry.getResult().getInst() != D && "Inst occurs as NLPD value");
1630 }
1631
1632 for (const auto &DepKV : NonLocalDeps) {
1633 assert(DepKV.first != D && "Inst occurs in data structures");
1634 const PerInstNLInfo &INLD = DepKV.second;
1635 for (const auto &Entry : INLD.first)
1636 assert(Entry.getResult().getInst() != D &&
1637 "Inst occurs in data structures");
1638 }
1639
1640 for (const auto &DepKV : ReverseLocalDeps) {
1641 assert(DepKV.first != D && "Inst occurs in data structures");
1642 for (Instruction *Inst : DepKV.second)
1643 assert(Inst != D && "Inst occurs in data structures");
1644 }
1645
1646 for (const auto &DepKV : ReverseNonLocalDeps) {
1647 assert(DepKV.first != D && "Inst occurs in data structures");
1648 for (Instruction *Inst : DepKV.second)
1649 assert(Inst != D && "Inst occurs in data structures");
1650 }
1651
1652 for (const auto &DepKV : ReverseNonLocalPtrDeps) {
1653 assert(DepKV.first != D && "Inst occurs in rev NLPD map");
1654
1655 for (ValueIsLoadPair P : DepKV.second)
1656 assert(P != ValueIsLoadPair(D, false) && P != ValueIsLoadPair(D, true) &&
1657 "Inst occurs in ReverseNonLocalPtrDeps map");
1658 }
1659 #endif
1660 }
1661
1662 char MemoryDependenceAnalysis::PassID;
1663
1664 MemoryDependenceResults
run(Function & F,AnalysisManager<Function> & AM)1665 MemoryDependenceAnalysis::run(Function &F, AnalysisManager<Function> &AM) {
1666 auto &AA = AM.getResult<AAManager>(F);
1667 auto &AC = AM.getResult<AssumptionAnalysis>(F);
1668 auto &TLI = AM.getResult<TargetLibraryAnalysis>(F);
1669 auto &DT = AM.getResult<DominatorTreeAnalysis>(F);
1670 return MemoryDependenceResults(AA, AC, TLI, DT);
1671 }
1672
1673 char MemoryDependenceWrapperPass::ID = 0;
1674
1675 INITIALIZE_PASS_BEGIN(MemoryDependenceWrapperPass, "memdep",
1676 "Memory Dependence Analysis", false, true)
INITIALIZE_PASS_DEPENDENCY(AssumptionCacheTracker)1677 INITIALIZE_PASS_DEPENDENCY(AssumptionCacheTracker)
1678 INITIALIZE_PASS_DEPENDENCY(AAResultsWrapperPass)
1679 INITIALIZE_PASS_DEPENDENCY(DominatorTreeWrapperPass)
1680 INITIALIZE_PASS_DEPENDENCY(TargetLibraryInfoWrapperPass)
1681 INITIALIZE_PASS_END(MemoryDependenceWrapperPass, "memdep",
1682 "Memory Dependence Analysis", false, true)
1683
1684 MemoryDependenceWrapperPass::MemoryDependenceWrapperPass() : FunctionPass(ID) {
1685 initializeMemoryDependenceWrapperPassPass(*PassRegistry::getPassRegistry());
1686 }
~MemoryDependenceWrapperPass()1687 MemoryDependenceWrapperPass::~MemoryDependenceWrapperPass() {}
1688
releaseMemory()1689 void MemoryDependenceWrapperPass::releaseMemory() {
1690 MemDep.reset();
1691 }
1692
getAnalysisUsage(AnalysisUsage & AU) const1693 void MemoryDependenceWrapperPass::getAnalysisUsage(AnalysisUsage &AU) const {
1694 AU.setPreservesAll();
1695 AU.addRequired<AssumptionCacheTracker>();
1696 AU.addRequired<DominatorTreeWrapperPass>();
1697 AU.addRequiredTransitive<AAResultsWrapperPass>();
1698 AU.addRequiredTransitive<TargetLibraryInfoWrapperPass>();
1699 }
1700
runOnFunction(Function & F)1701 bool MemoryDependenceWrapperPass::runOnFunction(Function &F) {
1702 auto &AA = getAnalysis<AAResultsWrapperPass>().getAAResults();
1703 auto &AC = getAnalysis<AssumptionCacheTracker>().getAssumptionCache(F);
1704 auto &TLI = getAnalysis<TargetLibraryInfoWrapperPass>().getTLI();
1705 auto &DT = getAnalysis<DominatorTreeWrapperPass>().getDomTree();
1706 MemDep.emplace(AA, AC, TLI, DT);
1707 return false;
1708 }
1709