64#define DEBUG_TYPE "memcpyopt"
67 "enable-memcpyopt-without-libcalls",
cl::Hidden,
68 cl::desc(
"Enable memcpyopt even when libcalls are disabled"));
70STATISTIC(NumMemCpyInstr,
"Number of memcpy instructions deleted");
71STATISTIC(NumMemSetInfer,
"Number of memsets inferred");
72STATISTIC(NumMoveToCpy,
"Number of memmoves converted to memcpy");
73STATISTIC(NumCpyToSet,
"Number of memcpys converted to memset");
74STATISTIC(NumCallSlot,
"Number of call slot optimizations performed");
103 bool isProfitableToUseMemset(
const DataLayout &
DL)
const;
108bool MemsetRange::isProfitableToUseMemset(
const DataLayout &
DL)
const {
110 if (TheStores.size() >= 4 || End-Start >= 16)
return true;
113 if (TheStores.size() < 2)
return false;
118 if (!isa<StoreInst>(SI))
123 if (TheStores.size() == 2)
return false;
135 unsigned Bytes =
unsigned(End-Start);
136 unsigned MaxIntSize =
DL.getLargestLegalIntTypeSizeInBits() / 8;
139 unsigned NumPointerStores = Bytes / MaxIntSize;
142 unsigned NumByteStores = Bytes % MaxIntSize;
147 return TheStores.size() > NumPointerStores+NumByteStores;
167 bool empty()
const {
return Ranges.empty(); }
169 void addInst(int64_t OffsetFromFirst,
Instruction *Inst) {
170 if (
auto *SI = dyn_cast<StoreInst>(Inst))
171 addStore(OffsetFromFirst, SI);
173 addMemSet(OffsetFromFirst, cast<MemSetInst>(Inst));
176 void addStore(int64_t OffsetFromFirst,
StoreInst *SI) {
177 TypeSize StoreSize =
DL.getTypeStoreSize(
SI->getOperand(0)->getType());
180 SI->getPointerOperand(),
SI->getAlign(), SI);
183 void addMemSet(int64_t OffsetFromFirst,
MemSetInst *MSI) {
184 int64_t
Size = cast<ConstantInt>(MSI->
getLength())->getZExtValue();
197void MemsetRanges::addRange(int64_t Start, int64_t
Size,
Value *
Ptr,
199 int64_t End = Start+
Size;
202 Ranges, [=](
const MemsetRange &O) {
return O.End < Start; });
207 if (
I ==
Ranges.end() || End < I->Start) {
208 MemsetRange &
R = *
Ranges.insert(
I, MemsetRange());
212 R.Alignment = Alignment;
213 R.TheStores.push_back(Inst);
218 I->TheStores.push_back(Inst);
222 if (
I->Start <= Start &&
I->End >= End)
231 if (Start < I->Start) {
234 I->Alignment = Alignment;
242 range_iterator NextI =
I;
243 while (++NextI !=
Ranges.end() && End >= NextI->Start) {
245 I->TheStores.append(NextI->TheStores.begin(), NextI->TheStores.end());
246 if (NextI->End >
I->End)
262 assert(Start->getParent() == End->getParent() &&
"Must be in same block");
264 if (Start->getFunction()->doesNotThrow())
269 bool RequiresNoCaptureBeforeUnwind;
271 RequiresNoCaptureBeforeUnwind) &&
272 !RequiresNoCaptureBeforeUnwind)
282 I->eraseFromParent();
293 assert(Start->getBlock() == End->getBlock() &&
"Only local supported");
295 make_range(++Start->getIterator(), End->getIterator())) {
296 Instruction *
I = cast<MemoryUseOrDef>(MA).getMemoryInst();
298 auto *II = dyn_cast<IntrinsicInst>(
I);
299 if (II && II->getIntrinsicID() == Intrinsic::lifetime_start &&
300 SkippedLifetimeStart && !*SkippedLifetimeStart) {
301 *SkippedLifetimeStart =
I;
316 if (isa<MemoryUse>(End)) {
320 return Start->getBlock() != End->getBlock() ||
322 make_range(std::next(Start->getIterator()), End->getIterator()),
324 if (isa<MemoryUse>(&Acc))
326 Instruction *AccInst =
327 cast<MemoryUseOrDef>(&Acc)->getMemoryInst();
328 return isModSet(AA.getModRefInfo(AccInst, Loc));
334 End->getDefiningAccess(), Loc, AA);
348 if (
auto *SI = dyn_cast<StoreInst>(StartInst))
349 if (
DL.getTypeStoreSize(
SI->getOperand(0)->getType()).isScalable())
369 for (++BI; !BI->isTerminator(); ++BI) {
370 auto *CurrentAcc = cast_or_null<MemoryUseOrDef>(
373 MemInsertPoint = CurrentAcc;
374 if (
auto *CurrentDef = dyn_cast<MemoryDef>(CurrentAcc))
375 LastMemDef = CurrentDef;
380 if (
auto *CB = dyn_cast<CallBase>(BI)) {
381 if (CB->onlyAccessesInaccessibleMemory())
385 if (!isa<StoreInst>(BI) && !isa<MemSetInst>(BI)) {
389 if (BI->mayWriteToMemory() || BI->mayReadFromMemory())
394 if (
auto *NextStore = dyn_cast<StoreInst>(BI)) {
396 if (!NextStore->isSimple())
break;
398 Value *StoredVal = NextStore->getValueOperand();
406 if (
DL.getTypeStoreSize(StoredVal->
getType()).isScalable())
411 if (isa<UndefValue>(ByteVal) && StoredByte)
412 ByteVal = StoredByte;
413 if (ByteVal != StoredByte)
417 std::optional<int64_t>
Offset =
424 auto *MSI = cast<MemSetInst>(BI);
426 if (MSI->isVolatile() || ByteVal != MSI->getValue() ||
427 !isa<ConstantInt>(MSI->getLength()))
431 std::optional<int64_t>
Offset =
448 Ranges.addInst(0, StartInst);
458 for (
const MemsetRange &Range : Ranges) {
459 if (
Range.TheStores.size() == 1)
continue;
462 if (!
Range.isProfitableToUseMemset(
DL))
467 StartPtr =
Range.StartPtr;
476 dbgs() <<
"With: " << *AMemSet <<
'\n');
477 if (!
Range.TheStores.empty())
480 assert(LastMemDef && MemInsertPoint &&
481 "Both LastMemDef and MemInsertPoint need to be set");
485 AMemSet, LastMemDef, MemInsertPoint)
487 AMemSet, LastMemDef, MemInsertPoint));
490 MemInsertPoint = NewDef;
494 eraseInstruction(SI);
516 auto *
I = dyn_cast<Instruction>(
Arg);
517 if (
I &&
I->getParent() ==
SI->getParent()) {
519 if (
I ==
P)
return false;
524 if (!AddArg(
SI->getPointerOperand()))
538 for (
auto I = --
SI->getIterator(),
E =
P->getIterator();
I !=
E; --
I) {
548 bool NeedLift =
false;
570 else if (
const auto *Call = dyn_cast<CallBase>(
C)) {
576 }
else if (isa<LoadInst>(
C) || isa<StoreInst>(
C) || isa<VAArgInst>(
C)) {
582 MemLocs.push_back(
ML);
589 for (
Value *Op :
C->operands())
603 MemInsertPoint = cast<MemoryUseOrDef>(--MA->getIterator());
619 assert(MemInsertPoint &&
"Must have found insert point");
641 if (
T->isAggregateType() &&
643 (TLI->
has(LibFunc_memcpy) && TLI->
has(LibFunc_memmove)))) {
664 if (!moveUp(SI,
P, LI))
675 bool UseMemMove =
false;
685 SI->getPointerOperand(),
SI->getAlign(),
689 SI->getPointerOperand(),
SI->getAlign(),
691 M->copyMetadata(*SI, LLVMContext::MD_DIAssignID);
693 LLVM_DEBUG(
dbgs() <<
"Promoting " << *LI <<
" to " << *SI <<
" => "
699 MSSAU->
insertDef(cast<MemoryDef>(NewAccess),
true);
701 eraseInstruction(SI);
702 eraseInstruction(LI);
706 BBI =
M->getIterator();
715 auto GetCall = [&]() ->
CallInst * {
718 if (
auto *LoadClobber = dyn_cast<MemoryUseOrDef>(
720 return dyn_cast_or_null<CallInst>(LoadClobber->getMemoryInst());
724 bool Changed = performCallSlotOptzn(
725 LI, SI,
SI->getPointerOperand()->stripPointerCasts(),
727 DL.getTypeStoreSize(
SI->getOperand(0)->getType()),
728 std::min(
SI->getAlign(), LI->
getAlign()), BAA, GetCall);
730 eraseInstruction(SI);
731 eraseInstruction(LI);
740 if (!
SI->isSimple())
return false;
748 if (
SI->getMetadata(LLVMContext::MD_nontemporal))
753 Value *StoredVal =
SI->getValueOperand();
761 if (
auto *LI = dyn_cast<LoadInst>(StoredVal))
762 return processStoreOfLoad(SI, LI,
DL, BBI);
777 auto *
V =
SI->getOperand(0);
779 if (
Instruction *
I = tryMergingIntoMemset(SI,
SI->getPointerOperand(),
781 BBI =
I->getIterator();
788 auto *
T =
V->getType();
789 if (
T->isAggregateType()) {
792 auto *
M =
Builder.CreateMemSet(
SI->getPointerOperand(), ByteVal,
Size,
794 M->copyMetadata(*SI, LLVMContext::MD_DIAssignID);
796 LLVM_DEBUG(
dbgs() <<
"Promoting " << *SI <<
" to " << *M <<
"\n");
802 M, StoreDef->getDefiningAccess(), StoreDef);
803 MSSAU->
insertDef(cast<MemoryDef>(NewAccess),
false);
805 eraseInstruction(SI);
809 BBI =
M->getIterator();
823 BBI =
I->getIterator();
832bool MemCpyOptPass::performCallSlotOptzn(
Instruction *cpyLoad,
856 auto *srcAlloca = dyn_cast<AllocaInst>(cpySrc);
860 ConstantInt *srcArraySize = dyn_cast<ConstantInt>(srcAlloca->getArraySize());
865 uint64_t srcSize =
DL.getTypeAllocSize(srcAlloca->getAllocatedType()) *
868 if (cpySize < srcSize)
877 if (
F->isIntrinsic() &&
F->getIntrinsicID() == Intrinsic::lifetime_start)
881 if (
C->getParent() != cpyStore->
getParent()) {
895 LLVM_DEBUG(
dbgs() <<
"Call Slot: Dest pointer modified after call\n");
902 if (SkippedLifetimeStart) {
904 dyn_cast<Instruction>(SkippedLifetimeStart->
getOperand(1));
905 if (LifetimeArg && LifetimeArg->getParent() ==
C->getParent() &&
906 C->comesBefore(LifetimeArg))
915 LLVM_DEBUG(
dbgs() <<
"Call Slot: Dest pointer not dereferenceable\n");
934 LLVM_DEBUG(
dbgs() <<
"Call Slot: Dest may be visible through unwinding\n");
939 Align srcAlign = srcAlloca->getAlign();
940 bool isDestSufficientlyAligned = srcAlign <= cpyDestAlign;
943 if (!isDestSufficientlyAligned && !isa<AllocaInst>(cpyDest)) {
944 LLVM_DEBUG(
dbgs() <<
"Call Slot: Dest not sufficiently aligned\n");
953 while (!srcUseList.empty()) {
954 User *
U = srcUseList.pop_back_val();
956 if (isa<BitCastInst>(U) || isa<AddrSpaceCastInst>(U)) {
960 if (
const auto *
G = dyn_cast<GetElementPtrInst>(U)) {
961 if (!
G->hasAllZeroIndices())
967 if (
const auto *
IT = dyn_cast<IntrinsicInst>(U))
968 if (
IT->isLifetimeStartOrEnd())
971 if (U !=
C && U != cpyLoad)
977 bool SrcIsCaptured =
any_of(
C->args(), [&](
Use &U) {
978 return U->stripPointerCasts() == cpySrc &&
979 !C->doesNotCapture(C->getArgOperandNo(&U));
1000 make_range(++
C->getIterator(),
C->getParent()->end())) {
1002 if (
auto *II = dyn_cast<IntrinsicInst>(&
I)) {
1003 if (II->getIntrinsicID() == Intrinsic::lifetime_end &&
1004 II->getArgOperand(1)->stripPointerCasts() == srcAlloca &&
1005 cast<ConstantInt>(II->getArgOperand(0))->uge(srcSize))
1010 if (isa<ReturnInst>(&
I))
1030 auto *
GEP = dyn_cast<GetElementPtrInst>(cpyDest);
1031 if (
GEP &&
GEP->hasAllConstantIndices() &&
1055 for (
unsigned ArgI = 0; ArgI <
C->arg_size(); ++ArgI)
1056 if (
C->getArgOperand(ArgI)->stripPointerCasts() == cpySrc &&
1058 C->getArgOperand(ArgI)->getType()->getPointerAddressSpace())
1062 bool changedArgument =
false;
1063 for (
unsigned ArgI = 0; ArgI <
C->arg_size(); ++ArgI)
1064 if (
C->getArgOperand(ArgI)->stripPointerCasts() == cpySrc) {
1068 changedArgument =
true;
1069 if (
C->getArgOperand(ArgI)->getType() == Dest->
getType())
1070 C->setArgOperand(ArgI, Dest);
1073 Dest,
C->getArgOperand(ArgI)->getType(),
1077 if (!changedArgument)
1081 if (!isDestSufficientlyAligned) {
1082 assert(isa<AllocaInst>(cpyDest) &&
"Can only increase alloca alignment!");
1083 cast<AllocaInst>(cpyDest)->setAlignment(srcAlign);
1086 if (SkippedLifetimeStart) {
1095 unsigned KnownIDs[] = {LLVMContext::MD_tbaa, LLVMContext::MD_alias_scope,
1096 LLVMContext::MD_noalias,
1097 LLVMContext::MD_invariant_group,
1098 LLVMContext::MD_access_group};
1100 if (cpyLoad != cpyStore)
1109bool MemCpyOptPass::processMemCpyMemCpyDependence(
MemCpyInst *M,
1128 auto *MDepLen = dyn_cast<ConstantInt>(MDep->
getLength());
1129 auto *MLen = dyn_cast<ConstantInt>(
M->getLength());
1130 if (!MDepLen || !MLen || MDepLen->getZExtValue() < MLen->getZExtValue())
1154 bool UseMemMove =
false;
1159 LLVM_DEBUG(
dbgs() <<
"MemCpyOptPass: Forwarding memcpy->memcpy src:\n"
1160 << *MDep <<
'\n' << *M <<
'\n');
1167 NewM =
Builder.CreateMemMove(
M->getRawDest(),
M->getDestAlign(),
1169 M->getLength(),
M->isVolatile());
1170 else if (isa<MemCpyInlineInst>(M)) {
1174 NewM =
Builder.CreateMemCpyInline(
1178 NewM =
Builder.CreateMemCpy(
M->getRawDest(),
M->getDestAlign(),
1180 M->getLength(),
M->isVolatile());
1186 MSSAU->
insertDef(cast<MemoryDef>(NewAccess),
true);
1189 eraseInstruction(M);
1208bool MemCpyOptPass::processMemSetMemCpyDependence(
MemCpyInst *MemCpy,
1238 if (DestSize == SrcSize) {
1239 eraseInstruction(MemSet);
1250 if (
auto *SrcSizeC = dyn_cast<ConstantInt>(SrcSize))
1265 Value *SizeDiff =
Builder.CreateSub(DestSize, SrcSize);
1274 MemSet->
getOperand(1), MemsetLen, Alignment);
1277 "MemCpy must be a MemoryDef");
1284 NewMemSet, LastDef->getDefiningAccess(), LastDef);
1285 MSSAU->
insertDef(cast<MemoryDef>(NewAccess),
true);
1287 eraseInstruction(MemSet);
1298 if (
auto *II = dyn_cast_or_null<IntrinsicInst>(Def->getMemoryInst())) {
1299 if (II->getIntrinsicID() == Intrinsic::lifetime_start) {
1300 auto *LTSize = cast<ConstantInt>(II->getArgOperand(0));
1302 if (
auto *CSize = dyn_cast<ConstantInt>(
Size)) {
1304 LTSize->getZExtValue() >= CSize->getZExtValue())
1314 const DataLayout &
DL = Alloca->getModule()->getDataLayout();
1315 if (std::optional<TypeSize> AllocaSize =
1316 Alloca->getAllocationSize(
DL))
1317 if (*AllocaSize == LTSize->getValue())
1339bool MemCpyOptPass::performMemCpyToMemSetOptzn(
MemCpyInst *MemCpy,
1350 if (MemSetSize != CopySize) {
1355 auto *CMemSetSize = dyn_cast<ConstantInt>(MemSetSize);
1360 auto *CCopySize = dyn_cast<ConstantInt>(CopySize);
1363 if (CCopySize->getZExtValue() > CMemSetSize->getZExtValue()) {
1369 bool CanReduceSize =
false;
1373 if (
auto *MD = dyn_cast<MemoryDef>(Clobber))
1375 CanReduceSize =
true;
1379 CopySize = MemSetSize;
1390 MSSAU->
insertDef(cast<MemoryDef>(NewAccess),
true);
1402 if (
M->isVolatile())
return false;
1405 if (
M->getSource() ==
M->getDest()) {
1407 eraseInstruction(M);
1412 if (
auto *GV = dyn_cast<GlobalVariable>(
M->getSource()))
1413 if (GV->isConstant() && GV->hasDefinitiveInitializer())
1415 M->getModule()->getDataLayout())) {
1418 M->getRawDest(), ByteVal,
M->getLength(),
M->getDestAlign(),
false);
1423 MSSAU->
insertDef(cast<MemoryDef>(NewAccess),
true);
1425 eraseInstruction(M);
1442 if (
auto *MD = dyn_cast<MemoryDef>(DestClobber))
1443 if (
auto *MDep = dyn_cast_or_null<MemSetInst>(MD->getMemoryInst()))
1444 if (DestClobber->
getBlock() ==
M->getParent())
1445 if (processMemSetMemCpyDependence(M, MDep, BAA))
1458 if (
auto *MD = dyn_cast<MemoryDef>(SrcClobber)) {
1460 if (
auto *CopySize = dyn_cast<ConstantInt>(
M->getLength())) {
1461 if (
auto *
C = dyn_cast<CallInst>(
MI)) {
1462 if (performCallSlotOptzn(M, M,
M->getDest(),
M->getSource(),
1464 M->getDestAlign().valueOrOne(), BAA,
1467 <<
" call: " << *
C <<
"\n"
1468 <<
" memcpy: " << *M <<
"\n");
1469 eraseInstruction(M);
1475 if (
auto *MDep = dyn_cast<MemCpyInst>(
MI))
1476 return processMemCpyMemCpyDependence(M, MDep, BAA);
1477 if (
auto *MDep = dyn_cast<MemSetInst>(
MI)) {
1478 if (performMemCpyToMemSetOptzn(M, MDep, BAA)) {
1480 eraseInstruction(M);
1489 eraseInstruction(M);
1500bool MemCpyOptPass::processMemMove(
MemMoveInst *M) {
1505 LLVM_DEBUG(
dbgs() <<
"MemCpyOptPass: Optimizing memmove -> memcpy: " << *M
1509 Type *ArgTys[3] = {
M->getRawDest()->getType(),
1510 M->getRawSource()->getType(),
1511 M->getLength()->getType() };
1513 Intrinsic::memcpy, ArgTys));
1523bool MemCpyOptPass::processByValArgument(
CallBase &CB,
unsigned ArgNo) {
1528 TypeSize ByValSize =
DL.getTypeAllocSize(ByValTy);
1537 if (
auto *MD = dyn_cast<MemoryDef>(Clobber))
1538 MDep = dyn_cast_or_null<MemCpyInst>(MD->getMemoryInst());
1548 auto *C1 = dyn_cast<ConstantInt>(MDep->
getLength());
1556 if (!ByValAlign)
return false;
1561 if ((!MemDepAlign || *MemDepAlign < *ByValAlign) &&
1587 TmpCast = TmpBitCast;
1590 LLVM_DEBUG(
dbgs() <<
"MemCpyOptPass: Forwarding memcpy to byval:\n"
1591 <<
" " << *MDep <<
"\n"
1592 <<
" " << CB <<
"\n");
1601bool MemCpyOptPass::iterateOnFunction(
Function &
F) {
1602 bool MadeChange =
false;
1617 bool RepeatInstruction =
false;
1619 if (
auto *SI = dyn_cast<StoreInst>(
I))
1620 MadeChange |= processStore(SI, BI);
1621 else if (
auto *M = dyn_cast<MemSetInst>(
I))
1622 RepeatInstruction = processMemSet(M, BI);
1623 else if (
auto *M = dyn_cast<MemCpyInst>(
I))
1624 RepeatInstruction = processMemCpy(M, BI);
1625 else if (
auto *M = dyn_cast<MemMoveInst>(
I))
1626 RepeatInstruction = processMemMove(M);
1627 else if (
auto *CB = dyn_cast<CallBase>(
I)) {
1628 for (
unsigned i = 0, e = CB->
arg_size(); i != e; ++i)
1630 MadeChange |= processByValArgument(*CB, i);
1634 if (RepeatInstruction) {
1635 if (BI != BB.begin())
1652 bool MadeChange =
runImpl(
F, &TLI, AA, AC, DT, &MSSA->getMSSA());
1665 bool MadeChange =
false;
1675 if (!iterateOnFunction(
F))
MachineBasicBlock MachineBasicBlock::iterator DebugLoc DL
amdgpu Simplify well known AMD library false FunctionCallee Value * Arg
static cl::opt< ITMode > IT(cl::desc("IT block support"), cl::Hidden, cl::init(DefaultIT), cl::values(clEnumValN(DefaultIT, "arm-default-it", "Generate any type of IT block"), clEnumValN(RestrictedIT, "arm-restrict-it", "Disallow complex IT blocks")))
static GCRegistry::Add< CoreCLRGC > E("coreclr", "CoreCLR-compatible GC")
This file contains the declarations for the subclasses of Constant, which represent the different fla...
This file defines the DenseSet and SmallDenseSet classes.
This is the interface for a simple mod/ref and alias analysis over globals.
static bool mayBeVisibleThroughUnwinding(Value *V, Instruction *Start, Instruction *End)
static bool accessedBetween(BatchAAResults &AA, MemoryLocation Loc, const MemoryUseOrDef *Start, const MemoryUseOrDef *End, Instruction **SkippedLifetimeStart=nullptr)
static bool hasUndefContents(MemorySSA *MSSA, BatchAAResults &AA, Value *V, MemoryDef *Def, Value *Size)
Determine whether the instruction has undefined content for the given Size, either because it was fre...
static cl::opt< bool > EnableMemCpyOptWithoutLibcalls("enable-memcpyopt-without-libcalls", cl::Hidden, cl::desc("Enable memcpyopt even when libcalls are disabled"))
static bool writtenBetween(MemorySSA *MSSA, BatchAAResults &AA, MemoryLocation Loc, const MemoryUseOrDef *Start, const MemoryUseOrDef *End)
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.
This header defines various interfaces for pass management in LLVM.
assert(ImpDefSCC.getReg()==AMDGPU::SCC &&ImpDefSCC.isDef())
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.
ModRefInfo getModRefInfo(const Instruction *I, const std::optional< MemoryLocation > &OptLoc)
Check whether or not an instruction may read or write the optionally specified memory location.
Class for arbitrary precision integers.
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.
A function analysis which provides an AssumptionCache.
A cache of @llvm.assume calls within a function.
LLVM Basic Block Representation.
InstListType::iterator iterator
Instruction iterators...
This class is a wrapper over an AAResults, and it is intended to be used only when there are no IR ch...
bool isMustAlias(const MemoryLocation &LocA, const MemoryLocation &LocB)
ModRefInfo getModRefInfo(const Instruction *I, const std::optional< MemoryLocation > &OptLoc)
ModRefInfo callCapturesBefore(const Instruction *I, const MemoryLocation &MemLoc, DominatorTree *DT)
This class represents a no-op cast from one type to another.
Represents analyses that only rely on functions' control flow.
Base class for all callable instructions (InvokeInst and CallInst) Holds everything related to callin...
bool isByValArgument(unsigned ArgNo) const
Determine whether this argument is passed by value.
MaybeAlign getParamAlign(unsigned ArgNo) const
Extract the alignment for a call or parameter (0=unknown).
Type * getParamByValType(unsigned ArgNo) const
Extract the byval type for a call or parameter.
Value * getArgOperand(unsigned i) const
void setArgOperand(unsigned i, Value *v)
unsigned arg_size() const
Function * getCaller()
Helper to get the caller (the parent function).
This class represents a function call, abstracting a target machine's calling convention.
static CastInst * CreatePointerCast(Value *S, Type *Ty, const Twine &Name, BasicBlock *InsertAtEnd)
Create a BitCast AddrSpaceCast, or a PtrToInt cast instruction.
This is the shared class of boolean and integer constants.
uint64_t getZExtValue() const
Return the constant as a 64-bit unsigned integer value after it has been zero extended as appropriate...
static Constant * getNullValue(Type *Ty)
Constructor to create a '0' constant of arbitrary type.
A parsed version of the target data layout string in and methods for querying it.
Implements a dense probed hash-table based set.
Analysis pass which computes a DominatorTree.
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.
Module * getParent()
Get the module that this global value is contained inside of...
This provides a uniform API for creating instructions and inserting them into a basic block: either a...
void mergeDIAssignID(ArrayRef< const Instruction * > SourceInstructions)
Merge the DIAssignID metadata from this instruction and those attached to instructions in SourceInstr...
const DebugLoc & getDebugLoc() const
Return the debug location for this node as a DebugLoc.
const Module * getModule() const
Return the module owning the function this instruction belongs to or nullptr it the function does not...
const BasicBlock * getParent() const
void setDebugLoc(DebugLoc Loc)
Set the debug location information for this instruction.
void copyMetadata(const Instruction &SrcInst, ArrayRef< unsigned > WL=ArrayRef< unsigned >())
Copy metadata from SrcInst to this instruction.
void moveBefore(Instruction *MovePos)
Unlink this instruction from its current basic block and insert it into the basic block that MovePos ...
An instruction for reading from memory.
Value * getPointerOperand()
Align getAlign() const
Return the alignment of the access that is being performed.
static LocationSize precise(uint64_t Value)
This class wraps the llvm.memcpy intrinsic.
PreservedAnalyses run(Function &F, FunctionAnalysisManager &AM)
bool runImpl(Function &F, TargetLibraryInfo *TLI, AAResults *AA, AssumptionCache *AC, DominatorTree *DT, MemorySSA *MSSA)
Value * getLength() const
Value * getRawDest() const
Value * getDest() const
This is just like getRawDest, but it strips off any cast instructions (including addrspacecast) that ...
MaybeAlign getDestAlign() const
This class wraps the llvm.memmove intrinsic.
This class wraps the llvm.memset and llvm.memset.inline intrinsics.
Value * getRawSource() const
Return the arguments to the instruction.
MaybeAlign getSourceAlign() const
Value * getSource() const
This is just like getRawSource, but it strips off any cast instructions that feed it,...
BasicBlock * getBlock() const
Represents a read-write access to memory, whether it is a must-alias, or a may-alias.
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.
static MemoryLocation getForSource(const MemTransferInst *MTI)
Return a location representing the source of a memory transfer.
static MemoryLocation getForDest(const MemIntrinsic *MI)
Return a location representing the destination of a memory set or transfer.
An analysis that produces MemorySSA for a function.
MemorySSA * getMemorySSA() const
Get handle on MemorySSA.
MemoryUseOrDef * createMemoryAccessBefore(Instruction *I, MemoryAccess *Definition, MemoryUseOrDef *InsertPt)
Create a MemoryAccess in MemorySSA before or after an existing MemoryAccess.
void insertDef(MemoryDef *Def, bool RenameUses=false)
Insert a definition into the MemorySSA IR.
void moveAfter(MemoryUseOrDef *What, MemoryUseOrDef *Where)
void removeMemoryAccess(MemoryAccess *, bool OptimizePhis=false)
Remove a MemoryAccess from MemorySSA, including updating all definitions and uses.
MemoryUseOrDef * createMemoryAccessAfter(Instruction *I, MemoryAccess *Definition, MemoryAccess *InsertPt)
void moveBefore(MemoryUseOrDef *What, MemoryUseOrDef *Where)
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.
bool dominates(const MemoryAccess *A, const MemoryAccess *B) const
Given two memory accesses in potentially different blocks, determine whether MemoryAccess A dominates...
void verifyMemorySSA(VerificationLevel=VerificationLevel::Fast) const
Verify that MemorySSA is self consistent (IE definitions dominate all uses, uses appear in the right ...
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.
Class that has the common methods + fields of memory uses/defs.
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.
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.
typename SuperClass::const_iterator const_iterator
typename SuperClass::iterator iterator
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.
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.
static constexpr TypeSize getFixed(ScalarTy ExactSize)
The instances of the Type class are immutable: once they are created, they are never changed.
unsigned getIntegerBitWidth() const
unsigned getPointerAddressSpace() const
Get the address space of this pointer or pointer vector type.
Type * getScalarType() const
If this is a vector type, return the element type, otherwise return 'this'.
A Use represents the edge between a Value definition and its users.
Value * getOperand(unsigned i) const
LLVM Value Representation.
Type * getType() const
All values are typed, get the type of this value.
bool hasOneUse() const
Return true if there is exactly one use of this value.
const Value * stripPointerCasts() const
Strip off pointer casts, all-zero GEPs and address space casts.
StringRef getName() const
Return a constant reference to the value's name.
constexpr ScalarTy getFixedValue() const
constexpr bool isScalable() const
Returns whether the quantity is scaled by a runtime quantity (vscale).
static constexpr bool isKnownGE(const FixedOrScalableQuantity &LHS, const FixedOrScalableQuantity &RHS)
reverse_self_iterator getReverseIterator()
self_iterator getIterator()
This provides a very simple, boring adaptor for a begin and end iterator into a range type.
constexpr char Args[]
Key for Kernel::Metadata::mArgs.
@ C
The default llvm calling convention, compatible with C.
Function * getDeclaration(Module *M, ID id, ArrayRef< Type * > Tys=std::nullopt)
Create or insert an LLVM Function declaration for an intrinsic, and return it.
const_iterator begin(StringRef path, Style style=Style::native)
Get begin iterator over path.
const_iterator end(StringRef path)
Get end iterator over path.
This is an optimization pass for GlobalISel generic memory operations.
bool isDereferenceableAndAlignedPointer(const Value *V, Type *Ty, Align Alignment, const DataLayout &DL, const Instruction *CtxI=nullptr, AssumptionCache *AC=nullptr, const DominatorTree *DT=nullptr, const TargetLibraryInfo *TLI=nullptr)
Returns true if V is always a dereferenceable pointer with alignment greater or equal than requested.
auto partition_point(R &&Range, Predicate P)
Binary search for the first iterator in a range where a predicate is false.
iterator_range< T > make_range(T x, T y)
Convenience function for iterating over sub-ranges.
bool PointerMayBeCapturedBefore(const Value *V, bool ReturnCaptures, bool StoreCaptures, const Instruction *I, const DominatorTree *DT, bool IncludeI=false, unsigned MaxUsesToExplore=0, const LoopInfo *LI=nullptr)
PointerMayBeCapturedBefore - Return true if this pointer value may be captured by the enclosing funct...
void append_range(Container &C, Range &&R)
Wrapper function to append a range to a container.
const Value * getUnderlyingObject(const Value *V, unsigned MaxLookup=6)
This method strips off any GEP address adjustments and pointer casts from the specified value,...
bool any_of(R &&range, UnaryPredicate P)
Provide wrappers to std::any_of which take ranges instead of having to pass begin/end explicitly.
auto reverse(ContainerTy &&C)
Align getOrEnforceKnownAlignment(Value *V, MaybeAlign PrefAlign, const DataLayout &DL, const Instruction *CxtI=nullptr, AssumptionCache *AC=nullptr, const DominatorTree *DT=nullptr)
Try to ensure that the alignment of V is at least PrefAlign bytes.
bool isModSet(const ModRefInfo MRI)
raw_ostream & dbgs()
dbgs() - This returns a reference to a raw_ostream for debugging messages.
bool isModOrRefSet(const ModRefInfo MRI)
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...
void combineMetadata(Instruction *K, const Instruction *J, ArrayRef< unsigned > KnownIDs, bool DoesKMove)
Combine the metadata of two instructions so that K can replace J.
ModRefInfo
Flags indicating whether a memory access modifies or references memory.
bool VerifyMemorySSA
Enables verification of MemorySSA.
bool isIdentifiedFunctionLocal(const Value *V)
Return true if V is umabigously identified at the function-level.
bool isGuaranteedToTransferExecutionToSuccessor(const Instruction *I)
Return true if this function can prove that the instruction I will always transfer execution to one o...
decltype(auto) cast(const From &Val)
cast<X> - Return the argument parameter cast to the specified type.
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...
Align commonAlignment(Align A, uint64_t Offset)
Returns the alignment that satisfies both alignments.
std::optional< int64_t > isPointerOffset(const Value *Ptr1, const Value *Ptr2, const DataLayout &DL)
If Ptr1 is provably equal to Ptr2 plus a constant offset, return that offset.
This struct is a compact representation of a valid (non-zero power of two) alignment.
This struct is a compact representation of a valid (power of two) or undefined (0) alignment.
Align valueOrOne() const
For convenience, returns a valid alignment or 1 if undefined.