104#define DEBUG_TYPE "loop-idiom"
106STATISTIC(NumMemSet,
"Number of memset's formed from loop stores");
107STATISTIC(NumMemCpy,
"Number of memcpy's formed from loop load+stores");
108STATISTIC(NumMemMove,
"Number of memmove's formed from loop load+stores");
110 NumShiftUntilBitTest,
111 "Number of uncountable loops recognized as 'shift until bitttest' idiom");
113 "Number of uncountable loops recognized as 'shift until zero' idiom");
118 cl::desc(
"Options to disable Loop Idiom Recognize Pass."),
125 cl::desc(
"Proceed with loop idiom recognize pass, but do "
126 "not convert loop(s) to memset."),
133 cl::desc(
"Proceed with loop idiom recognize pass, but do "
134 "not convert loop(s) to memcpy."),
139 "use-lir-code-size-heurs",
140 cl::desc(
"Use loop idiom recognition code size heuristics when compiling"
146class LoopIdiomRecognize {
147 Loop *CurLoop =
nullptr;
156 bool ApplyCodeSizeHeuristics;
157 std::unique_ptr<MemorySSAUpdater> MSSAU;
166 : AA(AA), DT(DT), LI(LI), SE(SE), TLI(TLI),
TTI(
TTI),
DL(
DL), ORE(ORE) {
168 MSSAU = std::make_unique<MemorySSAUpdater>(MSSA);
171 bool runOnLoop(
Loop *L);
177 StoreListMap StoreRefsForMemset;
178 StoreListMap StoreRefsForMemsetPattern;
179 StoreList StoreRefsForMemcpy;
181 bool HasMemsetPattern;
185 enum LegalStoreKind {
190 UnorderedAtomicMemcpy,
198 bool runOnCountableLoop();
203 LegalStoreKind isLegalStore(
StoreInst *SI);
204 enum class ForMemset {
No,
Yes };
208 template <
typename MemInst>
209 bool processLoopMemIntrinsic(
211 bool (LoopIdiomRecognize::*Processor)(MemInst *,
const SCEV *),
212 const SCEV *BECount);
216 bool processLoopStridedStore(
Value *DestPtr,
const SCEV *StoreSizeSCEV,
221 bool IsNegStride,
bool IsLoopMemset =
false);
222 bool processLoopStoreOfLoopLoad(
StoreInst *SI,
const SCEV *BECount);
223 bool processLoopStoreOfLoopLoad(
Value *DestPtr,
Value *SourcePtr,
229 const SCEV *BECount);
230 bool avoidLIRForMultiBlockLoop(
bool IsMemset =
false,
231 bool IsLoopMemset =
false);
237 bool runOnNoncountableLoop();
239 bool recognizePopcount();
242 bool recognizeAndInsertFFS();
247 bool IsCntPhiUsedOutsideLoop);
249 bool recognizeShiftUntilBitTest();
250 bool recognizeShiftUntilZero();
262 const auto *
DL = &L.getHeader()->getModule()->getDataLayout();
269 LoopIdiomRecognize LIR(&AR.
AA, &AR.
DT, &AR.
LI, &AR.
SE, &AR.
TLI, &AR.
TTI,
271 if (!LIR.runOnLoop(&L))
282 I->eraseFromParent();
291bool LoopIdiomRecognize::runOnLoop(
Loop *L) {
295 if (!
L->getLoopPreheader())
300 if (
Name ==
"memset" ||
Name ==
"memcpy")
304 ApplyCodeSizeHeuristics =
307 HasMemset = TLI->
has(LibFunc_memset);
308 HasMemsetPattern = TLI->
has(LibFunc_memset_pattern16);
309 HasMemcpy = TLI->
has(LibFunc_memcpy);
311 if (HasMemset || HasMemsetPattern || HasMemcpy)
313 return runOnCountableLoop();
315 return runOnNoncountableLoop();
318bool LoopIdiomRecognize::runOnCountableLoop() {
320 assert(!isa<SCEVCouldNotCompute>(BECount) &&
321 "runOnCountableLoop() called on a loop without a predictable"
322 "backedge-taken count");
326 if (
const SCEVConstant *BECst = dyn_cast<SCEVConstant>(BECount))
327 if (BECst->getAPInt() == 0)
345 bool MadeChange =
false;
353 MadeChange |= runOnLoopBlock(BB, BECount, ExitBlocks);
377 if (!
C || isa<ConstantExpr>(
C))
386 if (
DL->isBigEndian())
402 unsigned ArraySize = 16 /
Size;
407LoopIdiomRecognize::LegalStoreKind
408LoopIdiomRecognize::isLegalStore(
StoreInst *SI) {
410 if (
SI->isVolatile())
411 return LegalStoreKind::None;
413 if (!
SI->isUnordered())
414 return LegalStoreKind::None;
417 if (
SI->getMetadata(LLVMContext::MD_nontemporal))
418 return LegalStoreKind::None;
420 Value *StoredVal =
SI->getValueOperand();
421 Value *StorePtr =
SI->getPointerOperand();
426 return LegalStoreKind::None;
434 return LegalStoreKind::None;
440 dyn_cast<SCEVAddRecExpr>(SE->
getSCEV(StorePtr));
442 return LegalStoreKind::None;
445 if (!isa<SCEVConstant>(StoreEv->
getOperand(1)))
446 return LegalStoreKind::None;
457 bool UnorderedAtomic =
SI->isUnordered() && !
SI->isSimple();
466 return LegalStoreKind::Memset;
473 return LegalStoreKind::MemsetPattern;
481 unsigned StoreSize =
DL->getTypeStoreSize(
SI->getValueOperand()->getType());
482 if (StoreSize != Stride && StoreSize != -Stride)
483 return LegalStoreKind::None;
486 LoadInst *LI = dyn_cast<LoadInst>(
SI->getValueOperand());
490 return LegalStoreKind::None;
493 return LegalStoreKind::None;
501 return LegalStoreKind::None;
505 return LegalStoreKind::None;
508 UnorderedAtomic = UnorderedAtomic || LI->
isAtomic();
509 return UnorderedAtomic ? LegalStoreKind::UnorderedAtomicMemcpy
510 : LegalStoreKind::Memcpy;
513 return LegalStoreKind::None;
516void LoopIdiomRecognize::collectStores(
BasicBlock *BB) {
517 StoreRefsForMemset.clear();
518 StoreRefsForMemsetPattern.clear();
519 StoreRefsForMemcpy.clear();
526 switch (isLegalStore(SI)) {
527 case LegalStoreKind::None:
530 case LegalStoreKind::Memset: {
533 StoreRefsForMemset[
Ptr].push_back(SI);
535 case LegalStoreKind::MemsetPattern: {
538 StoreRefsForMemsetPattern[
Ptr].push_back(SI);
540 case LegalStoreKind::Memcpy:
541 case LegalStoreKind::UnorderedAtomicMemcpy:
542 StoreRefsForMemcpy.push_back(SI);
545 assert(
false &&
"unhandled return value");
554bool LoopIdiomRecognize::runOnLoopBlock(
564 bool MadeChange =
false;
571 for (
auto &SL : StoreRefsForMemset)
572 MadeChange |= processLoopStores(SL.second, BECount, ForMemset::Yes);
574 for (
auto &SL : StoreRefsForMemsetPattern)
575 MadeChange |= processLoopStores(SL.second, BECount, ForMemset::No);
578 for (
auto &SI : StoreRefsForMemcpy)
579 MadeChange |= processLoopStoreOfLoopLoad(SI, BECount);
581 MadeChange |= processLoopMemIntrinsic<MemCpyInst>(
582 BB, &LoopIdiomRecognize::processLoopMemCpy, BECount);
583 MadeChange |= processLoopMemIntrinsic<MemSetInst>(
584 BB, &LoopIdiomRecognize::processLoopMemSet, BECount);
591 const SCEV *BECount, ForMemset For) {
599 for (
unsigned i = 0, e = SL.
size(); i < e; ++i) {
600 assert(SL[i]->
isSimple() &&
"Expected only non-volatile stores.");
602 Value *FirstStoredVal = SL[i]->getValueOperand();
603 Value *FirstStorePtr = SL[i]->getPointerOperand();
605 cast<SCEVAddRecExpr>(SE->
getSCEV(FirstStorePtr));
607 unsigned FirstStoreSize =
DL->getTypeStoreSize(SL[i]->getValueOperand()->
getType());
610 if (FirstStride == FirstStoreSize || -FirstStride == FirstStoreSize) {
615 Value *FirstSplatValue =
nullptr;
616 Constant *FirstPatternValue =
nullptr;
618 if (For == ForMemset::Yes)
623 assert((FirstSplatValue || FirstPatternValue) &&
624 "Expected either splat value or pattern value.");
632 for (j = i + 1;
j <
e; ++
j)
634 for (j = i;
j > 0; --
j)
637 for (
auto &k : IndexQueue) {
638 assert(SL[k]->
isSimple() &&
"Expected only non-volatile stores.");
639 Value *SecondStorePtr = SL[
k]->getPointerOperand();
641 cast<SCEVAddRecExpr>(SE->
getSCEV(SecondStorePtr));
644 if (FirstStride != SecondStride)
647 Value *SecondStoredVal = SL[
k]->getValueOperand();
648 Value *SecondSplatValue =
nullptr;
649 Constant *SecondPatternValue =
nullptr;
651 if (For == ForMemset::Yes)
656 assert((SecondSplatValue || SecondPatternValue) &&
657 "Expected either splat value or pattern value.");
660 if (For == ForMemset::Yes) {
661 if (isa<UndefValue>(FirstSplatValue))
662 FirstSplatValue = SecondSplatValue;
663 if (FirstSplatValue != SecondSplatValue)
666 if (isa<UndefValue>(FirstPatternValue))
667 FirstPatternValue = SecondPatternValue;
668 if (FirstPatternValue != SecondPatternValue)
673 ConsecutiveChain[SL[i]] = SL[
k];
682 bool Changed =
false;
693 unsigned StoreSize = 0;
696 while (Tails.
count(
I) || Heads.count(
I)) {
697 if (TransformedStores.
count(
I))
701 StoreSize +=
DL->getTypeStoreSize(
I->getValueOperand()->getType());
703 I = ConsecutiveChain[
I];
713 if (StoreSize != Stride && StoreSize != -Stride)
716 bool IsNegStride = StoreSize == -Stride;
720 if (processLoopStridedStore(StorePtr, StoreSizeSCEV,
722 HeadStore, AdjacentStores, StoreEv, BECount,
724 TransformedStores.
insert(AdjacentStores.
begin(), AdjacentStores.
end());
734template <
typename MemInst>
735bool LoopIdiomRecognize::processLoopMemIntrinsic(
737 bool (LoopIdiomRecognize::*Processor)(MemInst *,
const SCEV *),
738 const SCEV *BECount) {
739 bool MadeChange =
false;
743 if (MemInst *
MI = dyn_cast<MemInst>(Inst)) {
745 if (!(this->*Processor)(
MI, BECount))
759bool LoopIdiomRecognize::processLoopMemCpy(
MemCpyInst *MCI,
760 const SCEV *BECount) {
771 if (!Dest || !Source)
786 if ((SizeInBytes >> 32) != 0)
792 dyn_cast<SCEVConstant>(StoreEv->
getOperand(1));
794 dyn_cast<SCEVConstant>(LoadEv->
getOperand(1));
795 if (!ConstStoreStride || !ConstLoadStride)
804 if (SizeInBytes != StoreStrideValue && SizeInBytes != -StoreStrideValue) {
807 <<
ore::NV(
"Inst",
"memcpy") <<
" in "
809 <<
" function will not be hoisted: "
810 <<
ore::NV(
"Reason",
"memcpy size is not equal to stride");
815 int64_t StoreStrideInt = StoreStrideValue.
getSExtValue();
818 if (StoreStrideInt != LoadStrideInt)
821 return processLoopStoreOfLoopLoad(
828bool LoopIdiomRecognize::processLoopMemSet(
MemSetInst *MSI,
829 const SCEV *BECount) {
844 if (!Ev || Ev->
getLoop() != CurLoop)
853 if (!PointerStrideSCEV || !MemsetSizeSCEV)
856 bool IsNegStride =
false;
857 const bool IsConstantSize = isa<ConstantInt>(MSI->
getLength());
859 if (IsConstantSize) {
870 if (SizeInBytes != Stride && SizeInBytes != -Stride)
873 IsNegStride = SizeInBytes == -Stride;
881 if (
Pointer->getType()->getPointerAddressSpace() != 0) {
894 const SCEV *PositiveStrideSCEV =
897 LLVM_DEBUG(
dbgs() <<
" MemsetSizeSCEV: " << *MemsetSizeSCEV <<
"\n"
898 <<
" PositiveStrideSCEV: " << *PositiveStrideSCEV
901 if (PositiveStrideSCEV != MemsetSizeSCEV) {
904 const SCEV *FoldedPositiveStride =
906 const SCEV *FoldedMemsetSize =
910 <<
" FoldedMemsetSize: " << *FoldedMemsetSize <<
"\n"
911 <<
" FoldedPositiveStride: " << *FoldedPositiveStride
914 if (FoldedPositiveStride != FoldedMemsetSize) {
931 BECount, IsNegStride,
true);
939 const SCEV *BECount,
const SCEV *StoreSizeSCEV,
949 const SCEVConstant *BECst = dyn_cast<SCEVConstant>(BECount);
950 const SCEVConstant *ConstSize = dyn_cast<SCEVConstant>(StoreSizeSCEV);
951 if (BECst && ConstSize)
973 Type *IntPtr,
const SCEV *StoreSizeSCEV,
976 if (!StoreSizeSCEV->
isOne()) {
991 const SCEV *StoreSizeSCEV,
Loop *CurLoop,
993 const SCEV *TripCountSCEV =
1002bool LoopIdiomRecognize::processLoopStridedStore(
1006 const SCEV *BECount,
bool IsNegStride,
bool IsLoopMemset) {
1014 assert((SplatValue || PatternValue) &&
1015 "Expected either splat value or pattern value.");
1026 Type *DestInt8PtrTy =
Builder.getInt8PtrTy(DestAS);
1029 bool Changed =
false;
1037 if (!Expander.isSafeToExpand(Start))
1046 Expander.expandCodeFor(Start, DestInt8PtrTy, Preheader->
getTerminator());
1058 StoreSizeSCEV, *AA, Stores))
1061 if (avoidLIRForMultiBlockLoop(
true, IsLoopMemset))
1066 const SCEV *NumBytesS =
1067 getNumBytes(BECount, IntIdxTy, StoreSizeSCEV, CurLoop,
DL, SE);
1071 if (!Expander.isSafeToExpand(NumBytesS))
1075 Expander.expandCodeFor(NumBytesS, IntIdxTy, Preheader->
getTerminator());
1081 AATags = AATags.
merge(
Store->getAAMetadata());
1082 if (
auto CI = dyn_cast<ConstantInt>(NumBytes))
1083 AATags = AATags.
extendTo(CI->getZExtValue());
1087 NewCall =
Builder.CreateMemSet(
1088 BasePtr, SplatValue, NumBytes,
MaybeAlign(StoreAlignment),
1092 Type *Int8PtrTy = DestInt8PtrTy;
1094 StringRef FuncName =
"memset_pattern16";
1096 Builder.getVoidTy(), Int8PtrTy, Int8PtrTy, IntIdxTy);
1103 PatternValue,
".memset_pattern");
1107 NewCall =
Builder.CreateCall(MSP, {
BasePtr, PatternPtr, NumBytes});
1114 MemoryAccess *NewMemAcc = MSSAU->createMemoryAccessInBB(
1116 MSSAU->insertDef(cast<MemoryDef>(NewMemAcc),
true);
1120 <<
" from store to: " << *Ev <<
" at: " << *TheStore
1126 R <<
"Transformed loop-strided store in "
1128 <<
" function into a call to "
1131 if (!Stores.empty())
1133 for (
auto *
I : Stores) {
1134 R <<
ore::NV(
"FromBlock",
I->getParent()->getName())
1142 for (
auto *
I : Stores) {
1144 MSSAU->removeMemoryAccess(
I,
true);
1148 MSSAU->getMemorySSA()->verifyMemorySSA();
1150 ExpCleaner.markResultUsed();
1157bool LoopIdiomRecognize::processLoopStoreOfLoopLoad(
StoreInst *SI,
1158 const SCEV *BECount) {
1159 assert(
SI->isUnordered() &&
"Expected only non-volatile non-ordered stores.");
1161 Value *StorePtr =
SI->getPointerOperand();
1163 unsigned StoreSize =
DL->getTypeStoreSize(
SI->getValueOperand()->getType());
1166 LoadInst *LI = cast<LoadInst>(
SI->getValueOperand());
1176 return processLoopStoreOfLoopLoad(StorePtr, LoadPtr, StoreSizeSCEV,
1178 StoreEv, LoadEv, BECount);
1182class MemmoveVerifier {
1184 explicit MemmoveVerifier(
const Value &LoadBasePtr,
const Value &StoreBasePtr,
1187 LoadBasePtr.stripPointerCasts(), LoadOff,
DL)),
1189 StoreBasePtr.stripPointerCasts(), StoreOff,
DL)),
1190 IsSameObject(BP1 == BP2) {}
1192 bool loadAndStoreMayFormMemmove(
unsigned StoreSize,
bool IsNegStride,
1194 bool IsMemCpy)
const {
1198 if ((!IsNegStride && LoadOff <= StoreOff) ||
1199 (IsNegStride && LoadOff >= StoreOff))
1205 DL.getTypeSizeInBits(TheLoad.
getType()).getFixedValue() / 8;
1206 if (BP1 != BP2 || LoadSize != int64_t(StoreSize))
1208 if ((!IsNegStride && LoadOff < StoreOff + int64_t(StoreSize)) ||
1209 (IsNegStride && LoadOff + LoadSize > StoreOff))
1217 int64_t LoadOff = 0;
1218 int64_t StoreOff = 0;
1223 const bool IsSameObject;
1227bool LoopIdiomRecognize::processLoopStoreOfLoopLoad(
1236 if (isa<MemCpyInlineInst>(TheStore))
1248 bool Changed =
false;
1251 Type *IntIdxTy =
Builder.getIntNTy(
DL->getIndexSizeInBits(StrAS));
1254 const SCEVConstant *ConstStoreSize = dyn_cast<SCEVConstant>(StoreSizeSCEV);
1257 assert(ConstStoreSize &&
"store size is expected to be a constant");
1260 bool IsNegStride = StoreSize == -Stride;
1273 Value *StoreBasePtr = Expander.expandCodeFor(
1286 IgnoredInsts.
insert(TheStore);
1288 bool IsMemCpy = isa<MemCpyInst>(TheStore);
1289 const StringRef InstRemark = IsMemCpy ?
"memcpy" :
"load and store";
1291 bool LoopAccessStore =
1293 StoreSizeSCEV, *AA, IgnoredInsts);
1294 if (LoopAccessStore) {
1300 IgnoredInsts.
insert(TheLoad);
1302 BECount, StoreSizeSCEV, *AA, IgnoredInsts)) {
1306 <<
ore::NV(
"Inst", InstRemark) <<
" in "
1308 <<
" function will not be hoisted: "
1309 <<
ore::NV(
"Reason",
"The loop may access store location");
1313 IgnoredInsts.
erase(TheLoad);
1326 Value *LoadBasePtr = Expander.expandCodeFor(
1331 MemmoveVerifier
Verifier(*LoadBasePtr, *StoreBasePtr, *
DL);
1332 if (IsMemCpy && !
Verifier.IsSameObject)
1333 IgnoredInsts.
erase(TheStore);
1335 StoreSizeSCEV, *AA, IgnoredInsts)) {
1338 <<
ore::NV(
"Inst", InstRemark) <<
" in "
1340 <<
" function will not be hoisted: "
1341 <<
ore::NV(
"Reason",
"The loop may access load location");
1346 bool UseMemMove = IsMemCpy ?
Verifier.IsSameObject : LoopAccessStore;
1348 if (!
Verifier.loadAndStoreMayFormMemmove(StoreSize, IsNegStride, *TheLoad,
1352 if (avoidLIRForMultiBlockLoop())
1357 const SCEV *NumBytesS =
1358 getNumBytes(BECount, IntIdxTy, StoreSizeSCEV, CurLoop,
DL, SE);
1361 Expander.expandCodeFor(NumBytesS, IntIdxTy, Preheader->
getTerminator());
1365 AATags = AATags.
merge(StoreAATags);
1366 if (
auto CI = dyn_cast<ConstantInt>(NumBytes))
1367 AATags = AATags.
extendTo(CI->getZExtValue());
1377 NewCall =
Builder.CreateMemMove(
1378 StoreBasePtr, StoreAlign, LoadBasePtr, LoadAlign, NumBytes,
1382 Builder.CreateMemCpy(StoreBasePtr, StoreAlign, LoadBasePtr, LoadAlign,
1383 NumBytes,
false, AATags.
TBAA,
1391 assert((StoreAlign && LoadAlign) &&
1392 "Expect unordered load/store to have align.");
1393 if (*StoreAlign < StoreSize || *LoadAlign < StoreSize)
1406 NewCall =
Builder.CreateElementUnorderedAtomicMemCpy(
1407 StoreBasePtr, *StoreAlign, LoadBasePtr, *LoadAlign, NumBytes, StoreSize,
1413 MemoryAccess *NewMemAcc = MSSAU->createMemoryAccessInBB(
1415 MSSAU->insertDef(cast<MemoryDef>(NewMemAcc),
true);
1419 <<
" from load ptr=" << *LoadEv <<
" at: " << *TheLoad
1421 <<
" from store ptr=" << *StoreEv <<
" at: " << *TheStore
1427 <<
"Formed a call to "
1429 <<
"() intrinsic from " <<
ore::NV(
"Inst", InstRemark)
1440 MSSAU->removeMemoryAccess(TheStore,
true);
1443 MSSAU->getMemorySSA()->verifyMemorySSA();
1448 ExpCleaner.markResultUsed();
1455bool LoopIdiomRecognize::avoidLIRForMultiBlockLoop(
bool IsMemset,
1456 bool IsLoopMemset) {
1457 if (ApplyCodeSizeHeuristics && CurLoop->
getNumBlocks() > 1) {
1458 if (CurLoop->
isOutermost() && (!IsMemset || !IsLoopMemset)) {
1460 <<
" : LIR " << (IsMemset ?
"Memset" :
"Memcpy")
1461 <<
" avoided: multi-block top-level loop\n");
1469bool LoopIdiomRecognize::runOnNoncountableLoop() {
1472 <<
"] Noncountable Loop %"
1475 return recognizePopcount() || recognizeAndInsertFFS() ||
1476 recognizeShiftUntilBitTest() || recognizeShiftUntilZero();
1486 bool JmpOnZero =
false) {
1495 if (!CmpZero || !CmpZero->
isZero())
1506 return Cond->getOperand(0);
1515 auto *PhiX = dyn_cast<PHINode>(VarX);
1516 if (PhiX && PhiX->getParent() == LoopEntry &&
1517 (PhiX->getOperand(0) == DefX || PhiX->
getOperand(1) == DefX))
1553 Value *VarX1, *VarX0;
1556 DefX2 = CountInst =
nullptr;
1557 VarX1 = VarX0 =
nullptr;
1558 PhiX = CountPhi =
nullptr;
1564 dyn_cast<BranchInst>(LoopEntry->
getTerminator()), LoopEntry))
1565 DefX2 = dyn_cast<Instruction>(
T);
1572 if (!DefX2 || DefX2->
getOpcode() != Instruction::And)
1577 if ((SubOneOp = dyn_cast<BinaryOperator>(DefX2->
getOperand(0))))
1581 SubOneOp = dyn_cast<BinaryOperator>(DefX2->
getOperand(1));
1583 if (!SubOneOp || SubOneOp->
getOperand(0) != VarX1)
1589 (SubOneOp->
getOpcode() == Instruction::Add &&
1602 CountInst =
nullptr;
1605 if (Inst.
getOpcode() != Instruction::Add)
1609 if (!Inc || !Inc->
isOne())
1617 bool LiveOutLoop =
false;
1619 if ((cast<Instruction>(U))->
getParent() != LoopEntry) {
1639 auto *PreCondBr = dyn_cast<BranchInst>(PreCondBB->
getTerminator());
1644 CntInst = CountInst;
1684 Value *VarX =
nullptr;
1693 dyn_cast<BranchInst>(LoopEntry->
getTerminator()), LoopEntry))
1694 DefX = dyn_cast<Instruction>(
T);
1699 if (!DefX || !DefX->
isShift())
1701 IntrinID = DefX->
getOpcode() == Instruction::Shl ? Intrinsic::cttz :
1704 if (!Shft || !Shft->
isOne())
1729 if (Inst.
getOpcode() != Instruction::Add)
1753bool LoopIdiomRecognize::recognizeAndInsertFFS() {
1765 size_t IdiomCanonicalSize = 6;
1768 CntInst, CntPhi, DefX))
1771 bool IsCntPhiUsedOutsideLoop =
false;
1773 if (!CurLoop->
contains(cast<Instruction>(U))) {
1774 IsCntPhiUsedOutsideLoop =
true;
1777 bool IsCntInstUsedOutsideLoop =
false;
1779 if (!CurLoop->
contains(cast<Instruction>(U))) {
1780 IsCntInstUsedOutsideLoop =
true;
1785 if (IsCntInstUsedOutsideLoop && IsCntPhiUsedOutsideLoop)
1791 bool ZeroCheck =
false;
1800 if (!IsCntPhiUsedOutsideLoop) {
1804 auto *PreCondBI = dyn_cast<BranchInst>(PreCondBB->getTerminator());
1829 std::distance(InstWithoutDebugIt.begin(), InstWithoutDebugIt.end());
1834 if (HeaderSize != IdiomCanonicalSize &&
1838 transformLoopToCountable(IntrinID, PH, CntInst, CntPhi, InitX, DefX,
1840 IsCntPhiUsedOutsideLoop);
1848bool LoopIdiomRecognize::recognizePopcount() {
1862 if (LoopBody->
size() >= 20) {
1872 if (!EntryBI || EntryBI->isConditional())
1880 auto *PreCondBI = dyn_cast<BranchInst>(PreCondBB->getTerminator());
1881 if (!PreCondBI || PreCondBI->isUnconditional())
1890 transformLoopToPopcount(PreCondBB, CntInst, CntPhi, Val);
1896 Value *Ops[] = {Val};
1952void LoopIdiomRecognize::transformLoopToCountable(
1955 bool ZeroCheck,
bool IsCntPhiUsedOutsideLoop) {
1969 if (IsCntPhiUsedOutsideLoop) {
1970 if (DefX->
getOpcode() == Instruction::AShr)
1971 InitXNext =
Builder.CreateAShr(InitX, 1);
1972 else if (DefX->
getOpcode() == Instruction::LShr)
1973 InitXNext =
Builder.CreateLShr(InitX, 1);
1974 else if (DefX->
getOpcode() == Instruction::Shl)
1975 InitXNext =
Builder.CreateShl(InitX, 1);
1985 Value *NewCount = Count;
1986 if (IsCntPhiUsedOutsideLoop)
1989 NewCount =
Builder.CreateZExtOrTrunc(NewCount, CntInst->
getType());
1992 if (cast<ConstantInt>(CntInst->
getOperand(1))->isOne()) {
1995 ConstantInt *InitConst = dyn_cast<ConstantInt>(CntInitVal);
1996 if (!InitConst || !InitConst->
isZero())
1997 NewCount =
Builder.CreateAdd(NewCount, CntInitVal);
2001 NewCount =
Builder.CreateSub(CntInitVal, NewCount);
2014 ICmpInst *LbCond = cast<ICmpInst>(LbBr->getCondition());
2018 Builder.SetInsertPoint(LbCond);
2033 if (IsCntPhiUsedOutsideLoop)
2043void LoopIdiomRecognize::transformLoopToPopcount(
BasicBlock *PreCondBB,
2047 auto *PreCondBr = cast<BranchInst>(PreCondBB->
getTerminator());
2056 Value *PopCnt, *PopCntZext, *NewCount, *TripCnt;
2059 NewCount = PopCntZext =
2060 Builder.CreateZExtOrTrunc(PopCnt, cast<IntegerType>(CntPhi->
getType()));
2062 if (NewCount != PopCnt)
2063 (cast<Instruction>(NewCount))->setDebugLoc(
DL);
2070 ConstantInt *InitConst = dyn_cast<ConstantInt>(CntInitVal);
2071 if (!InitConst || !InitConst->
isZero()) {
2072 NewCount =
Builder.CreateAdd(NewCount, CntInitVal);
2073 (cast<Instruction>(NewCount))->setDebugLoc(
DL);
2082 ICmpInst *PreCond = cast<ICmpInst>(PreCondBr->getCondition());
2084 Value *Opnd0 = PopCntZext;
2089 ICmpInst *NewPreCond = cast<ICmpInst>(
2091 PreCondBr->setCondition(NewPreCond);
2119 ICmpInst *LbCond = cast<ICmpInst>(LbBr->getCondition());
2124 Builder.SetInsertPoint(LbCond);
2127 "tcdec",
false,
true));
2155 : SubPattern(SP), L(L) {}
2157 template <
typename ITy>
bool match(ITy *V) {
2158 return L->isLoopInvariant(V) && SubPattern.match(V);
2163template <
typename Ty>
2194 " Performing shift-until-bittest idiom detection.\n");
2204 assert(LoopPreheaderBB &&
"There is always a loop preheader.");
2206 using namespace PatternMatch;
2211 Value *CmpLHS, *CmpRHS;
2222 auto MatchVariableBitMask = [&]() {
2231 auto MatchConstantBitMask = [&]() {
2237 auto MatchDecomposableConstantBitMask = [&]() {
2245 if (!MatchVariableBitMask() && !MatchConstantBitMask() &&
2246 !MatchDecomposableConstantBitMask()) {
2252 auto *CurrXPN = dyn_cast<PHINode>(CurrX);
2253 if (!CurrXPN || CurrXPN->getParent() != LoopHeaderBB) {
2258 BaseX = CurrXPN->getIncomingValueForBlock(LoopPreheaderBB);
2260 dyn_cast<Instruction>(CurrXPN->getIncomingValueForBlock(LoopHeaderBB));
2263 "Expected BaseX to be avaliable in the preheader!");
2274 "Should only get equality predicates here.");
2284 if (TrueBB != LoopHeaderBB) {
2343bool LoopIdiomRecognize::recognizeShiftUntilBitTest() {
2344 bool MadeChange =
false;
2346 Value *
X, *BitMask, *BitPos, *XCurr;
2351 " shift-until-bittest idiom detection failed.\n");
2361 assert(LoopPreheaderBB &&
"There is always a loop preheader.");
2364 assert(SuccessorBB &&
"There is only a single successor.");
2370 Type *Ty =
X->getType();
2384 " Intrinsic is too costly, not beneficial\n");
2399 BitPos->
getName() +
".lowbitmask");
2401 Builder.CreateOr(LowBitMask, BitMask, BitPos->
getName() +
".mask");
2402 Value *XMasked =
Builder.CreateAnd(
X, Mask,
X->getName() +
".masked");
2404 IntrID, Ty, {XMasked,
Builder.getTrue()},
2405 nullptr, XMasked->
getName() +
".numleadingzeros");
2408 XMasked->
getName() +
".numactivebits",
true,
2410 Value *XMaskedLeadingOnePos =
2412 XMasked->
getName() +
".leadingonepos",
false,
2416 BitPos, XMaskedLeadingOnePos, CurLoop->
getName() +
".backedgetakencount",
2420 Value *LoopTripCount =
2422 CurLoop->
getName() +
".tripcount",
true,
2431 if (
auto *
I = dyn_cast<Instruction>(NewX))
2432 I->copyIRFlags(XNext,
true);
2444 NewXNext =
Builder.CreateShl(
X, LoopTripCount);
2453 if (
auto *
I = dyn_cast<Instruction>(NewXNext))
2454 I->copyIRFlags(XNext,
true);
2473 true, Bitwidth != 2);
2476 auto *IVCheck =
Builder.CreateICmpEQ(IVNext, LoopTripCount,
2477 CurLoop->
getName() +
".ivcheck");
2478 Builder.CreateCondBr(IVCheck, SuccessorBB, LoopHeaderBB);
2483 IV->addIncoming(IVNext, LoopHeaderBB);
2494 ++NumShiftUntilBitTest;
2530 const SCEV *&ExtraOffsetExpr,
2531 bool &InvertedCond) {
2533 " Performing shift-until-zero idiom detection.\n");
2546 assert(LoopPreheaderBB &&
"There is always a loop preheader.");
2548 using namespace PatternMatch;
2557 !
match(ValShiftedIsZero,
2571 IntrinID = ValShifted->
getOpcode() == Instruction::Shl ? Intrinsic::cttz
2580 else if (
match(NBits,
2584 ExtraOffsetExpr = SE->
getSCEV(ExtraOffset);
2591 auto *IVPN = dyn_cast<PHINode>(
IV);
2592 if (!IVPN || IVPN->getParent() != LoopHeaderBB) {
2597 Start = IVPN->getIncomingValueForBlock(LoopPreheaderBB);
2598 IVNext = dyn_cast<Instruction>(IVPN->getIncomingValueForBlock(LoopHeaderBB));
2608 "Should only get equality predicates here.");
2619 if (FalseBB != LoopHeaderBB) {
2630 if (ValShifted->
getOpcode() == Instruction::AShr &&
2694bool LoopIdiomRecognize::recognizeShiftUntilZero() {
2695 bool MadeChange =
false;
2701 const SCEV *ExtraOffsetExpr;
2704 Start, Val, ExtraOffsetExpr, InvertedCond)) {
2706 " shift-until-zero idiom detection failed.\n");
2716 assert(LoopPreheaderBB &&
"There is always a loop preheader.");
2719 assert(SuccessorBB &&
"There is only a single successor.");
2722 Builder.SetCurrentDebugLocation(
IV->getDebugLoc());
2738 " Intrinsic is too costly, not beneficial\n");
2745 bool OffsetIsZero =
false;
2746 if (
auto *ExtraOffsetExprC = dyn_cast<SCEVConstant>(ExtraOffsetExpr))
2747 OffsetIsZero = ExtraOffsetExprC->isZero();
2752 IntrID, Ty, {Val,
Builder.getFalse()},
2753 nullptr, Val->
getName() +
".numleadingzeros");
2756 Val->
getName() +
".numactivebits",
true,
2760 Expander.setInsertPoint(&*
Builder.GetInsertPoint());
2761 Value *ExtraOffset = Expander.expandCodeFor(ExtraOffsetExpr);
2764 ValNumActiveBits, ExtraOffset, ValNumActiveBits->
getName() +
".offset",
2765 OffsetIsZero,
true);
2766 Value *IVFinal =
Builder.CreateIntrinsic(Intrinsic::smax, {Ty},
2767 {ValNumActiveBitsOffset, Start},
2768 nullptr,
"iv.final");
2770 auto *LoopBackedgeTakenCount = cast<Instruction>(
Builder.CreateSub(
2771 IVFinal, Start, CurLoop->
getName() +
".backedgetakencount",
2772 OffsetIsZero,
true));
2776 Value *LoopTripCount =
2778 CurLoop->
getName() +
".tripcount",
true,
2784 IV->replaceUsesOutsideBlock(IVFinal, LoopHeaderBB);
2790 auto *CIV =
Builder.CreatePHI(Ty, 2, CurLoop->
getName() +
".iv");
2796 true, Bitwidth != 2);
2799 auto *CIVCheck =
Builder.CreateICmpEQ(CIVNext, LoopTripCount,
2800 CurLoop->
getName() +
".ivcheck");
2801 auto *NewIVCheck = CIVCheck;
2803 NewIVCheck =
Builder.CreateNot(CIVCheck);
2804 NewIVCheck->takeName(ValShiftedIsZero);
2808 auto *IVDePHId =
Builder.CreateAdd(CIV, Start,
"",
false,
2810 IVDePHId->takeName(
IV);
2814 Builder.CreateCondBr(CIVCheck, SuccessorBB, LoopHeaderBB);
2819 CIV->addIncoming(CIVNext, LoopHeaderBB);
2827 IV->replaceAllUsesWith(IVDePHId);
2828 IV->eraseFromParent();
2837 ++NumShiftUntilZero;
MachineBasicBlock MachineBasicBlock::iterator DebugLoc DL
This file implements a class to represent arbitrary precision integral constant values and operations...
static const Function * getParent(const Value *V)
SmallVector< MachineOperand, 4 > Cond
static GCRegistry::Add< OcamlGC > B("ocaml", "ocaml 3.10-compatible GC")
static GCRegistry::Add< CoreCLRGC > E("coreclr", "CoreCLR-compatible GC")
This file contains the declarations for the subclasses of Constant, which represent the different fla...
static cl::opt< TargetTransformInfo::TargetCostKind > CostKind("cost-kind", cl::desc("Target cost kind"), cl::init(TargetTransformInfo::TCK_RecipThroughput), cl::values(clEnumValN(TargetTransformInfo::TCK_RecipThroughput, "throughput", "Reciprocal throughput"), clEnumValN(TargetTransformInfo::TCK_Latency, "latency", "Instruction latency"), clEnumValN(TargetTransformInfo::TCK_CodeSize, "code-size", "Code size"), clEnumValN(TargetTransformInfo::TCK_SizeAndLatency, "size-latency", "Code size and latency")))
This file defines the DenseMap class.
static GCMetadataPrinterRegistry::Add< ErlangGCPrinter > X("erlang", "erlang-compatible garbage collector")
static bool mayLoopAccessLocation(Value *Ptr, ModRefInfo Access, Loop *L, const SCEV *BECount, unsigned StoreSize, AliasAnalysis &AA, SmallPtrSetImpl< Instruction * > &Ignored)
mayLoopAccessLocation - Return true if the specified loop might access the specified pointer location...
This file defines an InstructionCost class that is used when calculating the cost of an instruction,...
static Value * matchCondition(BranchInst *BI, BasicBlock *LoopEntry, bool JmpOnZero=false)
Check if the given conditional branch is based on the comparison between a variable and zero,...
static PHINode * getRecurrenceVar(Value *VarX, Instruction *DefX, BasicBlock *LoopEntry)
static cl::opt< bool, true > DisableLIRPMemset("disable-" DEBUG_TYPE "-memset", cl::desc("Proceed with loop idiom recognize pass, but do " "not convert loop(s) to memset."), cl::location(DisableLIRP::Memset), cl::init(false), cl::ReallyHidden)
static cl::opt< bool > UseLIRCodeSizeHeurs("use-lir-code-size-heurs", cl::desc("Use loop idiom recognition code size heuristics when compiling" "with -Os/-Oz"), cl::init(true), cl::Hidden)
static CallInst * createFFSIntrinsic(IRBuilder<> &IRBuilder, Value *Val, const DebugLoc &DL, bool ZeroCheck, Intrinsic::ID IID)
static bool detectShiftUntilBitTestIdiom(Loop *CurLoop, Value *&BaseX, Value *&BitMask, Value *&BitPos, Value *&CurrX, Instruction *&NextX)
Return true if the idiom is detected in the loop.
static bool detectPopcountIdiom(Loop *CurLoop, BasicBlock *PreCondBB, Instruction *&CntInst, PHINode *&CntPhi, Value *&Var)
Return true iff the idiom is detected in the loop.
static Constant * getMemSetPatternValue(Value *V, const DataLayout *DL)
getMemSetPatternValue - If a strided store of the specified value is safe to turn into a memset_patte...
static cl::opt< bool, true > DisableLIRPMemcpy("disable-" DEBUG_TYPE "-memcpy", cl::desc("Proceed with loop idiom recognize pass, but do " "not convert loop(s) to memcpy."), cl::location(DisableLIRP::Memcpy), cl::init(false), cl::ReallyHidden)
static CallInst * createPopcntIntrinsic(IRBuilder<> &IRBuilder, Value *Val, const DebugLoc &DL)
static const SCEV * getNumBytes(const SCEV *BECount, Type *IntPtr, const SCEV *StoreSizeSCEV, Loop *CurLoop, const DataLayout *DL, ScalarEvolution *SE)
Compute the number of bytes as a SCEV from the backedge taken count.
static bool detectShiftUntilZeroIdiom(Loop *CurLoop, const DataLayout &DL, Intrinsic::ID &IntrinID, Value *&InitX, Instruction *&CntInst, PHINode *&CntPhi, Instruction *&DefX)
Return true if the idiom is detected in the loop.
static const SCEV * getStartForNegStride(const SCEV *Start, const SCEV *BECount, Type *IntPtr, const SCEV *StoreSizeSCEV, ScalarEvolution *SE)
static APInt getStoreStride(const SCEVAddRecExpr *StoreEv)
match_LoopInvariant< Ty > m_LoopInvariant(const Ty &M, const Loop *L)
Matches if the value is loop-invariant.
static cl::opt< bool, true > DisableLIRPAll("disable-" DEBUG_TYPE "-all", cl::desc("Options to disable Loop Idiom Recognize Pass."), cl::location(DisableLIRP::All), cl::init(false), cl::ReallyHidden)
static void deleteDeadInstruction(Instruction *I)
static DebugLoc getDebugLoc(MachineBasicBlock::instr_iterator FirstMI, MachineBasicBlock::instr_iterator LastMI)
Return the first found DebugLoc that has a DILocation, given a range of instructions.
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.
assert(ImpDefSCC.getReg()==AMDGPU::SCC &&ImpDefSCC.isDef())
static bool isSimple(Instruction *I)
verify safepoint Safepoint IR Verifier
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)
static SymbolRef::Type getType(const Symbol *Sym)
static const uint32_t IV[8]
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.
unsigned getBitWidth() const
Return the number of bits in the APInt.
int64_t getSExtValue() const
Get sign extended value.
A container for analyses that lazily runs them and caches their results.
static ArrayType * get(Type *ElementType, uint64_t NumElements)
This static method is the primary way to construct an ArrayType.
LLVM Basic Block Representation.
iterator begin()
Instruction iterator methods.
iterator_range< filter_iterator< BasicBlock::const_iterator, std::function< bool(const Instruction &)> > > instructionsWithoutDebug(bool SkipPseudoOp=true) const
Return a const iterator range over the instructions in the block, skipping any debug instructions.
const Instruction * getFirstNonPHI() const
Returns a pointer to the first instruction in this block that is not a PHINode instruction.
const Instruction & front() const
const BasicBlock * getSinglePredecessor() const
Return the predecessor of this block if it has a single predecessor block.
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...
BinaryOps getOpcode() const
Conditional or Unconditional Branch instruction.
bool isConditional() const
BasicBlock * getSuccessor(unsigned i) const
Value * getCondition() const
Function * getCalledFunction() const
Returns the function called, or null if this is an indirect function invocation or the function signa...
This class represents a function call, abstracting a target machine's calling convention.
void setPredicate(Predicate P)
Set the predicate for this instruction to the specified value.
Predicate
This enumeration lists the possible predicates for CmpInst subclasses.
@ ICMP_SLE
signed less or equal
@ ICMP_UGT
unsigned greater than
Predicate getInversePredicate() const
For example, EQ -> NE, UGT -> ULE, SLT -> SGE, OEQ -> UNE, UGT -> OLE, OLT -> UGE,...
Predicate getPredicate() const
Return the predicate for this instruction.
static Constant * get(ArrayType *T, ArrayRef< Constant * > V)
static Constant * getBitCast(Constant *C, Type *Ty, bool OnlyIfReduced=false)
static Constant * getExactLogBase2(Constant *C)
If C is a scalar/fixed width vector of known powers of 2, then this function returns a new scalar/fix...
This is the shared class of boolean and integer constants.
bool isMinusOne() const
This function will return true iff every bit in this constant is set to true.
bool isOne() const
This is just a convenience method to make client code smaller for a common case.
bool isZero() const
This is just a convenience method to make client code smaller for a common code.
static Constant * get(Type *Ty, uint64_t V, bool IsSigned=false)
If Ty is a vector type, return a Constant with a splat of the given value.
uint64_t getZExtValue() const
Return the constant as a 64-bit unsigned integer value after it has been zero extended as appropriate...
static ConstantInt * getBool(LLVMContext &Context, bool V)
This is an important base class in LLVM.
static Constant * getAllOnesValue(Type *Ty)
A parsed version of the target data layout string in and methods for querying it.
Concrete subclass of DominatorTreeBase that is used to compute a normal dominator tree.
bool dominates(const BasicBlock *BB, const Use &U) const
Return true if the (end of the) basic block BB dominates the use U.
A handy container for a FunctionType+Callee-pointer pair, which can be passed around as a single enti...
void setAlignment(Align Align)
Sets the alignment attribute of the GlobalObject.
void setUnnamedAddr(UnnamedAddr Val)
Module * getParent()
Get the module that this global value is contained inside of...
@ PrivateLinkage
Like Internal, but omit from symbol table.
This instruction compares its operands according to the predicate given to the constructor.
bool isEquality() const
Return true if this predicate is either EQ or NE.
ConstantInt * getInt1(bool V)
Get a constant value representing either true or false.
BasicBlock * GetInsertBlock() const
CallInst * CreateCall(FunctionType *FTy, Value *Callee, ArrayRef< Value * > Args=std::nullopt, const Twine &Name="", MDNode *FPMathTag=nullptr)
This provides a uniform API for creating instructions and inserting them into a basic block: either a...
bool hasNoUnsignedWrap() const LLVM_READONLY
Determine whether the no unsigned wrap flag is set.
bool hasNoSignedWrap() const LLVM_READONLY
Determine whether the no signed wrap flag is set.
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...
bool isAtomic() const LLVM_READONLY
Return true if this instruction has an AtomicOrdering of unordered or higher.
const BasicBlock * getParent() const
const Function * getFunction() const
Return the function this instruction belongs to.
AAMDNodes getAAMetadata() const
Returns the AA metadata for this instruction.
unsigned getOpcode() const
Returns a member of one of the enums like Instruction::Add.
SymbolTableList< Instruction >::iterator eraseFromParent()
This method unlinks 'this' from the containing basic block and deletes it.
void setDebugLoc(DebugLoc Loc)
Set the debug location information for this instruction.
This class provides an interface for updating the loop pass manager based on mutations to the loop ne...
An instruction for reading from memory.
Value * getPointerOperand()
bool isVolatile() const
Return true if this is a load from a volatile memory location.
Align getAlign() const
Return the alignment of the access that is being performed.
static LocationSize precise(uint64_t Value)
static constexpr LocationSize afterPointer()
Any location after the base pointer (but still within the underlying object).
bool contains(const LoopT *L) const
Return true if the specified loop is contained within in this loop.
bool isOutermost() const
Return true if the loop does not have a parent (natural) loop.
unsigned getNumBlocks() const
Get the number of blocks in this loop in constant time.
unsigned getNumBackEdges() const
Calculate the number of back edges to the loop header.
BlockT * getHeader() const
BlockT * getExitBlock() const
If getExitBlocks would return exactly one block, return that block.
BlockT * getLoopPreheader() const
If there is a preheader for this loop, return it.
ArrayRef< BlockT * > getBlocks() const
Get a list of the basic blocks which make up this loop.
void getUniqueExitBlocks(SmallVectorImpl< BlockT * > &ExitBlocks) const
Return all unique successor blocks of this loop.
block_iterator block_begin() const
PreservedAnalyses run(Loop &L, LoopAnalysisManager &AM, LoopStandardAnalysisResults &AR, LPMUpdater &U)
LoopT * getLoopFor(const BlockT *BB) const
Return the inner most loop that BB lives in.
Represents a single loop in the control flow graph.
bool isLoopInvariant(const Value *V) const
Return true if the specified value is loop invariant.
StringRef getName() const
This class implements a map that also provides access to all stored values in a deterministic order.
This class wraps the llvm.memcpy intrinsic.
Value * getLength() 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.memset and llvm.memset.inline intrinsics.
MaybeAlign getSourceAlign() const
Value * getSource() const
This is just like getRawSource, but it strips off any cast instructions that feed it,...
Representation for a specific memory location.
An analysis that produces MemorySSA for a function.
Encapsulates MemorySSA, including all data associated with memory accesses.
A Module instance is used to store all the information related to an LLVM module.
void addIncoming(Value *V, BasicBlock *BB)
Add an incoming value to the end of the PHI list.
static PHINode * Create(Type *Ty, unsigned NumReservedValues, const Twine &NameStr="", Instruction *InsertBefore=nullptr)
Constructors - NumReservedValues is a hint for the number of incoming edges that this phi node will h...
Value * getIncomingValueForBlock(const BasicBlock *BB) const
static PoisonValue * get(Type *T)
Static factory methods - Return an 'poison' object of the specified type.
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.
This node represents a polynomial recurrence on the trip count of the specified loop.
const SCEV * getStart() const
bool isAffine() const
Return true if this represents an expression A + B*x where A and B are loop invariant values.
const Loop * getLoop() const
This class represents a constant integer value.
ConstantInt * getValue() const
const APInt & getAPInt() const
Helper to remove instructions inserted during SCEV expansion, unless they are marked as used.
This class uses information about analyze scalars to rewrite expressions in canonical form.
const SCEV * getOperand(unsigned i) const
This class represents an analyzed expression in the program.
bool isOne() const
Return true if the expression is a constant one.
bool isNonConstantNegative() const
Return true if the specified scev is negated, but not a constant.
The main scalar evolution driver.
bool isKnownNonNegative(const SCEV *S)
Test if the given expression is known to be non-negative.
const SCEV * getNegativeSCEV(const SCEV *V, SCEV::NoWrapFlags Flags=SCEV::FlagAnyWrap)
Return the SCEV object corresponding to -V.
const SCEV * getBackedgeTakenCount(const Loop *L, ExitCountKind Kind=Exact)
If the specified loop has a predictable backedge-taken count, return it, otherwise return a SCEVCould...
const SCEV * getZero(Type *Ty)
Return a SCEV for the constant 0 of a specific type.
const SCEV * getConstant(ConstantInt *V)
const SCEV * getSCEV(Value *V)
Return a SCEV expression for the full generality of the specified expression.
const SCEV * getTripCountFromExitCount(const SCEV *ExitCount)
A version of getTripCountFromExitCount below which always picks an evaluation type which can not resu...
void forgetLoop(const Loop *L)
This method should be called by the client when it has changed a loop in a way that may effect Scalar...
bool isLoopInvariant(const SCEV *S, const Loop *L)
Return true if the value of the given SCEV is unchanging in the specified loop.
const SCEV * getMinusSCEV(const SCEV *LHS, const SCEV *RHS, SCEV::NoWrapFlags Flags=SCEV::FlagAnyWrap, unsigned Depth=0)
Return LHS-RHS.
bool hasLoopInvariantBackedgeTakenCount(const Loop *L)
Return true if the specified loop has an analyzable loop-invariant backedge-taken count.
const SCEV * applyLoopGuards(const SCEV *Expr, const Loop *L)
Try to apply information from loop guards for L to Expr.
const SCEV * getMulExpr(SmallVectorImpl< const SCEV * > &Ops, SCEV::NoWrapFlags Flags=SCEV::FlagAnyWrap, unsigned Depth=0)
Get a canonical multiply expression, or something simpler if possible.
const SCEV * getTruncateOrZeroExtend(const SCEV *V, Type *Ty, unsigned Depth=0)
Return a SCEV corresponding to a conversion of the input value to the specified type.
A vector that has set insertion semantics.
size_type count(const key_type &key) const
Count the number of elements of a given key in the SetVector.
bool insert(const value_type &X)
Insert a new element into the SetVector.
Simple and conservative implementation of LoopSafetyInfo that can give false-positive answers to its ...
void computeLoopSafetyInfo(const Loop *CurLoop) override
Computes safety information for a loop checks loop body & header for the possibility of may throw exc...
bool anyBlockMayThrow() const override
Returns true iff any block of the loop for which this info is contains an instruction that may throw ...
A templated base class for SmallPtrSet which provides the typesafe interface that is common across al...
bool erase(PtrType Ptr)
erase - If the set contains the specified pointer, remove it and return true, otherwise return false.
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.
This class consists of common code factored out of the SmallVector class to reduce code duplication b...
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()
Value * getPointerOperand()
StringRef - Represent a constant reference to a string, i.e.
Provides information about what library functions are available for the current target.
bool has(LibFunc F) const
Tests whether a library function is available.
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.
unsigned getScalarSizeInBits() const LLVM_READONLY
If this is a vector type, return the getPrimitiveSizeInBits value for the element type.
Type * getScalarType() const
If this is a vector type, return the element type, otherwise return 'this'.
static UndefValue * get(Type *T)
Static factory methods - Return an 'undef' object of the specified type.
void setOperand(unsigned i, Value *Val)
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.
void replaceAllUsesWith(Value *V)
Change all uses of this to point to a new Value.
iterator_range< user_iterator > users()
void replaceUsesOutsideBlock(Value *V, BasicBlock *BB)
replaceUsesOutsideBlock - Go through the uses list for this definition and make each use point to "V"...
LLVMContext & getContext() const
All values hold a context through their type.
StringRef getName() const
Return a constant reference to the value's name.
void takeName(Value *V)
Transfer the name from V to this value.
Value handle that is nullable, but tries to track the Value.
constexpr ScalarTy getFixedValue() const
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.
constexpr char Args[]
Key for Kernel::Metadata::mArgs.
constexpr char Attrs[]
Key for Kernel::Metadata::mAttrs.
constexpr std::underlying_type_t< E > Mask()
Get a bitmask with 1s in all places up to the high-order bit of E's largest value.
@ 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.
BinaryOp_match< LHS, RHS, Instruction::And > m_And(const LHS &L, const RHS &R)
BinaryOp_match< LHS, RHS, Instruction::Add > m_Add(const LHS &L, const RHS &R)
cst_pred_ty< is_power2 > m_Power2()
Match an integer or vector power-of-2.
BinaryOp_match< LHS, RHS, Instruction::And, true > m_c_And(const LHS &L, const RHS &R)
Matches an And with LHS and RHS in either order.
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.
cst_pred_ty< is_one > m_One()
Match an integer 1 or a vector with all elements equal to 1.
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)
brc_match< Cond_t, bind_ty< BasicBlock >, bind_ty< BasicBlock > > m_Br(const Cond_t &C, BasicBlock *&T, BasicBlock *&F)
BinaryOp_match< LHS, RHS, Instruction::Add, true > m_c_Add(const LHS &L, const RHS &R)
Matches a Add with LHS and RHS in either order.
class_match< Value > m_Value()
Match an arbitrary value and ignore it.
BinOpPred_match< LHS, RHS, is_shift_op > m_Shift(const LHS &L, const RHS &R)
Matches shift operations.
BinaryOp_match< LHS, RHS, Instruction::Shl > m_Shl(const LHS &L, const RHS &R)
class_match< BasicBlock > m_BasicBlock()
Match an arbitrary basic block value and ignore it.
is_zero m_Zero()
Match any null constant or a vector with all elements equal to 0.
BinaryOp_match< LHS, RHS, Instruction::Sub > m_Sub(const LHS &L, const RHS &R)
cst_pred_ty< icmp_pred_with_threshold > m_SpecificInt_ICMP(ICmpInst::Predicate Predicate, const APInt &Threshold)
Match an integer or vector with every element comparing 'pred' (eg/ne/...) to Threshold.
initializer< Ty > init(const Ty &Val)
LocationClass< Ty > location(Ty &L)
DiagnosticInfoOptimizationBase::setExtraArgs setExtraArgs
DiagnosticInfoOptimizationBase::Argument NV
This is an optimization pass for GlobalISel generic memory operations.
bool RecursivelyDeleteTriviallyDeadInstructions(Value *V, const TargetLibraryInfo *TLI=nullptr, MemorySSAUpdater *MSSAU=nullptr, std::function< void(Value *)> AboutToDeleteCallback=std::function< void(Value *)>())
If the specified value is a trivially dead instruction, delete it.
iterator_range< T > make_range(T x, T y)
Convenience function for iterating over sub-ranges.
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,...
bool inferNonMandatoryLibFuncAttrs(Module *M, StringRef Name, const TargetLibraryInfo &TLI)
Analyze the name and prototype of the given function and set any applicable attributes.
bool isLibFuncEmittable(const Module *M, const TargetLibraryInfo *TLI, LibFunc TheLibFunc)
Check whether the library function is available on target and also that it in the current Module is a...
bool isMustProgress(const Loop *L)
Return true if this loop can be assumed to make progress.
raw_ostream & dbgs()
dbgs() - This returns a reference to a raw_ostream for debugging messages.
bool isModOrRefSet(const ModRefInfo MRI)
FunctionCallee getOrInsertLibFunc(Module *M, const TargetLibraryInfo &TLI, LibFunc TheLibFunc, FunctionType *T, AttributeList AttributeList)
Calls getOrInsertFunction() and then makes sure to add mandatory argument attributes.
ModRefInfo
Flags indicating whether a memory access modifies or references memory.
@ ModRef
The access may reference and may modify the value stored in memory.
@ Mod
The access may modify the value stored in memory.
bool VerifyMemorySSA
Enables verification of MemorySSA.
bool isConsecutiveAccess(Value *A, Value *B, const DataLayout &DL, ScalarEvolution &SE, bool CheckType=true)
Returns true if the memory operations A and B are consecutive.
bool isKnownNonNegative(const Value *V, const DataLayout &DL, unsigned Depth=0, AssumptionCache *AC=nullptr, const Instruction *CxtI=nullptr, const DominatorTree *DT=nullptr, bool UseInstrInfo=true)
Returns true if the give value is known to be non-negative.
PreservedAnalyses getLoopPassPreservedAnalyses()
Returns the minimum set of Analyses that all loop passes must preserve.
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...
bool decomposeBitTestICmp(Value *LHS, Value *RHS, CmpInst::Predicate &Pred, Value *&X, APInt &Mask, bool LookThroughTrunc=true)
Decompose an icmp into the form ((X & Mask) pred 0) if possible.
void swap(llvm::BitVector &LHS, llvm::BitVector &RHS)
Implement std::swap in terms of BitVector swap.
A collection of metadata nodes that might be associated with a memory access used by the alias-analys...
MDNode * TBAAStruct
The tag for type-based alias analysis (tbaa struct).
MDNode * Scope
The tag for alias scope specification (used with noalias).
MDNode * TBAA
The tag for type-based alias analysis.
AAMDNodes merge(const AAMDNodes &Other) const
Given two sets of AAMDNodes applying to potentially different locations, determine the best AAMDNodes...
MDNode * NoAlias
The tag specifying the noalias scope.
AAMDNodes extendTo(ssize_t Len) const
Create a new AAMDNode that describes this AAMDNode after extending it to apply to a series of bytes o...
This struct is a compact representation of a valid (non-zero power of two) alignment.
static bool Memset
When true, Memset is disabled.
static bool All
When true, the entire pass is disabled.
static bool Memcpy
When true, Memcpy is disabled.
The adaptor from a function pass to a loop pass computes these analyses and makes them available to t...
TargetTransformInfo & TTI
This struct is a compact representation of a valid (power of two) or undefined (0) alignment.
Match loop-invariant value.
match_LoopInvariant(const SubPattern_t &SP, const Loop *L)