88using namespace PatternMatch;
90#define DEBUG_TYPE "dse"
92STATISTIC(NumRemainingStores,
"Number of stores remaining after DSE");
93STATISTIC(NumRedundantStores,
"Number of redundant stores deleted");
94STATISTIC(NumFastStores,
"Number of stores deleted");
95STATISTIC(NumFastOther,
"Number of other instrs removed");
96STATISTIC(NumCompletePartials,
"Number of stores dead by later partials");
97STATISTIC(NumModifiedStores,
"Number of stores modified");
98STATISTIC(NumCFGChecks,
"Number of stores modified");
102 "Number of times a valid candidate is returned from getDomMemoryDef");
104 "Number iterations check for reads in getDomMemoryDef");
107 "Controls which MemoryDefs are eliminated.");
112 cl::desc(
"Enable partial-overwrite tracking in DSE"));
117 cl::desc(
"Enable partial store merging in DSE"));
121 cl::desc(
"The number of memory instructions to scan for "
122 "dead store elimination (default = 150)"));
125 cl::desc(
"The maximum number of steps while walking upwards to find "
126 "MemoryDefs that may be killed (default = 90)"));
130 cl::desc(
"The maximum number candidates that only partially overwrite the "
131 "killing MemoryDef to consider"
136 cl::desc(
"The number of MemoryDefs we consider as candidates to eliminated "
137 "other stores per basic block (default = 5000)"));
142 "The cost of a step in the same basic block as the killing MemoryDef"
148 cl::desc(
"The cost of a step in a different basic "
149 "block than the killing MemoryDef"
154 cl::desc(
"The maximum number of blocks to check when trying to prove that "
155 "all paths to an exit go through a killing block (default = 50)"));
165 cl::desc(
"Allow DSE to optimize memory accesses."));
177 if (isa<StoreInst>(
I))
181 switch (II->getIntrinsicID()) {
182 default:
return false;
183 case Intrinsic::memset:
184 case Intrinsic::memcpy:
185 case Intrinsic::memcpy_element_unordered_atomic:
186 case Intrinsic::memset_element_unordered_atomic:
203 return isa<AnyMemSetInst>(
I);
221enum OverwriteResult {
225 OW_PartialEarlierWithFullLater,
239 const auto *KillingII = dyn_cast<IntrinsicInst>(KillingI);
240 const auto *DeadII = dyn_cast<IntrinsicInst>(DeadI);
241 if (KillingII ==
nullptr || DeadII ==
nullptr)
243 if (KillingII->getIntrinsicID() != DeadII->getIntrinsicID())
245 if (KillingII->getIntrinsicID() == Intrinsic::masked_store) {
248 cast<VectorType>(KillingII->getArgOperand(0)->getType());
249 VectorType *DeadTy = cast<VectorType>(DeadII->getArgOperand(0)->getType());
250 if (KillingTy->getScalarSizeInBits() != DeadTy->getScalarSizeInBits())
253 if (KillingTy->getElementCount() != DeadTy->getElementCount())
258 if (KillingPtr != DeadPtr && !AA.
isMustAlias(KillingPtr, DeadPtr))
262 if (KillingII->getArgOperand(3) != DeadII->getArgOperand(3))
281 int64_t KillingOff, int64_t DeadOff,
292 KillingOff < int64_t(DeadOff + DeadSize) &&
293 int64_t(KillingOff + KillingSize) >= DeadOff) {
296 auto &IM = IOL[DeadI];
297 LLVM_DEBUG(
dbgs() <<
"DSE: Partial overwrite: DeadLoc [" << DeadOff <<
", "
298 << int64_t(DeadOff + DeadSize) <<
") KillingLoc ["
299 << KillingOff <<
", " << int64_t(KillingOff + KillingSize)
306 int64_t KillingIntStart = KillingOff;
307 int64_t KillingIntEnd = KillingOff + KillingSize;
311 auto ILI = IM.lower_bound(KillingIntStart);
312 if (ILI != IM.end() && ILI->second <= KillingIntEnd) {
316 KillingIntStart = std::min(KillingIntStart, ILI->second);
317 KillingIntEnd = std::max(KillingIntEnd, ILI->first);
326 while (ILI != IM.end() && ILI->second <= KillingIntEnd) {
327 assert(ILI->second > KillingIntStart &&
"Unexpected interval");
328 KillingIntEnd = std::max(KillingIntEnd, ILI->first);
333 IM[KillingIntEnd] = KillingIntStart;
336 if (ILI->second <= DeadOff && ILI->first >= int64_t(DeadOff + DeadSize)) {
337 LLVM_DEBUG(
dbgs() <<
"DSE: Full overwrite from partials: DeadLoc ["
338 << DeadOff <<
", " << int64_t(DeadOff + DeadSize)
339 <<
") Composite KillingLoc [" << ILI->second <<
", "
340 << ILI->first <<
")\n");
341 ++NumCompletePartials;
349 int64_t(DeadOff + DeadSize) > KillingOff &&
350 uint64_t(KillingOff - DeadOff) + KillingSize <= DeadSize) {
351 LLVM_DEBUG(
dbgs() <<
"DSE: Partial overwrite a dead load [" << DeadOff
352 <<
", " << int64_t(DeadOff + DeadSize)
353 <<
") by a killing store [" << KillingOff <<
", "
354 << int64_t(KillingOff + KillingSize) <<
")\n");
356 return OW_PartialEarlierWithFullLater;
369 (KillingOff > DeadOff && KillingOff < int64_t(DeadOff + DeadSize) &&
370 int64_t(KillingOff + KillingSize) >= int64_t(DeadOff + DeadSize)))
383 (KillingOff <= DeadOff && int64_t(KillingOff + KillingSize) > DeadOff)) {
384 assert(int64_t(KillingOff + KillingSize) < int64_t(DeadOff + DeadSize) &&
385 "Expect to be handled as OW_Complete");
405 using BlockAddressPair = std::pair<BasicBlock *, PHITransAddr>;
417 if (
auto *MemSet = dyn_cast<MemSetInst>(SecondI))
422 auto *MemLocPtr =
const_cast<Value *
>(MemLoc.
Ptr);
427 bool isFirstBlock =
true;
430 while (!WorkList.
empty()) {
442 assert(
B == SecondBB &&
"first block is not the store block");
444 isFirstBlock =
false;
450 for (; BI != EI; ++BI) {
452 if (
I->mayWriteToMemory() &&
I != SecondI)
458 "Should not hit the entry block because SI must be dominated by LI");
468 auto Inserted = Visited.
insert(std::make_pair(Pred, TranslatedPtr));
469 if (!Inserted.second) {
472 if (TranslatedPtr != Inserted.first->second)
477 WorkList.
push_back(std::make_pair(Pred, PredAddr));
486 uint64_t NewSizeInBits,
bool IsOverwriteEnd) {
488 uint64_t DeadSliceSizeInBits = OldSizeInBits - NewSizeInBits;
490 OldOffsetInBits + (IsOverwriteEnd ? NewSizeInBits : 0);
491 auto SetDeadFragExpr = [](
auto *Assign,
495 uint64_t RelativeOffset = DeadFragment.OffsetInBits -
496 Assign->getExpression()
501 Assign->getExpression(), RelativeOffset, DeadFragment.SizeInBits)) {
502 Assign->setExpression(*
NewExpr);
508 DIExpression::get(Assign->getContext(), std::nullopt),
509 DeadFragment.OffsetInBits, DeadFragment.SizeInBits);
510 Assign->setExpression(Expr);
511 Assign->setKillLocation();
518 auto GetDeadLink = [&Ctx, &LinkToNothing]() {
521 return LinkToNothing;
533 auto InsertAssignForOverlap = [&](
auto *Assign) {
534 std::optional<DIExpression::FragmentInfo> NewFragment;
536 DeadSliceSizeInBits, Assign,
541 Assign->setKillAddress();
542 Assign->setAssignId(GetDeadLink());
546 if (NewFragment->SizeInBits == 0)
550 auto *NewAssign =
static_cast<decltype(Assign)
>(Assign->clone());
551 NewAssign->insertAfter(Assign);
552 NewAssign->setAssignId(GetDeadLink());
554 SetDeadFragExpr(NewAssign, *NewFragment);
555 NewAssign->setKillAddress();
557 for_each(Linked, InsertAssignForOverlap);
558 for_each(LinkedDVRAssigns, InsertAssignForOverlap);
562 uint64_t &DeadSize, int64_t KillingStart,
563 uint64_t KillingSize,
bool IsOverwriteEnd) {
564 auto *DeadIntrinsic = cast<AnyMemIntrinsic>(DeadI);
565 Align PrefAlign = DeadIntrinsic->getDestAlign().valueOrOne();
581 int64_t ToRemoveStart = 0;
585 if (IsOverwriteEnd) {
590 ToRemoveStart = KillingStart + Off;
591 if (DeadSize <=
uint64_t(ToRemoveStart - DeadStart))
593 ToRemoveSize = DeadSize -
uint64_t(ToRemoveStart - DeadStart);
595 ToRemoveStart = DeadStart;
597 "Not overlapping accesses?");
598 ToRemoveSize = KillingSize -
uint64_t(DeadStart - KillingStart);
603 if (ToRemoveSize <= (PrefAlign.
value() - Off))
605 ToRemoveSize -= PrefAlign.
value() - Off;
608 "Should preserve selected alignment");
611 assert(ToRemoveSize > 0 &&
"Shouldn't reach here if nothing to remove");
612 assert(DeadSize > ToRemoveSize &&
"Can't remove more than original size");
614 uint64_t NewSize = DeadSize - ToRemoveSize;
615 if (
auto *AMI = dyn_cast<AtomicMemIntrinsic>(DeadI)) {
618 const uint32_t ElementSize = AMI->getElementSizeInBytes();
619 if (0 != NewSize % ElementSize)
624 << (IsOverwriteEnd ?
"END" :
"BEGIN") <<
": " << *DeadI
625 <<
"\n KILLER [" << ToRemoveStart <<
", "
626 << int64_t(ToRemoveStart + ToRemoveSize) <<
")\n");
628 Value *DeadWriteLength = DeadIntrinsic->getLength();
629 Value *TrimmedLength = ConstantInt::get(DeadWriteLength->
getType(), NewSize);
630 DeadIntrinsic->setLength(TrimmedLength);
631 DeadIntrinsic->setDestAlignment(PrefAlign);
633 Value *OrigDest = DeadIntrinsic->getRawDest();
634 if (!IsOverwriteEnd) {
635 Value *Indices[1] = {
636 ConstantInt::get(DeadWriteLength->
getType(), ToRemoveSize)};
640 NewDestGEP->
setDebugLoc(DeadIntrinsic->getDebugLoc());
641 DeadIntrinsic->setDest(NewDestGEP);
650 DeadStart += ToRemoveSize;
657 int64_t &DeadStart,
uint64_t &DeadSize) {
662 int64_t KillingStart = OII->second;
663 uint64_t KillingSize = OII->first - KillingStart;
665 assert(OII->first - KillingStart >= 0 &&
"Size expected to be positive");
667 if (KillingStart > DeadStart &&
670 (
uint64_t)(KillingStart - DeadStart) < DeadSize &&
673 KillingSize >= DeadSize - (
uint64_t)(KillingStart - DeadStart)) {
674 if (
tryToShorten(DeadI, DeadStart, DeadSize, KillingStart, KillingSize,
685 int64_t &DeadStart,
uint64_t &DeadSize) {
690 int64_t KillingStart = OII->second;
691 uint64_t KillingSize = OII->first - KillingStart;
693 assert(OII->first - KillingStart >= 0 &&
"Size expected to be positive");
695 if (KillingStart <= DeadStart &&
698 KillingSize > (
uint64_t)(DeadStart - KillingStart)) {
701 assert(KillingSize - (
uint64_t)(DeadStart - KillingStart) < DeadSize &&
702 "Should have been handled as OW_Complete");
703 if (
tryToShorten(DeadI, DeadStart, DeadSize, KillingStart, KillingSize,
714 int64_t KillingOffset, int64_t DeadOffset,
741 unsigned BitOffsetDiff = (KillingOffset - DeadOffset) * 8;
742 unsigned LShiftAmount =
743 DL.isBigEndian() ? DeadValue.
getBitWidth() - BitOffsetDiff - KillingBits
746 LShiftAmount + KillingBits);
749 APInt Merged = (DeadValue & ~Mask) | (KillingValue << LShiftAmount);
751 <<
"\n Killing: " << *KillingI
752 <<
"\n Merged Value: " << Merged <<
'\n');
762 switch (II->getIntrinsicID()) {
763 case Intrinsic::lifetime_start:
764 case Intrinsic::lifetime_end:
765 case Intrinsic::invariant_end:
766 case Intrinsic::launder_invariant_group:
767 case Intrinsic::assume:
769 case Intrinsic::dbg_declare:
770 case Intrinsic::dbg_label:
771 case Intrinsic::dbg_value:
781bool canSkipDef(
MemoryDef *
D,
bool DefVisibleToCaller) {
785 if (
auto *CB = dyn_cast<CallBase>(DI))
786 if (CB->onlyAccessesInaccessibleMemory())
791 if (DI->
mayThrow() && !DefVisibleToCaller)
799 if (isa<FenceInst>(DI))
803 if (isNoopIntrinsic(DI))
832 bool ContainsIrreducibleLoops;
855 bool AnyUnreachableExit;
860 bool ShouldIterateEndOfFunctionDSE;
866 DSEState(
const DSEState &) =
delete;
867 DSEState &operator=(
const DSEState &) =
delete;
872 :
F(
F), AA(AA), EI(DT, &LI), BatchAA(AA, &EI), MSSA(MSSA), DT(DT),
873 PDT(PDT), TLI(TLI),
DL(
F.
getParent()->getDataLayout()), LI(LI) {
878 PostOrderNumbers[BB] = PO++;
881 if (
I.mayThrow() && !MA)
882 ThrowingBlocks.
insert(
I.getParent());
884 auto *MD = dyn_cast_or_null<MemoryDef>(MA);
886 (getLocForWrite(&
I) || isMemTerminatorInst(&
I)))
894 if (AI.hasPassPointeeByValueCopyAttr())
895 InvisibleToCallerAfterRet.
insert({&AI,
true});
901 return isa<UnreachableInst>(E->getTerminator());
907 if (
auto *CB = dyn_cast<CallBase>(
I)) {
910 (F == LibFunc_memset_chk || F == LibFunc_memcpy_chk)) {
919 if (
const auto *Len = dyn_cast<ConstantInt>(CB->getArgOperand(2)))
934 OverwriteResult isOverwrite(
const Instruction *KillingI,
938 int64_t &KillingOff, int64_t &DeadOff) {
942 if (!isGuaranteedLoopIndependent(DeadI, KillingI, DeadLoc))
946 strengthenLocationSize(KillingI, KillingLoc.
Size);
954 if (DeadUndObj == KillingUndObj && KillingLocSize.
isPrecise() &&
956 std::optional<TypeSize> KillingUndObjSize =
958 if (KillingUndObjSize && *KillingUndObjSize == KillingLocSize.
getValue())
967 const auto *KillingMemI = dyn_cast<MemIntrinsic>(KillingI);
968 const auto *DeadMemI = dyn_cast<MemIntrinsic>(DeadI);
969 if (KillingMemI && DeadMemI) {
970 const Value *KillingV = KillingMemI->getLength();
971 const Value *DeadV = DeadMemI->getLength();
972 if (KillingV == DeadV && BatchAA.
isMustAlias(DeadLoc, KillingLoc))
985 const bool AnyScalable =
997 if (KillingSize >= DeadSize)
1004 if (Off >= 0 && (
uint64_t)Off + DeadSize <= KillingSize)
1010 if (DeadUndObj != KillingUndObj) {
1026 const Value *DeadBasePtr =
1028 const Value *KillingBasePtr =
1033 if (DeadBasePtr != KillingBasePtr)
1051 if (DeadOff >= KillingOff) {
1054 if (
uint64_t(DeadOff - KillingOff) + DeadSize <= KillingSize)
1058 else if ((
uint64_t)(DeadOff - KillingOff) < KillingSize)
1059 return OW_MaybePartial;
1063 else if ((
uint64_t)(KillingOff - DeadOff) < DeadSize) {
1064 return OW_MaybePartial;
1071 bool isInvisibleToCallerAfterRet(
const Value *V) {
1072 if (isa<AllocaInst>(V))
1074 auto I = InvisibleToCallerAfterRet.
insert({
V,
false});
1076 if (!isInvisibleToCallerOnUnwind(V)) {
1077 I.first->second =
false;
1082 return I.first->second;
1085 bool isInvisibleToCallerOnUnwind(
const Value *V) {
1086 bool RequiresNoCaptureBeforeUnwind;
1089 if (!RequiresNoCaptureBeforeUnwind)
1092 auto I = CapturedBeforeReturn.
insert({
V,
true});
1099 return !
I.first->second;
1102 std::optional<MemoryLocation> getLocForWrite(
Instruction *
I)
const {
1103 if (!
I->mayWriteToMemory())
1104 return std::nullopt;
1106 if (
auto *CB = dyn_cast<CallBase>(
I))
1115 assert(getLocForWrite(
I) &&
"Must have analyzable write");
1119 return SI->isUnordered();
1121 if (
auto *CB = dyn_cast<CallBase>(
I)) {
1123 if (
auto *
MI = dyn_cast<MemIntrinsic>(CB))
1124 return !
MI->isVolatile();
1128 if (CB->isLifetimeStartOrEnd())
1131 return CB->use_empty() && CB->willReturn() && CB->doesNotThrow() &&
1132 !CB->isTerminator();
1148 if (
auto *CB = dyn_cast<CallBase>(UseInst))
1149 if (CB->onlyAccessesInaccessibleMemory())
1152 int64_t InstWriteOffset, DepWriteOffset;
1153 if (
auto CC = getLocForWrite(UseInst))
1154 return isOverwrite(UseInst, DefInst, *
CC, DefLoc, InstWriteOffset,
1155 DepWriteOffset) == OW_Complete;
1160 bool isWriteAtEndOfFunction(
MemoryDef *Def) {
1162 << *
Def->getMemoryInst()
1163 <<
") is at the end the function \n");
1165 auto MaybeLoc = getLocForWrite(
Def->getMemoryInst());
1167 LLVM_DEBUG(
dbgs() <<
" ... could not get location for write.\n");
1173 auto PushMemUses = [&WorkList, &Visited](
MemoryAccess *Acc) {
1174 if (!Visited.
insert(Acc).second)
1177 WorkList.
push_back(cast<MemoryAccess>(
U.getUser()));
1180 for (
unsigned I = 0;
I < WorkList.
size();
I++) {
1187 if (isa<MemoryPhi>(UseAccess)) {
1191 if (!isGuaranteedLoopInvariant(MaybeLoc->Ptr))
1194 PushMemUses(cast<MemoryPhi>(UseAccess));
1199 Instruction *UseInst = cast<MemoryUseOrDef>(UseAccess)->getMemoryInst();
1200 if (isReadClobber(*MaybeLoc, UseInst)) {
1201 LLVM_DEBUG(
dbgs() <<
" ... hit read clobber " << *UseInst <<
".\n");
1205 if (
MemoryDef *UseDef = dyn_cast<MemoryDef>(UseAccess))
1206 PushMemUses(UseDef);
1214 std::optional<std::pair<MemoryLocation, bool>>
1222 if (
auto *CB = dyn_cast<CallBase>(
I)) {
1227 return std::nullopt;
1233 auto *CB = dyn_cast<CallBase>(
I);
1234 return CB && (CB->getIntrinsicID() == Intrinsic::lifetime_end ||
1242 std::optional<std::pair<MemoryLocation, bool>> MaybeTermLoc =
1243 getLocForTerminator(MaybeTerm);
1254 auto TermLoc = MaybeTermLoc->first;
1255 if (MaybeTermLoc->second) {
1259 int64_t InstWriteOffset = 0;
1260 int64_t DepWriteOffset = 0;
1261 return isOverwrite(MaybeTerm, AccessI, TermLoc, Loc, InstWriteOffset,
1262 DepWriteOffset) == OW_Complete;
1267 if (isNoopIntrinsic(UseInst))
1272 if (
auto SI = dyn_cast<StoreInst>(UseInst))
1278 if (
auto *CB = dyn_cast<CallBase>(UseInst))
1279 if (CB->onlyAccessesInaccessibleMemory())
1290 bool isGuaranteedLoopIndependent(
const Instruction *Current,
1300 if (!ContainsIrreducibleLoops && CurrentLI &&
1304 return isGuaranteedLoopInvariant(CurrentLoc.
Ptr);
1310 bool isGuaranteedLoopInvariant(
const Value *
Ptr) {
1311 Ptr =
Ptr->stripPointerCasts();
1312 if (
auto *
GEP = dyn_cast<GEPOperator>(
Ptr))
1313 if (
GEP->hasAllConstantIndices())
1314 Ptr =
GEP->getPointerOperand()->stripPointerCasts();
1316 if (
auto *
I = dyn_cast<Instruction>(
Ptr)) {
1317 return I->getParent()->isEntryBlock() ||
1318 (!ContainsIrreducibleLoops && !LI.
getLoopFor(
I->getParent()));
1329 std::optional<MemoryAccess *>
1332 unsigned &ScanLimit,
unsigned &WalkerStepLimit,
1333 bool IsMemTerm,
unsigned &PartialLimit) {
1334 if (ScanLimit == 0 || WalkerStepLimit == 0) {
1336 return std::nullopt;
1353 std::optional<MemoryLocation> CurrentLoc;
1354 for (;; Current = cast<MemoryDef>(Current)->getDefiningAccess()) {
1356 dbgs() <<
" visiting " << *Current;
1358 dbgs() <<
" (" << *cast<MemoryUseOrDef>(Current)->getMemoryInst()
1369 return std::nullopt;
1377 if (WalkerStepLimit <= StepCost) {
1379 return std::nullopt;
1381 WalkerStepLimit -= StepCost;
1385 if (isa<MemoryPhi>(Current)) {
1392 MemoryDef *CurrentDef = cast<MemoryDef>(Current);
1395 if (canSkipDef(CurrentDef, !isInvisibleToCallerOnUnwind(KillingUndObj))) {
1396 CanOptimize =
false;
1402 if (mayThrowBetween(KillingI, CurrentI, KillingUndObj)) {
1404 return std::nullopt;
1409 if (isDSEBarrier(KillingUndObj, CurrentI)) {
1411 return std::nullopt;
1418 if (!isa<IntrinsicInst>(CurrentI) && isReadClobber(KillingLoc, CurrentI))
1419 return std::nullopt;
1422 if (
any_of(Current->
uses(), [
this, &KillingLoc, StartAccess](
Use &U) {
1423 if (auto *UseOrDef = dyn_cast<MemoryUseOrDef>(U.getUser()))
1424 return !MSSA.dominates(StartAccess, UseOrDef) &&
1425 isReadClobber(KillingLoc, UseOrDef->getMemoryInst());
1429 return std::nullopt;
1434 CurrentLoc = getLocForWrite(CurrentI);
1435 if (!CurrentLoc || !isRemovable(CurrentI)) {
1436 CanOptimize =
false;
1443 if (!isGuaranteedLoopIndependent(CurrentI, KillingI, *CurrentLoc)) {
1445 CanOptimize =
false;
1453 if (!isMemTerminator(*CurrentLoc, CurrentI, KillingI)) {
1454 CanOptimize =
false;
1458 int64_t KillingOffset = 0;
1459 int64_t DeadOffset = 0;
1460 auto OR = isOverwrite(KillingI, CurrentI, KillingLoc, *CurrentLoc,
1461 KillingOffset, DeadOffset);
1467 (OR == OW_Complete || OR == OW_MaybePartial))
1473 CanOptimize =
false;
1478 if (OR == OW_Unknown || OR == OW_None)
1480 else if (OR == OW_MaybePartial) {
1485 if (PartialLimit <= 1) {
1486 WalkerStepLimit -= 1;
1487 LLVM_DEBUG(
dbgs() <<
" ... reached partial limit ... continue with next access\n");
1504 Instruction *MaybeDeadI = cast<MemoryDef>(MaybeDeadAccess)->getMemoryInst();
1505 LLVM_DEBUG(
dbgs() <<
" Checking for reads of " << *MaybeDeadAccess <<
" ("
1506 << *MaybeDeadI <<
")\n");
1511 WorkList.
insert(cast<MemoryAccess>(
U.getUser()));
1513 PushMemUses(MaybeDeadAccess);
1516 for (
unsigned I = 0;
I < WorkList.
size();
I++) {
1521 if (ScanLimit < (WorkList.
size() -
I)) {
1523 return std::nullopt;
1526 NumDomMemDefChecks++;
1528 if (isa<MemoryPhi>(UseAccess)) {
1533 LLVM_DEBUG(
dbgs() <<
" ... skipping, dominated by killing block\n");
1537 PushMemUses(UseAccess);
1541 Instruction *UseInst = cast<MemoryUseOrDef>(UseAccess)->getMemoryInst();
1547 LLVM_DEBUG(
dbgs() <<
" ... skipping, dominated by killing def\n");
1553 if (isMemTerminator(MaybeDeadLoc, MaybeDeadI, UseInst)) {
1556 <<
" ... skipping, memterminator invalidates following accesses\n");
1560 if (isNoopIntrinsic(cast<MemoryUseOrDef>(UseAccess)->getMemoryInst())) {
1562 PushMemUses(UseAccess);
1566 if (UseInst->
mayThrow() && !isInvisibleToCallerOnUnwind(KillingUndObj)) {
1568 return std::nullopt;
1573 if (isReadClobber(MaybeDeadLoc, UseInst)) {
1575 return std::nullopt;
1581 if (MaybeDeadAccess == UseAccess &&
1582 !isGuaranteedLoopInvariant(MaybeDeadLoc.
Ptr)) {
1583 LLVM_DEBUG(
dbgs() <<
" ... found not loop invariant self access\n");
1584 return std::nullopt;
1590 if (KillingDef == UseAccess || MaybeDeadAccess == UseAccess) {
1605 if (
MemoryDef *UseDef = dyn_cast<MemoryDef>(UseAccess)) {
1606 if (isCompleteOverwrite(MaybeDeadLoc, MaybeDeadI, UseInst)) {
1608 if (PostOrderNumbers.
find(MaybeKillingBlock)->second <
1609 PostOrderNumbers.
find(MaybeDeadAccess->
getBlock())->second) {
1610 if (!isInvisibleToCallerAfterRet(KillingUndObj)) {
1612 <<
" ... found killing def " << *UseInst <<
"\n");
1613 KillingDefs.
insert(UseInst);
1617 <<
" ... found preceeding def " << *UseInst <<
"\n");
1618 return std::nullopt;
1621 PushMemUses(UseDef);
1628 if (!isInvisibleToCallerAfterRet(KillingUndObj)) {
1631 KillingBlocks.
insert(KD->getParent());
1633 "Expected at least a single killing block");
1647 if (!AnyUnreachableExit)
1648 return std::nullopt;
1652 CommonPred =
nullptr;
1656 if (KillingBlocks.
count(CommonPred))
1657 return {MaybeDeadAccess};
1663 WorkList.
insert(CommonPred);
1666 if (!isa<UnreachableInst>(
R->getTerminator()))
1673 for (
unsigned I = 0;
I < WorkList.
size();
I++) {
1676 if (KillingBlocks.
count(Current))
1678 if (Current == MaybeDeadAccess->
getBlock())
1679 return std::nullopt;
1690 return std::nullopt;
1697 return {MaybeDeadAccess};
1708 while (!NowDeadInsts.
empty()) {
1718 bool IsMemDef = MA && isa<MemoryDef>(MA);
1721 auto *MD = cast<MemoryDef>(MA);
1723 if (
auto *SI = dyn_cast<StoreInst>(MD->getMemoryInst())) {
1724 if (
SI->getValueOperand()->getType()->isPointerTy()) {
1726 if (CapturedBeforeReturn.
erase(UO))
1727 ShouldIterateEndOfFunctionDSE =
true;
1728 InvisibleToCallerAfterRet.
erase(UO);
1733 Updater.removeMemoryAccess(MA);
1737 if (
I != IOLs.
end())
1738 I->second.erase(DeadInst);
1741 if (
Instruction *OpI = dyn_cast<Instruction>(O)) {
1765 const Value *KillingUndObj) {
1769 if (KillingUndObj && isInvisibleToCallerOnUnwind(KillingUndObj))
1774 return !ThrowingBlocks.
empty();
1785 if (DeadI->
mayThrow() && !isInvisibleToCallerOnUnwind(KillingUndObj))
1791 if (
auto *LI = dyn_cast<LoadInst>(DeadI))
1793 if (
auto *SI = dyn_cast<StoreInst>(DeadI))
1795 if (
auto *ARMW = dyn_cast<AtomicRMWInst>(DeadI))
1797 if (
auto *CmpXchg = dyn_cast<AtomicCmpXchgInst>(DeadI))
1807 bool eliminateDeadWritesAtEndOfFunction() {
1808 bool MadeChange =
false;
1811 <<
"Trying to eliminate MemoryDefs at the end of the function\n");
1813 ShouldIterateEndOfFunctionDSE =
false;
1819 auto DefLoc = getLocForWrite(DefI);
1820 if (!DefLoc || !isRemovable(DefI))
1829 if (!isInvisibleToCallerAfterRet(UO))
1832 if (isWriteAtEndOfFunction(Def)) {
1834 LLVM_DEBUG(
dbgs() <<
" ... MemoryDef is not accessed until the end "
1835 "of the function\n");
1841 }
while (ShouldIterateEndOfFunctionDSE);
1849 MemSetInst *MemSet = dyn_cast<MemSetInst>(DefI);
1854 if (!StoredConstant || !StoredConstant->
isNullValue())
1857 if (!isRemovable(DefI))
1861 if (
F.hasFnAttribute(Attribute::SanitizeMemory) ||
1862 F.hasFnAttribute(Attribute::SanitizeAddress) ||
1863 F.hasFnAttribute(Attribute::SanitizeHWAddress) ||
1864 F.getName() ==
"calloc")
1866 auto *
Malloc =
const_cast<CallInst *
>(dyn_cast<CallInst>(DefUO));
1869 auto *InnerCallee =
Malloc->getCalledFunction();
1873 if (!TLI.
getLibFunc(*InnerCallee, Func) || !TLI.
has(Func) ||
1874 Func != LibFunc_malloc)
1884 auto *MallocBB =
Malloc->getParent(),
1885 *MemsetBB = Memset->getParent();
1886 if (MallocBB == MemsetBB)
1888 auto *
Ptr = Memset->getArgOperand(0);
1889 auto *TI = MallocBB->getTerminator();
1895 if (Pred != ICmpInst::ICMP_EQ || MemsetBB != FalseBB)
1902 if (!shouldCreateCalloc(
Malloc, MemSet) ||
1907 Type *SizeTTy =
Malloc->getArgOperand(0)->getType();
1908 auto *Calloc =
emitCalloc(ConstantInt::get(SizeTTy, 1),
1909 Malloc->getArgOperand(0), IRB, TLI);
1915 Updater.createMemoryAccessAfter(cast<Instruction>(Calloc),
nullptr,
1917 auto *NewAccessMD = cast<MemoryDef>(NewAccess);
1918 Updater.insertDef(NewAccessMD,
true);
1919 Malloc->replaceAllUsesWith(Calloc);
1926 bool dominatingConditionImpliesValue(
MemoryDef *Def) {
1927 auto *StoreI = cast<StoreInst>(
Def->getMemoryInst());
1929 Value *StorePtr = StoreI->getPointerOperand();
1930 Value *StoreVal = StoreI->getValueOperand();
1937 if (!BI || !BI->isConditional())
1943 if (BI->getSuccessor(0) == BI->getSuccessor(1))
1948 if (!
match(BI->getCondition(),
1958 if (Pred == ICmpInst::ICMP_EQ &&
1963 if (Pred == ICmpInst::ICMP_NE &&
1972 return MSSA.
dominates(ClobAcc, LoadAcc);
1980 MemSetInst *MemSet = dyn_cast<MemSetInst>(DefI);
1981 Constant *StoredConstant =
nullptr;
1983 StoredConstant = dyn_cast<Constant>(
Store->getOperand(0));
1985 StoredConstant = dyn_cast<Constant>(MemSet->
getValue());
1989 if (!isRemovable(DefI))
1992 if (StoredConstant) {
1997 if (InitC && InitC == StoredConstant)
2005 if (dominatingConditionImpliesValue(Def))
2008 if (
auto *LoadI = dyn_cast<LoadInst>(
Store->getOperand(0))) {
2009 if (LoadI->getPointerOperand() ==
Store->getOperand(1)) {
2013 if (LoadAccess ==
Def->getDefiningAccess())
2028 for (
unsigned I = 1;
I < ToCheck.
size(); ++
I) {
2029 Current = ToCheck[
I];
2030 if (
auto PhiAccess = dyn_cast<MemoryPhi>(Current)) {
2032 for (
auto &
Use : PhiAccess->incoming_values())
2033 ToCheck.
insert(cast<MemoryAccess>(&
Use));
2039 assert(isa<MemoryDef>(Current) &&
2040 "Only MemoryDefs should reach here.");
2045 if (LoadAccess != Current)
2056 bool Changed =
false;
2057 for (
auto OI : IOL) {
2060 assert(isRemovable(DeadI) &&
"Expect only removable instruction");
2063 int64_t DeadStart = 0;
2077 bool eliminateRedundantStoresOfExistingValues() {
2078 bool MadeChange =
false;
2079 LLVM_DEBUG(
dbgs() <<
"Trying to eliminate MemoryDefs that write the "
2080 "already existing value\n");
2081 for (
auto *Def : MemDefs) {
2086 auto MaybeDefLoc = getLocForWrite(DefInst);
2087 if (!MaybeDefLoc || !isRemovable(DefInst))
2094 if (
Def->isOptimized())
2095 UpperDef = dyn_cast<MemoryDef>(
Def->getOptimized());
2097 UpperDef = dyn_cast<MemoryDef>(
Def->getDefiningAccess());
2102 auto IsRedundantStore = [&]() {
2105 if (
auto *MemSetI = dyn_cast<MemSetInst>(UpperInst)) {
2106 if (
auto *SI = dyn_cast<StoreInst>(DefInst)) {
2109 int64_t InstWriteOffset = 0;
2110 int64_t DepWriteOffset = 0;
2111 auto OR = isOverwrite(UpperInst, DefInst, UpperLoc, *MaybeDefLoc,
2112 InstWriteOffset, DepWriteOffset);
2114 return StoredByte && StoredByte == MemSetI->getOperand(1) &&
2121 if (!IsRedundantStore() || isReadClobber(*MaybeDefLoc, DefInst))
2123 LLVM_DEBUG(
dbgs() <<
"DSE: Remove No-Op Store:\n DEAD: " << *DefInst
2126 NumRedundantStores++;
2137 bool MadeChange =
false;
2139 DSEState State(
F, AA, MSSA, DT, PDT, TLI, LI);
2141 for (
unsigned I = 0;
I < State.MemDefs.size();
I++) {
2143 if (State.SkipStores.count(KillingDef))
2147 std::optional<MemoryLocation> MaybeKillingLoc;
2148 if (State.isMemTerminatorInst(KillingI)) {
2149 if (
auto KillingLoc = State.getLocForTerminator(KillingI))
2150 MaybeKillingLoc = KillingLoc->first;
2152 MaybeKillingLoc = State.getLocForWrite(KillingI);
2155 if (!MaybeKillingLoc) {
2156 LLVM_DEBUG(
dbgs() <<
"Failed to find analyzable write location for "
2157 << *KillingI <<
"\n");
2161 assert(KillingLoc.
Ptr &&
"KillingLoc should not be null");
2164 << *KillingDef <<
" (" << *KillingI <<
")\n");
2173 bool Shortend =
false;
2174 bool IsMemTerm = State.isMemTerminatorInst(KillingI);
2176 for (
unsigned I = 0;
I < ToCheck.
size();
I++) {
2178 if (State.SkipStores.count(Current))
2181 std::optional<MemoryAccess *> MaybeDeadAccess = State.getDomMemoryDef(
2182 KillingDef, Current, KillingLoc, KillingUndObj, ScanLimit,
2183 WalkerStepLimit, IsMemTerm, PartialLimit);
2185 if (!MaybeDeadAccess) {
2191 LLVM_DEBUG(
dbgs() <<
" Checking if we can kill " << *DeadAccess);
2192 if (isa<MemoryPhi>(DeadAccess)) {
2193 LLVM_DEBUG(
dbgs() <<
"\n ... adding incoming values to worklist\n");
2194 for (
Value *V : cast<MemoryPhi>(DeadAccess)->incoming_values()) {
2202 if (State.PostOrderNumbers[IncomingBlock] >
2203 State.PostOrderNumbers[PhiBlock])
2204 ToCheck.
insert(IncomingAccess);
2208 auto *DeadDefAccess = cast<MemoryDef>(DeadAccess);
2209 Instruction *DeadI = DeadDefAccess->getMemoryInst();
2211 ToCheck.
insert(DeadDefAccess->getDefiningAccess());
2212 NumGetDomMemoryDefPassed++;
2221 if (KillingUndObj != DeadUndObj)
2223 LLVM_DEBUG(
dbgs() <<
"DSE: Remove Dead Store:\n DEAD: " << *DeadI
2224 <<
"\n KILLER: " << *KillingI <<
'\n');
2225 State.deleteDeadInstruction(DeadI);
2230 int64_t KillingOffset = 0;
2231 int64_t DeadOffset = 0;
2232 OverwriteResult
OR = State.isOverwrite(
2233 KillingI, DeadI, KillingLoc, DeadLoc, KillingOffset, DeadOffset);
2234 if (OR == OW_MaybePartial) {
2235 auto Iter = State.IOLs.insert(
2236 std::make_pair<BasicBlock *, InstOverlapIntervalsTy>(
2238 auto &IOL = Iter.first->second;
2240 DeadOffset, DeadI, IOL);
2244 auto *DeadSI = dyn_cast<StoreInst>(DeadI);
2245 auto *KillingSI = dyn_cast<StoreInst>(KillingI);
2249 if (DeadSI && KillingSI && DT.
dominates(DeadSI, KillingSI)) {
2251 KillingSI, DeadSI, KillingOffset, DeadOffset, State.DL,
2252 State.BatchAA, &DT)) {
2255 DeadSI->setOperand(0, Merged);
2256 ++NumModifiedStores;
2262 State.deleteDeadInstruction(KillingSI);
2263 auto I = State.IOLs.find(DeadSI->getParent());
2264 if (
I != State.IOLs.end())
2265 I->second.erase(DeadSI);
2271 if (OR == OW_Complete) {
2272 LLVM_DEBUG(
dbgs() <<
"DSE: Remove Dead Store:\n DEAD: " << *DeadI
2273 <<
"\n KILLER: " << *KillingI <<
'\n');
2274 State.deleteDeadInstruction(DeadI);
2282 if (!Shortend && State.storeIsNoop(KillingDef, KillingUndObj)) {
2283 LLVM_DEBUG(
dbgs() <<
"DSE: Remove No-Op Store:\n DEAD: " << *KillingI
2285 State.deleteDeadInstruction(KillingI);
2286 NumRedundantStores++;
2292 if (!Shortend && State.tryFoldIntoCalloc(KillingDef, KillingUndObj)) {
2293 LLVM_DEBUG(
dbgs() <<
"DSE: Remove memset after forming calloc:\n"
2294 <<
" DEAD: " << *KillingI <<
'\n');
2295 State.deleteDeadInstruction(KillingI);
2302 for (
auto &KV : State.IOLs)
2303 MadeChange |= State.removePartiallyOverlappedStores(KV.second);
2305 MadeChange |= State.eliminateRedundantStoresOfExistingValues();
2306 MadeChange |= State.eliminateDeadWritesAtEndOfFunction();
2308 while (!State.ToRemove.empty()) {
2309 Instruction *DeadInst = State.ToRemove.pop_back_val();
2328 bool Changed = eliminateDeadStores(
F, AA, MSSA, DT, PDT, TLI, LI);
2330#ifdef LLVM_ENABLE_STATS
2333 NumRemainingStores += isa<StoreInst>(&
I);
MachineBasicBlock MachineBasicBlock::iterator DebugLoc DL
This file implements a class to represent arbitrary precision integral constant values and operations...
ReachingDefAnalysis InstSet & ToRemove
Expand Atomic instructions
static const Function * getParent(const Value *V)
static GCRegistry::Add< OcamlGC > B("ocaml", "ocaml 3.10-compatible GC")
static GCRegistry::Add< StatepointGC > D("statepoint-example", "an example strategy for statepoint")
This file contains the declarations for the subclasses of Constant, which represent the different fla...
static void shortenAssignment(Instruction *Inst, Value *OriginalDest, uint64_t OldOffsetInBits, uint64_t OldSizeInBits, uint64_t NewSizeInBits, bool IsOverwriteEnd)
static bool isShortenableAtTheEnd(Instruction *I)
Returns true if the end of this instruction can be safely shortened in length.
static cl::opt< bool > EnablePartialStoreMerging("enable-dse-partial-store-merging", cl::init(true), cl::Hidden, cl::desc("Enable partial store merging in DSE"))
static bool tryToShortenBegin(Instruction *DeadI, OverlapIntervalsTy &IntervalMap, int64_t &DeadStart, uint64_t &DeadSize)
std::map< int64_t, int64_t > OverlapIntervalsTy
static bool isShortenableAtTheBeginning(Instruction *I)
Returns true if the beginning of this instruction can be safely shortened in length.
static cl::opt< unsigned > MemorySSADefsPerBlockLimit("dse-memoryssa-defs-per-block-limit", cl::init(5000), cl::Hidden, cl::desc("The number of MemoryDefs we consider as candidates to eliminated " "other stores per basic block (default = 5000)"))
static Constant * tryToMergePartialOverlappingStores(StoreInst *KillingI, StoreInst *DeadI, int64_t KillingOffset, int64_t DeadOffset, const DataLayout &DL, BatchAAResults &AA, DominatorTree *DT)
static bool memoryIsNotModifiedBetween(Instruction *FirstI, Instruction *SecondI, BatchAAResults &AA, const DataLayout &DL, DominatorTree *DT)
Returns true if the memory which is accessed by the second instruction is not modified between the fi...
static OverwriteResult isMaskedStoreOverwrite(const Instruction *KillingI, const Instruction *DeadI, BatchAAResults &AA)
Check if two instruction are masked stores that completely overwrite one another.
static cl::opt< unsigned > MemorySSAOtherBBStepCost("dse-memoryssa-otherbb-cost", cl::init(5), cl::Hidden, cl::desc("The cost of a step in a different basic " "block than the killing MemoryDef" "(default = 5)"))
static bool tryToShorten(Instruction *DeadI, int64_t &DeadStart, uint64_t &DeadSize, int64_t KillingStart, uint64_t KillingSize, bool IsOverwriteEnd)
static cl::opt< unsigned > MemorySSAScanLimit("dse-memoryssa-scanlimit", cl::init(150), cl::Hidden, cl::desc("The number of memory instructions to scan for " "dead store elimination (default = 150)"))
static cl::opt< unsigned > MemorySSASameBBStepCost("dse-memoryssa-samebb-cost", cl::init(1), cl::Hidden, cl::desc("The cost of a step in the same basic block as the killing MemoryDef" "(default = 1)"))
static cl::opt< bool > EnablePartialOverwriteTracking("enable-dse-partial-overwrite-tracking", cl::init(true), cl::Hidden, cl::desc("Enable partial-overwrite tracking in DSE"))
static OverwriteResult isPartialOverwrite(const MemoryLocation &KillingLoc, const MemoryLocation &DeadLoc, int64_t KillingOff, int64_t DeadOff, Instruction *DeadI, InstOverlapIntervalsTy &IOL)
Return 'OW_Complete' if a store to the 'KillingLoc' location completely overwrites a store to the 'De...
static cl::opt< unsigned > MemorySSAPartialStoreLimit("dse-memoryssa-partial-store-limit", cl::init(5), cl::Hidden, cl::desc("The maximum number candidates that only partially overwrite the " "killing MemoryDef to consider" " (default = 5)"))
static std::optional< TypeSize > getPointerSize(const Value *V, const DataLayout &DL, const TargetLibraryInfo &TLI, const Function *F)
static bool tryToShortenEnd(Instruction *DeadI, OverlapIntervalsTy &IntervalMap, int64_t &DeadStart, uint64_t &DeadSize)
static cl::opt< unsigned > MemorySSAUpwardsStepLimit("dse-memoryssa-walklimit", cl::init(90), cl::Hidden, cl::desc("The maximum number of steps while walking upwards to find " "MemoryDefs that may be killed (default = 90)"))
static cl::opt< bool > OptimizeMemorySSA("dse-optimize-memoryssa", cl::init(true), cl::Hidden, cl::desc("Allow DSE to optimize memory accesses."))
static cl::opt< unsigned > MemorySSAPathCheckLimit("dse-memoryssa-path-check-limit", cl::init(50), cl::Hidden, cl::desc("The maximum number of blocks to check when trying to prove that " "all paths to an exit go through a killing block (default = 50)"))
This file provides an implementation of debug counters.
#define DEBUG_COUNTER(VARNAME, COUNTERNAME, DESC)
This file defines the DenseMap class.
This is the interface for a simple mod/ref and alias analysis over globals.
static void deleteDeadInstruction(Instruction *I)
This file implements a map that provides insertion order iteration.
This file provides utility analysis objects describing memory locations.
This file exposes an interface to building/using memory SSA to walk memory instructions using a use/d...
Module.h This file contains the declarations for the Module class.
Contains a collection of routines for determining if a given instruction is guaranteed to execute if ...
This header defines various interfaces for pass management in LLVM.
This file builds on the ADT/GraphTraits.h file to build a generic graph post order iterator.
assert(ImpDefSCC.getReg()==AMDGPU::SCC &&ImpDefSCC.isDef())
This file implements a set that has insertion order iteration characteristics.
This file defines the SmallPtrSet class.
This file defines the SmallVector class.
This file defines the 'Statistic' class, which is designed to be an easy way to expose various metric...
#define STATISTIC(VARNAME, DESC)
A manager for alias analyses.
Class for arbitrary precision integers.
APInt zext(unsigned width) const
Zero extend to a new width.
static APInt getBitsSet(unsigned numBits, unsigned loBit, unsigned hiBit)
Get a value with a block of bits set.
unsigned getBitWidth() const
Return the number of bits in the APInt.
The possible results of an alias query.
@ NoAlias
The two locations do not alias at all.
@ PartialAlias
The two locations alias, but only due to a partial overlap.
@ MustAlias
The two locations precisely alias each other.
constexpr int32_t getOffset() const
constexpr bool hasOffset() const
A container for analyses that lazily runs them and caches their results.
PassT::Result & getResult(IRUnitT &IR, ExtraArgTs... ExtraArgs)
Get the result of an analysis pass for a given IR unit.
This class represents an incoming formal argument to a Function.
LLVM Basic Block Representation.
const Function * getParent() const
Return the enclosing method, or null if none.
InstListType::iterator iterator
Instruction iterators...
const Instruction * getTerminator() const LLVM_READONLY
Returns the terminator instruction if the block is well formed or null if the block is not well forme...
This class is a wrapper over an AAResults, and it is intended to be used only when there are no IR ch...
AliasResult alias(const MemoryLocation &LocA, const MemoryLocation &LocB)
bool isMustAlias(const MemoryLocation &LocA, const MemoryLocation &LocB)
ModRefInfo getModRefInfo(const Instruction *I, const std::optional< MemoryLocation > &OptLoc)
Represents analyses that only rely on functions' control flow.
This class represents a function call, abstracting a target machine's calling convention.
Predicate
This enumeration lists the possible predicates for CmpInst subclasses.
This is an important base class in LLVM.
bool isNullValue() const
Return true if this is the value that would be returned by getNullValue.
static DIAssignID * getDistinct(LLVMContext &Context)
static std::optional< DIExpression * > createFragmentExpression(const DIExpression *Expr, unsigned OffsetInBits, unsigned SizeInBits)
Create a DIExpression to describe one part of an aggregate variable that is fragmented across multipl...
PreservedAnalyses run(Function &F, FunctionAnalysisManager &FAM)
A parsed version of the target data layout string in and methods for querying it.
static bool shouldExecute(unsigned CounterName)
iterator find(const_arg_type_t< KeyT > Val)
bool erase(const KeyT &Val)
std::pair< iterator, bool > insert(const std::pair< KeyT, ValueT > &KV)
DomTreeNodeBase * getIDom() const
Analysis pass which computes a DominatorTree.
NodeT * findNearestCommonDominator(NodeT *A, NodeT *B) const
Find nearest common dominator basic block for basic block A and B.
iterator_range< root_iterator > roots()
DomTreeNodeBase< NodeT > * getNode(const NodeT *BB) const
getNode - return the (Post)DominatorTree node for the specified basic block.
bool properlyDominates(const DomTreeNodeBase< NodeT > *A, const DomTreeNodeBase< NodeT > *B) const
properlyDominates - Returns true iff A dominates B and A != B.
Concrete subclass of DominatorTreeBase that is used to compute a normal dominator tree.
bool isReachableFromEntry(const Use &U) const
Provide an overload for a Use.
bool dominates(const BasicBlock *BB, const Use &U) const
Return true if the (end of the) basic block BB dominates the use U.
Context-sensitive CaptureInfo provider, which computes and caches the earliest common dominator closu...
void removeInstruction(Instruction *I)
const BasicBlock & getEntryBlock() const
static GetElementPtrInst * CreateInBounds(Type *PointeeType, Value *Ptr, ArrayRef< Value * > IdxList, const Twine &NameStr, BasicBlock::iterator InsertBefore)
Create an "inbounds" getelementptr.
bool isEquality() const
Return true if this predicate is either EQ or NE.
This provides a uniform API for creating instructions and inserting them into a basic block: either a...
bool mayThrow(bool IncludePhaseOneUnwind=false) const LLVM_READONLY
Return true if this instruction may throw an exception.
bool mayWriteToMemory() const LLVM_READONLY
Return true if this instruction may modify memory.
const Module * getModule() const
Return the module owning the function this instruction belongs to or nullptr it the function does not...
bool isAtomic() const LLVM_READONLY
Return true if this instruction has an AtomicOrdering of unordered or higher.
const BasicBlock * getParent() const
InstListType::iterator eraseFromParent()
This method unlinks 'this' from the containing basic block and deletes it.
bool mayReadFromMemory() const LLVM_READONLY
Return true if this instruction may read memory.
bool isIdenticalTo(const Instruction *I) const LLVM_READONLY
Return true if the specified instruction is exactly identical to the current one.
void setDebugLoc(DebugLoc Loc)
Set the debug location information for this instruction.
const_iterator begin() const
bool empty() const
empty - Return true when no intervals are mapped.
const_iterator end() const
A wrapper class for inspecting calls to intrinsic functions.
This is an important class for using LLVM in a threaded context.
static LocationSize precise(uint64_t Value)
TypeSize getValue() const
Analysis pass that exposes the LoopInfo for a function.
LoopT * getLoopFor(const BlockT *BB) const
Return the inner most loop that BB lives in.
Represents a single loop in the control flow graph.
This class implements a map that also provides access to all stored values in a deterministic order.
iterator find(const KeyT &Key)
Value * getLength() const
This class wraps the llvm.memset and llvm.memset.inline intrinsics.
BasicBlock * getBlock() const
Represents a read-write access to memory, whether it is a must-alias, or a may-alias.
void setOptimized(MemoryAccess *MA)
Representation for a specific memory location.
static MemoryLocation get(const LoadInst *LI)
Return a location with information about the memory reference by the given instruction.
LocationSize Size
The maximum size of the location, in address-units, or UnknownSize if the size is not known.
static MemoryLocation getAfter(const Value *Ptr, const AAMDNodes &AATags=AAMDNodes())
Return a location that may access any location after Ptr, while remaining within the underlying objec...
MemoryLocation getWithNewPtr(const Value *NewPtr) const
const Value * Ptr
The address of the start of the location.
static MemoryLocation getForDest(const MemIntrinsic *MI)
Return a location representing the destination of a memory set or transfer.
static std::optional< MemoryLocation > getOrNone(const Instruction *Inst)
An analysis that produces MemorySSA for a function.
MemoryAccess * getClobberingMemoryAccess(const Instruction *I, BatchAAResults &AA)
Given a memory Mod/Ref/ModRef'ing instruction, calling this will give you the nearest dominating Memo...
Encapsulates MemorySSA, including all data associated with memory accesses.
MemorySSAWalker * getSkipSelfWalker()
bool dominates(const MemoryAccess *A, const MemoryAccess *B) const
Given two memory accesses in potentially different blocks, determine whether MemoryAccess A dominates...
MemorySSAWalker * getWalker()
MemoryUseOrDef * getMemoryAccess(const Instruction *I) const
Given a memory Mod/Ref'ing instruction, get the MemorySSA access associated with it.
bool isLiveOnEntryDef(const MemoryAccess *MA) const
Return true if MA represents the live on entry value.
MemoryAccess * getDefiningAccess() const
Get the access that produces the memory state used by this Use.
Instruction * getMemoryInst() const
Get the instruction that this MemoryUse represents.
const DataLayout & getDataLayout() const
Get the data layout for the module's target platform.
PHITransAddr - An address value which tracks and handles phi translation.
Value * translateValue(BasicBlock *CurBB, BasicBlock *PredBB, const DominatorTree *DT, bool MustDominate)
translateValue - PHI translate the current address up the CFG from CurBB to Pred, updating our state ...
bool isPotentiallyPHITranslatable() const
isPotentiallyPHITranslatable - If this needs PHI translation, return true if we have some hope of doi...
bool needsPHITranslationFromBlock(BasicBlock *BB) const
needsPHITranslationFromBlock - Return true if moving from the specified BasicBlock to its predecessor...
static PoisonValue * get(Type *T)
Static factory methods - Return an 'poison' object of the specified type.
Analysis pass which computes a PostDominatorTree.
PostDominatorTree Class - Concrete subclass of DominatorTree that is used to compute the post-dominat...
bool dominates(const Instruction *I1, const Instruction *I2) const
Return true if I1 dominates I2.
A set of analyses that are preserved following a run of a transformation pass.
static PreservedAnalyses all()
Construct a special preserved set that preserves all passes.
void preserveSet()
Mark an analysis set as preserved.
void preserve()
Mark an analysis as preserved.
A vector that has set insertion semantics.
size_type size() const
Determine the number of elements in the SetVector.
bool insert(const value_type &X)
Insert a new element into the SetVector.
size_type count(ConstPtrType Ptr) const
count - Return 1 if the specified pointer is in the set, 0 otherwise.
std::pair< iterator, bool > insert(PtrType Ptr)
Inserts Ptr if and only if there is no element in the container equal to Ptr.
bool contains(ConstPtrType Ptr) const
SmallPtrSet - This class implements a set which is optimized for holding SmallSize or less elements.
A SetVector that performs no allocations if smaller than a certain size.
void push_back(const T &Elt)
This is a 'vector' (really, a variable-sized array), optimized for the case when the array is small.
An instruction for storing to memory.
Value * getValueOperand()
Analysis pass providing the TargetLibraryInfo.
Provides information about what library functions are available for the current target.
bool has(LibFunc F) const
Tests whether a library function is available.
bool getLibFunc(StringRef funcName, LibFunc &F) const
Searches for a particular function name.
static constexpr TypeSize getFixed(ScalarTy ExactSize)
The instances of the Type class are immutable: once they are created, they are never changed.
static IntegerType * getInt8Ty(LLVMContext &C)
bool isVoidTy() const
Return true if this is 'void'.
A Use represents the edge between a Value definition and its users.
LLVM Value Representation.
Type * getType() const
All values are typed, get the type of this value.
const Value * stripPointerCasts() const
Strip off pointer casts, all-zero GEPs and address space casts.
LLVMContext & getContext() const
All values hold a context through their type.
iterator_range< use_iterator > uses()
constexpr bool isScalable() const
Returns whether the quantity is scaled by a runtime quantity (vscale).
self_iterator getIterator()
#define llvm_unreachable(msg)
Marks that the current location is not supposed to be reachable.
bool match(Val *V, const Pattern &P)
bind_ty< Instruction > m_Instruction(Instruction *&I)
Match an instruction, capturing it if we match.
specificval_ty m_Specific(const Value *V)
Match if we have a specific specified value.
class_match< ConstantInt > m_ConstantInt()
Match an arbitrary ConstantInt and ignore it.
match_combine_and< LTy, RTy > m_CombineAnd(const LTy &L, const RTy &R)
Combine two pattern matchers matching L && R.
CmpClass_match< LHS, RHS, ICmpInst, ICmpInst::Predicate > m_ICmp(ICmpInst::Predicate &Pred, const LHS &L, const RHS &R)
OneOps_match< OpTy, Instruction::Load > m_Load(const OpTy &Op)
Matches LoadInst.
brc_match< Cond_t, bind_ty< BasicBlock >, bind_ty< BasicBlock > > m_Br(const Cond_t &C, BasicBlock *&T, BasicBlock *&F)
CmpClass_match< LHS, RHS, ICmpInst, ICmpInst::Predicate, true > m_c_ICmp(ICmpInst::Predicate &Pred, const LHS &L, const RHS &R)
Matches an ICmp with a predicate over LHS and RHS in either order.
class_match< Value > m_Value()
Match an arbitrary value and ignore it.
is_zero m_Zero()
Match any null constant or a vector with all elements equal to 0.
AssignmentMarkerRange getAssignmentMarkers(DIAssignID *ID)
Return a range of dbg.assign intrinsics which use \ID as an operand.
SmallVector< DbgVariableRecord * > getDVRAssignmentMarkers(const Instruction *Inst)
bool calculateFragmentIntersect(const DataLayout &DL, const Value *Dest, uint64_t SliceOffsetInBits, uint64_t SliceSizeInBits, const DbgAssignIntrinsic *DbgAssign, std::optional< DIExpression::FragmentInfo > &Result)
Calculate the fragment of the variable in DAI covered from (Dest + SliceOffsetInBits) to to (Dest + S...
initializer< Ty > init(const Ty &Val)
NodeAddr< DefNode * > Def
NodeAddr< FuncNode * > Func
This is an optimization pass for GlobalISel generic memory operations.
auto drop_begin(T &&RangeOrContainer, size_t N=1)
Return a range covering RangeOrContainer with the first N elements excluded.
UnaryFunction for_each(R &&Range, UnaryFunction F)
Provide wrappers to std::for_each which take ranges instead of having to pass begin/end explicitly.
Constant * getInitialValueOfAllocation(const Value *V, const TargetLibraryInfo *TLI, Type *Ty)
If this is a call to an allocation function that initializes memory to a fixed value,...
bool isStrongerThanMonotonic(AtomicOrdering AO)
bool isAligned(Align Lhs, uint64_t SizeInBytes)
Checks that SizeInBytes is a multiple of the alignment.
void salvageDebugInfo(const MachineRegisterInfo &MRI, MachineInstr &MI)
Assuming the instruction MI is going to be deleted, attempt to salvage debug users of MI by writing t...
Value * emitCalloc(Value *Num, Value *Size, IRBuilderBase &B, const TargetLibraryInfo &TLI)
Emit a call to the calloc function.
Value * GetPointerBaseWithConstantOffset(Value *Ptr, int64_t &Offset, const DataLayout &DL, bool AllowNonInbounds=true)
Analyze the specified pointer to see if it can be expressed as a base pointer plus a constant offset.
const Value * getUnderlyingObject(const Value *V, unsigned MaxLookup=6)
This method strips off any GEP address adjustments and pointer casts from the specified value,...
iterator_range< po_iterator< T > > post_order(const T &G)
bool isNoAliasCall(const Value *V)
Return true if this pointer is returned by a noalias function.
bool any_of(R &&range, UnaryPredicate P)
Provide wrappers to std::any_of which take ranges instead of having to pass begin/end explicitly.
bool isInstructionTriviallyDead(Instruction *I, const TargetLibraryInfo *TLI=nullptr)
Return true if the result produced by the instruction is not used, and the instruction will return.
bool getObjectSize(const Value *Ptr, uint64_t &Size, const DataLayout &DL, const TargetLibraryInfo *TLI, ObjectSizeOpts Opts={})
Compute the size of the object pointed by Ptr.
auto reverse(ContainerTy &&C)
bool isModSet(const ModRefInfo MRI)
bool PointerMayBeCaptured(const Value *V, bool ReturnCaptures, bool StoreCaptures, unsigned MaxUsesToExplore=0)
PointerMayBeCaptured - Return true if this pointer value may be captured by the enclosing function (w...
bool NullPointerIsDefined(const Function *F, unsigned AS=0)
Check whether null pointer dereferencing is considered undefined behavior for a given function or an ...
raw_ostream & dbgs()
dbgs() - This returns a reference to a raw_ostream for debugging messages.
bool AreStatisticsEnabled()
Check if statistics are enabled.
bool isNotVisibleOnUnwind(const Value *Object, bool &RequiresNoCaptureBeforeUnwind)
Return true if Object memory is not visible after an unwind, in the sense that program semantics cann...
uint64_t offsetToAlignment(uint64_t Value, Align Alignment)
Returns the offset to the next integer (mod 2**64) that is greater than or equal to Value and is a mu...
bool salvageKnowledge(Instruction *I, AssumptionCache *AC=nullptr, DominatorTree *DT=nullptr)
Calls BuildAssumeFromInst and if the resulting llvm.assume is valid insert if before I.
Value * getFreedOperand(const CallBase *CB, const TargetLibraryInfo *TLI)
If this if a call to a free function, return the freed operand.
Value * isBytewiseValue(Value *V, const DataLayout &DL)
If the specified value can be set by repeating the same byte in memory, return the i8 value that it i...
auto predecessors(const MachineBasicBlock *BB)
bool mayContainIrreducibleControl(const Function &F, const LoopInfo *LI)
bool isIdentifiedObject(const Value *V)
Return true if this pointer refers to a distinct and identifiable object.
bool isStrongerThan(AtomicOrdering AO, AtomicOrdering Other)
Returns true if ao is stronger than other as defined by the AtomicOrdering lattice,...
bool isRefSet(const ModRefInfo MRI)
This struct is a compact representation of a valid (non-zero power of two) alignment.
uint64_t value() const
This is a hole in the type system and should not be abused.
Holds the characteristics of one fragment of a larger variable.
Various options to control the behavior of getObjectSize.
bool NullIsUnknownSize
If this is true, null pointers in address space 0 will be treated as though they can't be evaluated.