1//===- Loads.cpp - Local load analysis ------------------------------------===//
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 defines simple local analyses for load instructions.
10//
11//===----------------------------------------------------------------------===//
12
13#include "llvm/Analysis/Loads.h"
14#include "llvm/Analysis/AliasAnalysis.h"
15#include "llvm/Analysis/AssumeBundleQueries.h"
16#include "llvm/Analysis/LoopInfo.h"
17#include "llvm/Analysis/MemoryBuiltins.h"
18#include "llvm/Analysis/MemoryLocation.h"
19#include "llvm/Analysis/ScalarEvolution.h"
20#include "llvm/Analysis/ScalarEvolutionExpressions.h"
21#include "llvm/Analysis/ValueTracking.h"
22#include "llvm/IR/DataLayout.h"
23#include "llvm/IR/IntrinsicInst.h"
24#include "llvm/IR/Module.h"
25#include "llvm/IR/Operator.h"
26
27using namespace llvm;
28
29static bool isAligned(const Value *Base, const APInt &Offset, Align Alignment,
30 const DataLayout &DL) {
31 Align BA = Base->getPointerAlignment(DL);
32 return BA >= Alignment && Offset.isAligned(A: BA);
33}
34
35/// Test if V is always a pointer to allocated and suitably aligned memory for
36/// a simple load or store.
37static bool isDereferenceableAndAlignedPointer(
38 const Value *V, Align Alignment, const APInt &Size, const DataLayout &DL,
39 const Instruction *CtxI, AssumptionCache *AC, const DominatorTree *DT,
40 const TargetLibraryInfo *TLI, SmallPtrSetImpl<const Value *> &Visited,
41 unsigned MaxDepth) {
42 assert(V->getType()->isPointerTy() && "Base must be pointer");
43
44 // Recursion limit.
45 if (MaxDepth-- == 0)
46 return false;
47
48 // Already visited? Bail out, we've likely hit unreachable code.
49 if (!Visited.insert(Ptr: V).second)
50 return false;
51
52 // Note that it is not safe to speculate into a malloc'd region because
53 // malloc may return null.
54
55 // For GEPs, determine if the indexing lands within the allocated object.
56 if (const GEPOperator *GEP = dyn_cast<GEPOperator>(Val: V)) {
57 const Value *Base = GEP->getPointerOperand();
58
59 APInt Offset(DL.getIndexTypeSizeInBits(Ty: GEP->getType()), 0);
60 if (!GEP->accumulateConstantOffset(DL, Offset) || Offset.isNegative() ||
61 !Offset.urem(RHS: APInt(Offset.getBitWidth(), Alignment.value()))
62 .isMinValue())
63 return false;
64
65 // If the base pointer is dereferenceable for Offset+Size bytes, then the
66 // GEP (== Base + Offset) is dereferenceable for Size bytes. If the base
67 // pointer is aligned to Align bytes, and the Offset is divisible by Align
68 // then the GEP (== Base + Offset == k_0 * Align + k_1 * Align) is also
69 // aligned to Align bytes.
70
71 // Offset and Size may have different bit widths if we have visited an
72 // addrspacecast, so we can't do arithmetic directly on the APInt values.
73 return isDereferenceableAndAlignedPointer(
74 V: Base, Alignment, Size: Offset + Size.sextOrTrunc(width: Offset.getBitWidth()), DL,
75 CtxI, AC, DT, TLI, Visited, MaxDepth);
76 }
77
78 // bitcast instructions are no-ops as far as dereferenceability is concerned.
79 if (const BitCastOperator *BC = dyn_cast<BitCastOperator>(Val: V)) {
80 if (BC->getSrcTy()->isPointerTy())
81 return isDereferenceableAndAlignedPointer(
82 V: BC->getOperand(i_nocapture: 0), Alignment, Size, DL, CtxI, AC, DT, TLI,
83 Visited, MaxDepth);
84 }
85
86 // Recurse into both hands of select.
87 if (const SelectInst *Sel = dyn_cast<SelectInst>(Val: V)) {
88 return isDereferenceableAndAlignedPointer(V: Sel->getTrueValue(), Alignment,
89 Size, DL, CtxI, AC, DT, TLI,
90 Visited, MaxDepth) &&
91 isDereferenceableAndAlignedPointer(V: Sel->getFalseValue(), Alignment,
92 Size, DL, CtxI, AC, DT, TLI,
93 Visited, MaxDepth);
94 }
95
96 bool CheckForNonNull, CheckForFreed;
97 APInt KnownDerefBytes(Size.getBitWidth(),
98 V->getPointerDereferenceableBytes(DL, CanBeNull&: CheckForNonNull,
99 CanBeFreed&: CheckForFreed));
100 if (KnownDerefBytes.getBoolValue() && KnownDerefBytes.uge(RHS: Size) &&
101 !CheckForFreed)
102 if (!CheckForNonNull ||
103 isKnownNonZero(V, Q: SimplifyQuery(DL, DT, AC, CtxI))) {
104 // As we recursed through GEPs to get here, we've incrementally checked
105 // that each step advanced by a multiple of the alignment. If our base is
106 // properly aligned, then the original offset accessed must also be.
107 APInt Offset(DL.getTypeStoreSizeInBits(Ty: V->getType()), 0);
108 return isAligned(Base: V, Offset, Alignment, DL);
109 }
110
111 /// TODO refactor this function to be able to search independently for
112 /// Dereferencability and Alignment requirements.
113
114
115 if (const auto *Call = dyn_cast<CallBase>(Val: V)) {
116 if (auto *RP = getArgumentAliasingToReturnedPointer(Call, MustPreserveNullness: true))
117 return isDereferenceableAndAlignedPointer(V: RP, Alignment, Size, DL, CtxI,
118 AC, DT, TLI, Visited, MaxDepth);
119
120 // If we have a call we can't recurse through, check to see if this is an
121 // allocation function for which we can establish an minimum object size.
122 // Such a minimum object size is analogous to a deref_or_null attribute in
123 // that we still need to prove the result non-null at point of use.
124 // NOTE: We can only use the object size as a base fact as we a) need to
125 // prove alignment too, and b) don't want the compile time impact of a
126 // separate recursive walk.
127 ObjectSizeOpts Opts;
128 // TODO: It may be okay to round to align, but that would imply that
129 // accessing slightly out of bounds was legal, and we're currently
130 // inconsistent about that. For the moment, be conservative.
131 Opts.RoundToAlign = false;
132 Opts.NullIsUnknownSize = true;
133 uint64_t ObjSize;
134 if (getObjectSize(Ptr: V, Size&: ObjSize, DL, TLI, Opts)) {
135 APInt KnownDerefBytes(Size.getBitWidth(), ObjSize);
136 if (KnownDerefBytes.getBoolValue() && KnownDerefBytes.uge(RHS: Size) &&
137 isKnownNonZero(V, Q: SimplifyQuery(DL, DT, AC, CtxI)) &&
138 !V->canBeFreed()) {
139 // As we recursed through GEPs to get here, we've incrementally
140 // checked that each step advanced by a multiple of the alignment. If
141 // our base is properly aligned, then the original offset accessed
142 // must also be.
143 APInt Offset(DL.getTypeStoreSizeInBits(Ty: V->getType()), 0);
144 return isAligned(Base: V, Offset, Alignment, DL);
145 }
146 }
147 }
148
149 // For gc.relocate, look through relocations
150 if (const GCRelocateInst *RelocateInst = dyn_cast<GCRelocateInst>(Val: V))
151 return isDereferenceableAndAlignedPointer(V: RelocateInst->getDerivedPtr(),
152 Alignment, Size, DL, CtxI, AC, DT,
153 TLI, Visited, MaxDepth);
154
155 if (const AddrSpaceCastOperator *ASC = dyn_cast<AddrSpaceCastOperator>(Val: V))
156 return isDereferenceableAndAlignedPointer(V: ASC->getOperand(i_nocapture: 0), Alignment,
157 Size, DL, CtxI, AC, DT, TLI,
158 Visited, MaxDepth);
159
160 if (CtxI) {
161 /// Look through assumes to see if both dereferencability and alignment can
162 /// be provent by an assume
163 RetainedKnowledge AlignRK;
164 RetainedKnowledge DerefRK;
165 if (getKnowledgeForValue(
166 V, AttrKinds: {Attribute::Dereferenceable, Attribute::Alignment}, AC,
167 Filter: [&](RetainedKnowledge RK, Instruction *Assume, auto) {
168 if (!isValidAssumeForContext(I: Assume, CxtI: CtxI, DT))
169 return false;
170 if (RK.AttrKind == Attribute::Alignment)
171 AlignRK = std::max(a: AlignRK, b: RK);
172 if (RK.AttrKind == Attribute::Dereferenceable)
173 DerefRK = std::max(a: DerefRK, b: RK);
174 if (AlignRK && DerefRK && AlignRK.ArgValue >= Alignment.value() &&
175 DerefRK.ArgValue >= Size.getZExtValue())
176 return true; // We have found what we needed so we stop looking
177 return false; // Other assumes may have better information. so
178 // keep looking
179 }))
180 return true;
181 }
182
183 // If we don't know, assume the worst.
184 return false;
185}
186
187bool llvm::isDereferenceableAndAlignedPointer(
188 const Value *V, Align Alignment, const APInt &Size, const DataLayout &DL,
189 const Instruction *CtxI, AssumptionCache *AC, const DominatorTree *DT,
190 const TargetLibraryInfo *TLI) {
191 // Note: At the moment, Size can be zero. This ends up being interpreted as
192 // a query of whether [Base, V] is dereferenceable and V is aligned (since
193 // that's what the implementation happened to do). It's unclear if this is
194 // the desired semantic, but at least SelectionDAG does exercise this case.
195
196 SmallPtrSet<const Value *, 32> Visited;
197 return ::isDereferenceableAndAlignedPointer(V, Alignment, Size, DL, CtxI, AC,
198 DT, TLI, Visited, MaxDepth: 16);
199}
200
201bool llvm::isDereferenceableAndAlignedPointer(
202 const Value *V, Type *Ty, Align Alignment, const DataLayout &DL,
203 const Instruction *CtxI, AssumptionCache *AC, const DominatorTree *DT,
204 const TargetLibraryInfo *TLI) {
205 // For unsized types or scalable vectors we don't know exactly how many bytes
206 // are dereferenced, so bail out.
207 if (!Ty->isSized() || Ty->isScalableTy())
208 return false;
209
210 // When dereferenceability information is provided by a dereferenceable
211 // attribute, we know exactly how many bytes are dereferenceable. If we can
212 // determine the exact offset to the attributed variable, we can use that
213 // information here.
214
215 APInt AccessSize(DL.getPointerTypeSizeInBits(V->getType()),
216 DL.getTypeStoreSize(Ty));
217 return isDereferenceableAndAlignedPointer(V, Alignment, Size: AccessSize, DL, CtxI,
218 AC, DT, TLI);
219}
220
221bool llvm::isDereferenceablePointer(const Value *V, Type *Ty,
222 const DataLayout &DL,
223 const Instruction *CtxI,
224 AssumptionCache *AC,
225 const DominatorTree *DT,
226 const TargetLibraryInfo *TLI) {
227 return isDereferenceableAndAlignedPointer(V, Ty, Alignment: Align(1), DL, CtxI, AC, DT,
228 TLI);
229}
230
231/// Test if A and B will obviously have the same value.
232///
233/// This includes recognizing that %t0 and %t1 will have the same
234/// value in code like this:
235/// \code
236/// %t0 = getelementptr \@a, 0, 3
237/// store i32 0, i32* %t0
238/// %t1 = getelementptr \@a, 0, 3
239/// %t2 = load i32* %t1
240/// \endcode
241///
242static bool AreEquivalentAddressValues(const Value *A, const Value *B) {
243 // Test if the values are trivially equivalent.
244 if (A == B)
245 return true;
246
247 // Test if the values come from identical arithmetic instructions.
248 // Use isIdenticalToWhenDefined instead of isIdenticalTo because
249 // this function is only used when one address use dominates the
250 // other, which means that they'll always either have the same
251 // value or one of them will have an undefined value.
252 if (isa<BinaryOperator>(Val: A) || isa<CastInst>(Val: A) || isa<PHINode>(Val: A) ||
253 isa<GetElementPtrInst>(Val: A))
254 if (const Instruction *BI = dyn_cast<Instruction>(Val: B))
255 if (cast<Instruction>(Val: A)->isIdenticalToWhenDefined(I: BI))
256 return true;
257
258 // Otherwise they may not be equivalent.
259 return false;
260}
261
262bool llvm::isDereferenceableAndAlignedInLoop(LoadInst *LI, Loop *L,
263 ScalarEvolution &SE,
264 DominatorTree &DT,
265 AssumptionCache *AC) {
266 auto &DL = LI->getDataLayout();
267 Value *Ptr = LI->getPointerOperand();
268
269 APInt EltSize(DL.getIndexTypeSizeInBits(Ty: Ptr->getType()),
270 DL.getTypeStoreSize(Ty: LI->getType()).getFixedValue());
271 const Align Alignment = LI->getAlign();
272
273 Instruction *HeaderFirstNonPHI = L->getHeader()->getFirstNonPHI();
274
275 // If given a uniform (i.e. non-varying) address, see if we can prove the
276 // access is safe within the loop w/o needing predication.
277 if (L->isLoopInvariant(V: Ptr))
278 return isDereferenceableAndAlignedPointer(V: Ptr, Alignment, Size: EltSize, DL,
279 CtxI: HeaderFirstNonPHI, AC, DT: &DT);
280
281 // Otherwise, check to see if we have a repeating access pattern where we can
282 // prove that all accesses are well aligned and dereferenceable.
283 auto *AddRec = dyn_cast<SCEVAddRecExpr>(Val: SE.getSCEV(V: Ptr));
284 if (!AddRec || AddRec->getLoop() != L || !AddRec->isAffine())
285 return false;
286 auto* Step = dyn_cast<SCEVConstant>(Val: AddRec->getStepRecurrence(SE));
287 if (!Step)
288 return false;
289
290 auto TC = SE.getSmallConstantMaxTripCount(L);
291 if (!TC)
292 return false;
293
294 // TODO: Handle overlapping accesses.
295 // We should be computing AccessSize as (TC - 1) * Step + EltSize.
296 if (EltSize.sgt(RHS: Step->getAPInt()))
297 return false;
298
299 // Compute the total access size for access patterns with unit stride and
300 // patterns with gaps. For patterns with unit stride, Step and EltSize are the
301 // same.
302 // For patterns with gaps (i.e. non unit stride), we are
303 // accessing EltSize bytes at every Step.
304 APInt AccessSize = TC * Step->getAPInt();
305
306 assert(SE.isLoopInvariant(AddRec->getStart(), L) &&
307 "implied by addrec definition");
308 Value *Base = nullptr;
309 if (auto *StartS = dyn_cast<SCEVUnknown>(Val: AddRec->getStart())) {
310 Base = StartS->getValue();
311 } else if (auto *StartS = dyn_cast<SCEVAddExpr>(Val: AddRec->getStart())) {
312 // Handle (NewBase + offset) as start value.
313 const auto *Offset = dyn_cast<SCEVConstant>(Val: StartS->getOperand(i: 0));
314 const auto *NewBase = dyn_cast<SCEVUnknown>(Val: StartS->getOperand(i: 1));
315 if (StartS->getNumOperands() == 2 && Offset && NewBase) {
316 // The following code below assumes the offset is unsigned, but GEP
317 // offsets are treated as signed so we can end up with a signed value
318 // here too. For example, suppose the initial PHI value is (i8 255),
319 // the offset will be treated as (i8 -1) and sign-extended to (i64 -1).
320 if (Offset->getAPInt().isNegative())
321 return false;
322
323 // For the moment, restrict ourselves to the case where the offset is a
324 // multiple of the requested alignment and the base is aligned.
325 // TODO: generalize if a case found which warrants
326 if (Offset->getAPInt().urem(RHS: Alignment.value()) != 0)
327 return false;
328 Base = NewBase->getValue();
329 bool Overflow = false;
330 AccessSize = AccessSize.uadd_ov(RHS: Offset->getAPInt(), Overflow);
331 if (Overflow)
332 return false;
333 }
334 }
335
336 if (!Base)
337 return false;
338
339 // For the moment, restrict ourselves to the case where the access size is a
340 // multiple of the requested alignment and the base is aligned.
341 // TODO: generalize if a case found which warrants
342 if (EltSize.urem(RHS: Alignment.value()) != 0)
343 return false;
344 return isDereferenceableAndAlignedPointer(V: Base, Alignment, Size: AccessSize, DL,
345 CtxI: HeaderFirstNonPHI, AC, DT: &DT);
346}
347
348/// Check if executing a load of this pointer value cannot trap.
349///
350/// If DT and ScanFrom are specified this method performs context-sensitive
351/// analysis and returns true if it is safe to load immediately before ScanFrom.
352///
353/// If it is not obviously safe to load from the specified pointer, we do
354/// a quick local scan of the basic block containing \c ScanFrom, to determine
355/// if the address is already accessed.
356///
357/// This uses the pointee type to determine how many bytes need to be safe to
358/// load from the pointer.
359bool llvm::isSafeToLoadUnconditionally(Value *V, Align Alignment, const APInt &Size,
360 const DataLayout &DL,
361 Instruction *ScanFrom,
362 AssumptionCache *AC,
363 const DominatorTree *DT,
364 const TargetLibraryInfo *TLI) {
365 // If DT is not specified we can't make context-sensitive query
366 const Instruction* CtxI = DT ? ScanFrom : nullptr;
367 if (isDereferenceableAndAlignedPointer(V, Alignment, Size, DL, CtxI, AC, DT,
368 TLI))
369 return true;
370
371 if (!ScanFrom)
372 return false;
373
374 if (Size.getBitWidth() > 64)
375 return false;
376 const TypeSize LoadSize = TypeSize::getFixed(ExactSize: Size.getZExtValue());
377
378 // Otherwise, be a little bit aggressive by scanning the local block where we
379 // want to check to see if the pointer is already being loaded or stored
380 // from/to. If so, the previous load or store would have already trapped,
381 // so there is no harm doing an extra load (also, CSE will later eliminate
382 // the load entirely).
383 BasicBlock::iterator BBI = ScanFrom->getIterator(),
384 E = ScanFrom->getParent()->begin();
385
386 // We can at least always strip pointer casts even though we can't use the
387 // base here.
388 V = V->stripPointerCasts();
389
390 while (BBI != E) {
391 --BBI;
392
393 // If we see a free or a call which may write to memory (i.e. which might do
394 // a free) the pointer could be marked invalid.
395 if (isa<CallInst>(Val: BBI) && BBI->mayWriteToMemory() &&
396 !isa<LifetimeIntrinsic>(Val: BBI) && !isa<DbgInfoIntrinsic>(Val: BBI))
397 return false;
398
399 Value *AccessedPtr;
400 Type *AccessedTy;
401 Align AccessedAlign;
402 if (LoadInst *LI = dyn_cast<LoadInst>(Val&: BBI)) {
403 // Ignore volatile loads. The execution of a volatile load cannot
404 // be used to prove an address is backed by regular memory; it can,
405 // for example, point to an MMIO register.
406 if (LI->isVolatile())
407 continue;
408 AccessedPtr = LI->getPointerOperand();
409 AccessedTy = LI->getType();
410 AccessedAlign = LI->getAlign();
411 } else if (StoreInst *SI = dyn_cast<StoreInst>(Val&: BBI)) {
412 // Ignore volatile stores (see comment for loads).
413 if (SI->isVolatile())
414 continue;
415 AccessedPtr = SI->getPointerOperand();
416 AccessedTy = SI->getValueOperand()->getType();
417 AccessedAlign = SI->getAlign();
418 } else
419 continue;
420
421 if (AccessedAlign < Alignment)
422 continue;
423
424 // Handle trivial cases.
425 if (AccessedPtr == V &&
426 TypeSize::isKnownLE(LHS: LoadSize, RHS: DL.getTypeStoreSize(Ty: AccessedTy)))
427 return true;
428
429 if (AreEquivalentAddressValues(A: AccessedPtr->stripPointerCasts(), B: V) &&
430 TypeSize::isKnownLE(LHS: LoadSize, RHS: DL.getTypeStoreSize(Ty: AccessedTy)))
431 return true;
432 }
433 return false;
434}
435
436bool llvm::isSafeToLoadUnconditionally(Value *V, Type *Ty, Align Alignment,
437 const DataLayout &DL,
438 Instruction *ScanFrom,
439 AssumptionCache *AC,
440 const DominatorTree *DT,
441 const TargetLibraryInfo *TLI) {
442 TypeSize TySize = DL.getTypeStoreSize(Ty);
443 if (TySize.isScalable())
444 return false;
445 APInt Size(DL.getIndexTypeSizeInBits(Ty: V->getType()), TySize.getFixedValue());
446 return isSafeToLoadUnconditionally(V, Alignment, Size, DL, ScanFrom, AC, DT,
447 TLI);
448}
449
450/// DefMaxInstsToScan - the default number of maximum instructions
451/// to scan in the block, used by FindAvailableLoadedValue().
452/// FindAvailableLoadedValue() was introduced in r60148, to improve jump
453/// threading in part by eliminating partially redundant loads.
454/// At that point, the value of MaxInstsToScan was already set to '6'
455/// without documented explanation.
456cl::opt<unsigned>
457llvm::DefMaxInstsToScan("available-load-scan-limit", cl::init(Val: 6), cl::Hidden,
458 cl::desc("Use this to specify the default maximum number of instructions "
459 "to scan backward from a given instruction, when searching for "
460 "available loaded value"));
461
462Value *llvm::FindAvailableLoadedValue(LoadInst *Load, BasicBlock *ScanBB,
463 BasicBlock::iterator &ScanFrom,
464 unsigned MaxInstsToScan,
465 BatchAAResults *AA, bool *IsLoad,
466 unsigned *NumScanedInst) {
467 // Don't CSE load that is volatile or anything stronger than unordered.
468 if (!Load->isUnordered())
469 return nullptr;
470
471 MemoryLocation Loc = MemoryLocation::get(LI: Load);
472 return findAvailablePtrLoadStore(Loc, AccessTy: Load->getType(), AtLeastAtomic: Load->isAtomic(),
473 ScanBB, ScanFrom, MaxInstsToScan, AA, IsLoadCSE: IsLoad,
474 NumScanedInst);
475}
476
477// Check if the load and the store have the same base, constant offsets and
478// non-overlapping access ranges.
479static bool areNonOverlapSameBaseLoadAndStore(const Value *LoadPtr,
480 Type *LoadTy,
481 const Value *StorePtr,
482 Type *StoreTy,
483 const DataLayout &DL) {
484 APInt LoadOffset(DL.getIndexTypeSizeInBits(Ty: LoadPtr->getType()), 0);
485 APInt StoreOffset(DL.getIndexTypeSizeInBits(Ty: StorePtr->getType()), 0);
486 const Value *LoadBase = LoadPtr->stripAndAccumulateConstantOffsets(
487 DL, Offset&: LoadOffset, /* AllowNonInbounds */ false);
488 const Value *StoreBase = StorePtr->stripAndAccumulateConstantOffsets(
489 DL, Offset&: StoreOffset, /* AllowNonInbounds */ false);
490 if (LoadBase != StoreBase)
491 return false;
492 auto LoadAccessSize = LocationSize::precise(Value: DL.getTypeStoreSize(Ty: LoadTy));
493 auto StoreAccessSize = LocationSize::precise(Value: DL.getTypeStoreSize(Ty: StoreTy));
494 ConstantRange LoadRange(LoadOffset,
495 LoadOffset + LoadAccessSize.toRaw());
496 ConstantRange StoreRange(StoreOffset,
497 StoreOffset + StoreAccessSize.toRaw());
498 return LoadRange.intersectWith(CR: StoreRange).isEmptySet();
499}
500
501static Value *getAvailableLoadStore(Instruction *Inst, const Value *Ptr,
502 Type *AccessTy, bool AtLeastAtomic,
503 const DataLayout &DL, bool *IsLoadCSE) {
504 // If this is a load of Ptr, the loaded value is available.
505 // (This is true even if the load is volatile or atomic, although
506 // those cases are unlikely.)
507 if (LoadInst *LI = dyn_cast<LoadInst>(Val: Inst)) {
508 // We can value forward from an atomic to a non-atomic, but not the
509 // other way around.
510 if (LI->isAtomic() < AtLeastAtomic)
511 return nullptr;
512
513 Value *LoadPtr = LI->getPointerOperand()->stripPointerCasts();
514 if (!AreEquivalentAddressValues(A: LoadPtr, B: Ptr))
515 return nullptr;
516
517 if (CastInst::isBitOrNoopPointerCastable(SrcTy: LI->getType(), DestTy: AccessTy, DL)) {
518 if (IsLoadCSE)
519 *IsLoadCSE = true;
520 return LI;
521 }
522 }
523
524 // If this is a store through Ptr, the value is available!
525 // (This is true even if the store is volatile or atomic, although
526 // those cases are unlikely.)
527 if (StoreInst *SI = dyn_cast<StoreInst>(Val: Inst)) {
528 // We can value forward from an atomic to a non-atomic, but not the
529 // other way around.
530 if (SI->isAtomic() < AtLeastAtomic)
531 return nullptr;
532
533 Value *StorePtr = SI->getPointerOperand()->stripPointerCasts();
534 if (!AreEquivalentAddressValues(A: StorePtr, B: Ptr))
535 return nullptr;
536
537 if (IsLoadCSE)
538 *IsLoadCSE = false;
539
540 Value *Val = SI->getValueOperand();
541 if (CastInst::isBitOrNoopPointerCastable(SrcTy: Val->getType(), DestTy: AccessTy, DL))
542 return Val;
543
544 TypeSize StoreSize = DL.getTypeSizeInBits(Ty: Val->getType());
545 TypeSize LoadSize = DL.getTypeSizeInBits(Ty: AccessTy);
546 if (TypeSize::isKnownLE(LHS: LoadSize, RHS: StoreSize))
547 if (auto *C = dyn_cast<Constant>(Val))
548 return ConstantFoldLoadFromConst(C, Ty: AccessTy, DL);
549 }
550
551 if (auto *MSI = dyn_cast<MemSetInst>(Val: Inst)) {
552 // Don't forward from (non-atomic) memset to atomic load.
553 if (AtLeastAtomic)
554 return nullptr;
555
556 // Only handle constant memsets.
557 auto *Val = dyn_cast<ConstantInt>(Val: MSI->getValue());
558 auto *Len = dyn_cast<ConstantInt>(Val: MSI->getLength());
559 if (!Val || !Len)
560 return nullptr;
561
562 // TODO: Handle offsets.
563 Value *Dst = MSI->getDest();
564 if (!AreEquivalentAddressValues(A: Dst, B: Ptr))
565 return nullptr;
566
567 if (IsLoadCSE)
568 *IsLoadCSE = false;
569
570 TypeSize LoadTypeSize = DL.getTypeSizeInBits(Ty: AccessTy);
571 if (LoadTypeSize.isScalable())
572 return nullptr;
573
574 // Make sure the read bytes are contained in the memset.
575 uint64_t LoadSize = LoadTypeSize.getFixedValue();
576 if ((Len->getValue() * 8).ult(RHS: LoadSize))
577 return nullptr;
578
579 APInt Splat = LoadSize >= 8 ? APInt::getSplat(NewLen: LoadSize, V: Val->getValue())
580 : Val->getValue().trunc(width: LoadSize);
581 ConstantInt *SplatC = ConstantInt::get(Context&: MSI->getContext(), V: Splat);
582 if (CastInst::isBitOrNoopPointerCastable(SrcTy: SplatC->getType(), DestTy: AccessTy, DL))
583 return SplatC;
584
585 return nullptr;
586 }
587
588 return nullptr;
589}
590
591Value *llvm::findAvailablePtrLoadStore(
592 const MemoryLocation &Loc, Type *AccessTy, bool AtLeastAtomic,
593 BasicBlock *ScanBB, BasicBlock::iterator &ScanFrom, unsigned MaxInstsToScan,
594 BatchAAResults *AA, bool *IsLoadCSE, unsigned *NumScanedInst) {
595 if (MaxInstsToScan == 0)
596 MaxInstsToScan = ~0U;
597
598 const DataLayout &DL = ScanBB->getDataLayout();
599 const Value *StrippedPtr = Loc.Ptr->stripPointerCasts();
600
601 while (ScanFrom != ScanBB->begin()) {
602 // We must ignore debug info directives when counting (otherwise they
603 // would affect codegen).
604 Instruction *Inst = &*--ScanFrom;
605 if (Inst->isDebugOrPseudoInst())
606 continue;
607
608 // Restore ScanFrom to expected value in case next test succeeds
609 ScanFrom++;
610
611 if (NumScanedInst)
612 ++(*NumScanedInst);
613
614 // Don't scan huge blocks.
615 if (MaxInstsToScan-- == 0)
616 return nullptr;
617
618 --ScanFrom;
619
620 if (Value *Available = getAvailableLoadStore(Inst, Ptr: StrippedPtr, AccessTy,
621 AtLeastAtomic, DL, IsLoadCSE))
622 return Available;
623
624 // Try to get the store size for the type.
625 if (StoreInst *SI = dyn_cast<StoreInst>(Val: Inst)) {
626 Value *StorePtr = SI->getPointerOperand()->stripPointerCasts();
627
628 // If both StrippedPtr and StorePtr reach all the way to an alloca or
629 // global and they are different, ignore the store. This is a trivial form
630 // of alias analysis that is important for reg2mem'd code.
631 if ((isa<AllocaInst>(Val: StrippedPtr) || isa<GlobalVariable>(Val: StrippedPtr)) &&
632 (isa<AllocaInst>(Val: StorePtr) || isa<GlobalVariable>(Val: StorePtr)) &&
633 StrippedPtr != StorePtr)
634 continue;
635
636 if (!AA) {
637 // When AA isn't available, but if the load and the store have the same
638 // base, constant offsets and non-overlapping access ranges, ignore the
639 // store. This is a simple form of alias analysis that is used by the
640 // inliner. FIXME: use BasicAA if possible.
641 if (areNonOverlapSameBaseLoadAndStore(
642 LoadPtr: Loc.Ptr, LoadTy: AccessTy, StorePtr: SI->getPointerOperand(),
643 StoreTy: SI->getValueOperand()->getType(), DL))
644 continue;
645 } else {
646 // If we have alias analysis and it says the store won't modify the
647 // loaded value, ignore the store.
648 if (!isModSet(MRI: AA->getModRefInfo(I: SI, OptLoc: Loc)))
649 continue;
650 }
651
652 // Otherwise the store that may or may not alias the pointer, bail out.
653 ++ScanFrom;
654 return nullptr;
655 }
656
657 // If this is some other instruction that may clobber Ptr, bail out.
658 if (Inst->mayWriteToMemory()) {
659 // If alias analysis claims that it really won't modify the load,
660 // ignore it.
661 if (AA && !isModSet(MRI: AA->getModRefInfo(I: Inst, OptLoc: Loc)))
662 continue;
663
664 // May modify the pointer, bail out.
665 ++ScanFrom;
666 return nullptr;
667 }
668 }
669
670 // Got to the start of the block, we didn't find it, but are done for this
671 // block.
672 return nullptr;
673}
674
675Value *llvm::FindAvailableLoadedValue(LoadInst *Load, BatchAAResults &AA,
676 bool *IsLoadCSE,
677 unsigned MaxInstsToScan) {
678 const DataLayout &DL = Load->getDataLayout();
679 Value *StrippedPtr = Load->getPointerOperand()->stripPointerCasts();
680 BasicBlock *ScanBB = Load->getParent();
681 Type *AccessTy = Load->getType();
682 bool AtLeastAtomic = Load->isAtomic();
683
684 if (!Load->isUnordered())
685 return nullptr;
686
687 // Try to find an available value first, and delay expensive alias analysis
688 // queries until later.
689 Value *Available = nullptr;
690 SmallVector<Instruction *> MustNotAliasInsts;
691 for (Instruction &Inst : make_range(x: ++Load->getReverseIterator(),
692 y: ScanBB->rend())) {
693 if (Inst.isDebugOrPseudoInst())
694 continue;
695
696 if (MaxInstsToScan-- == 0)
697 return nullptr;
698
699 Available = getAvailableLoadStore(Inst: &Inst, Ptr: StrippedPtr, AccessTy,
700 AtLeastAtomic, DL, IsLoadCSE);
701 if (Available)
702 break;
703
704 if (Inst.mayWriteToMemory())
705 MustNotAliasInsts.push_back(Elt: &Inst);
706 }
707
708 // If we found an available value, ensure that the instructions in between
709 // did not modify the memory location.
710 if (Available) {
711 MemoryLocation Loc = MemoryLocation::get(LI: Load);
712 for (Instruction *Inst : MustNotAliasInsts)
713 if (isModSet(MRI: AA.getModRefInfo(I: Inst, OptLoc: Loc)))
714 return nullptr;
715 }
716
717 return Available;
718}
719
720// Returns true if a use is either in an ICmp/PtrToInt or a Phi/Select that only
721// feeds into them.
722static bool isPointerUseReplacable(const Use &U) {
723 unsigned Limit = 40;
724 SmallVector<const User *> Worklist({U.getUser()});
725 SmallPtrSet<const User *, 8> Visited;
726
727 while (!Worklist.empty() && --Limit) {
728 auto *User = Worklist.pop_back_val();
729 if (!Visited.insert(Ptr: User).second)
730 continue;
731 if (isa<ICmpInst, PtrToIntInst>(Val: User))
732 continue;
733 if (isa<PHINode, SelectInst>(Val: User))
734 Worklist.append(in_start: User->user_begin(), in_end: User->user_end());
735 else
736 return false;
737 }
738
739 return Limit != 0;
740}
741
742// Returns true if `To` is a null pointer, constant dereferenceable pointer or
743// both pointers have the same underlying objects.
744static bool isPointerAlwaysReplaceable(const Value *From, const Value *To,
745 const DataLayout &DL) {
746 // This is not strictly correct, but we do it for now to retain important
747 // optimizations.
748 if (isa<ConstantPointerNull>(Val: To))
749 return true;
750 if (isa<Constant>(Val: To) &&
751 isDereferenceablePointer(V: To, Ty: Type::getInt8Ty(C&: To->getContext()), DL))
752 return true;
753 return getUnderlyingObjectAggressive(V: From) ==
754 getUnderlyingObjectAggressive(V: To);
755}
756
757bool llvm::canReplacePointersInUseIfEqual(const Use &U, const Value *To,
758 const DataLayout &DL) {
759 assert(U->getType() == To->getType() && "values must have matching types");
760 // Not a pointer, just return true.
761 if (!To->getType()->isPointerTy())
762 return true;
763
764 if (isPointerAlwaysReplaceable(From: &*U, To, DL))
765 return true;
766 return isPointerUseReplacable(U);
767}
768
769bool llvm::canReplacePointersIfEqual(const Value *From, const Value *To,
770 const DataLayout &DL) {
771 assert(From->getType() == To->getType() && "values must have matching types");
772 // Not a pointer, just return true.
773 if (!From->getType()->isPointerTy())
774 return true;
775
776 return isPointerAlwaysReplaceable(From, To, DL);
777}
778
779bool llvm::isDereferenceableReadOnlyLoop(Loop *L, ScalarEvolution *SE,
780 DominatorTree *DT,
781 AssumptionCache *AC) {
782 for (BasicBlock *BB : L->blocks()) {
783 for (Instruction &I : *BB) {
784 if (auto *LI = dyn_cast<LoadInst>(Val: &I)) {
785 if (!isDereferenceableAndAlignedInLoop(LI, L, SE&: *SE, DT&: *DT, AC))
786 return false;
787 } else if (I.mayReadFromMemory() || I.mayWriteToMemory() || I.mayThrow())
788 return false;
789 }
790 }
791 return true;
792}
793