107#define DEBUG_TYPE "loop-idiom"
109STATISTIC(NumMemSet,
"Number of memset's formed from loop stores");
110STATISTIC(NumMemCpy,
"Number of memcpy's formed from loop load+stores");
111STATISTIC(NumMemMove,
"Number of memmove's formed from loop load+stores");
113 NumShiftUntilBitTest,
114 "Number of uncountable loops recognized as 'shift until bitttest' idiom");
116 "Number of uncountable loops recognized as 'shift until zero' idiom");
121 cl::desc(
"Options to disable Loop Idiom Recognize Pass."),
128 cl::desc(
"Proceed with loop idiom recognize pass, but do "
129 "not convert loop(s) to memset."),
136 cl::desc(
"Proceed with loop idiom recognize pass, but do "
137 "not convert loop(s) to memcpy."),
142 "use-lir-code-size-heurs",
143 cl::desc(
"Use loop idiom recognition code size heuristics when compiling"
149class LoopIdiomRecognize {
150 Loop *CurLoop =
nullptr;
159 bool ApplyCodeSizeHeuristics;
160 std::unique_ptr<MemorySSAUpdater> MSSAU;
169 : AA(AA), DT(DT), LI(LI), SE(SE), TLI(TLI),
TTI(
TTI),
DL(
DL), ORE(ORE) {
171 MSSAU = std::make_unique<MemorySSAUpdater>(MSSA);
174 bool runOnLoop(
Loop *L);
180 StoreListMap StoreRefsForMemset;
181 StoreListMap StoreRefsForMemsetPattern;
182 StoreList StoreRefsForMemcpy;
184 bool HasMemsetPattern;
188 enum LegalStoreKind {
193 UnorderedAtomicMemcpy,
201 bool runOnCountableLoop();
206 LegalStoreKind isLegalStore(
StoreInst *SI);
207 enum class ForMemset {
No,
Yes };
211 template <
typename MemInst>
212 bool processLoopMemIntrinsic(
214 bool (LoopIdiomRecognize::*Processor)(MemInst *,
const SCEV *),
215 const SCEV *BECount);
219 bool processLoopStridedStore(
Value *DestPtr,
const SCEV *StoreSizeSCEV,
224 bool IsNegStride,
bool IsLoopMemset =
false);
225 bool processLoopStoreOfLoopLoad(
StoreInst *SI,
const SCEV *BECount);
226 bool processLoopStoreOfLoopLoad(
Value *DestPtr,
Value *SourcePtr,
232 const SCEV *BECount);
233 bool avoidLIRForMultiBlockLoop(
bool IsMemset =
false,
234 bool IsLoopMemset =
false);
240 bool runOnNoncountableLoop();
242 bool recognizePopcount();
245 bool recognizeAndInsertFFS();
250 bool IsCntPhiUsedOutsideLoop);
252 bool recognizeShiftUntilBitTest();
253 bool recognizeShiftUntilZero();
265 const auto *
DL = &L.getHeader()->getModule()->getDataLayout();
272 LoopIdiomRecognize LIR(&AR.
AA, &AR.
DT, &AR.
LI, &AR.
SE, &AR.
TLI, &AR.
TTI,
274 if (!LIR.runOnLoop(&L))
285 I->eraseFromParent();
294bool LoopIdiomRecognize::runOnLoop(
Loop *L) {
298 if (!
L->getLoopPreheader())
303 if (
Name ==
"memset" ||
Name ==
"memcpy")
307 ApplyCodeSizeHeuristics =
310 HasMemset = TLI->
has(LibFunc_memset);
311 HasMemsetPattern = TLI->
has(LibFunc_memset_pattern16);
312 HasMemcpy = TLI->
has(LibFunc_memcpy);
314 if (HasMemset || HasMemsetPattern || HasMemcpy)
316 return runOnCountableLoop();
318 return runOnNoncountableLoop();
321bool LoopIdiomRecognize::runOnCountableLoop() {
323 assert(!isa<SCEVCouldNotCompute>(BECount) &&
324 "runOnCountableLoop() called on a loop without a predictable"
325 "backedge-taken count");
329 if (
const SCEVConstant *BECst = dyn_cast<SCEVConstant>(BECount))
330 if (BECst->getAPInt() == 0)
348 bool MadeChange =
false;
356 MadeChange |= runOnLoopBlock(BB, BECount, ExitBlocks);
380 if (!
C || isa<ConstantExpr>(
C))
389 if (
DL->isBigEndian())
405 unsigned ArraySize = 16 /
Size;
410LoopIdiomRecognize::LegalStoreKind
411LoopIdiomRecognize::isLegalStore(
StoreInst *SI) {
413 if (
SI->isVolatile())
414 return LegalStoreKind::None;
416 if (!
SI->isUnordered())
417 return LegalStoreKind::None;
420 if (
SI->getMetadata(LLVMContext::MD_nontemporal))
421 return LegalStoreKind::None;
423 Value *StoredVal =
SI->getValueOperand();
424 Value *StorePtr =
SI->getPointerOperand();
429 return LegalStoreKind::None;
437 return LegalStoreKind::None;
443 dyn_cast<SCEVAddRecExpr>(SE->
getSCEV(StorePtr));
445 return LegalStoreKind::None;
448 if (!isa<SCEVConstant>(StoreEv->
getOperand(1)))
449 return LegalStoreKind::None;
460 bool UnorderedAtomic =
SI->isUnordered() && !
SI->isSimple();
469 return LegalStoreKind::Memset;
476 return LegalStoreKind::MemsetPattern;
484 unsigned StoreSize =
DL->getTypeStoreSize(
SI->getValueOperand()->getType());
485 if (StoreSize != Stride && StoreSize != -Stride)
486 return LegalStoreKind::None;
489 LoadInst *LI = dyn_cast<LoadInst>(
SI->getValueOperand());
493 return LegalStoreKind::None;
496 return LegalStoreKind::None;
504 return LegalStoreKind::None;
508 return LegalStoreKind::None;
511 UnorderedAtomic = UnorderedAtomic || LI->
isAtomic();
512 return UnorderedAtomic ? LegalStoreKind::UnorderedAtomicMemcpy
513 : LegalStoreKind::Memcpy;
516 return LegalStoreKind::None;
519void LoopIdiomRecognize::collectStores(
BasicBlock *BB) {
520 StoreRefsForMemset.clear();
521 StoreRefsForMemsetPattern.clear();
522 StoreRefsForMemcpy.clear();
529 switch (isLegalStore(SI)) {
530 case LegalStoreKind::None:
533 case LegalStoreKind::Memset: {
536 StoreRefsForMemset[
Ptr].push_back(SI);
538 case LegalStoreKind::MemsetPattern: {
541 StoreRefsForMemsetPattern[
Ptr].push_back(SI);
543 case LegalStoreKind::Memcpy:
544 case LegalStoreKind::UnorderedAtomicMemcpy:
545 StoreRefsForMemcpy.push_back(SI);
548 assert(
false &&
"unhandled return value");
557bool LoopIdiomRecognize::runOnLoopBlock(
567 bool MadeChange =
false;
574 for (
auto &SL : StoreRefsForMemset)
575 MadeChange |= processLoopStores(SL.second, BECount, ForMemset::Yes);
577 for (
auto &SL : StoreRefsForMemsetPattern)
578 MadeChange |= processLoopStores(SL.second, BECount, ForMemset::No);
581 for (
auto &SI : StoreRefsForMemcpy)
582 MadeChange |= processLoopStoreOfLoopLoad(SI, BECount);
584 MadeChange |= processLoopMemIntrinsic<MemCpyInst>(
585 BB, &LoopIdiomRecognize::processLoopMemCpy, BECount);
586 MadeChange |= processLoopMemIntrinsic<MemSetInst>(
587 BB, &LoopIdiomRecognize::processLoopMemSet, BECount);
594 const SCEV *BECount, ForMemset For) {
602 for (
unsigned i = 0, e = SL.
size(); i < e; ++i) {
603 assert(SL[i]->
isSimple() &&
"Expected only non-volatile stores.");
605 Value *FirstStoredVal = SL[i]->getValueOperand();
606 Value *FirstStorePtr = SL[i]->getPointerOperand();
608 cast<SCEVAddRecExpr>(SE->
getSCEV(FirstStorePtr));
610 unsigned FirstStoreSize =
DL->getTypeStoreSize(SL[i]->getValueOperand()->
getType());
613 if (FirstStride == FirstStoreSize || -FirstStride == FirstStoreSize) {
618 Value *FirstSplatValue =
nullptr;
619 Constant *FirstPatternValue =
nullptr;
621 if (For == ForMemset::Yes)
626 assert((FirstSplatValue || FirstPatternValue) &&
627 "Expected either splat value or pattern value.");
635 for (j = i + 1;
j <
e; ++
j)
637 for (j = i;
j > 0; --
j)
640 for (
auto &k : IndexQueue) {
641 assert(SL[k]->
isSimple() &&
"Expected only non-volatile stores.");
642 Value *SecondStorePtr = SL[
k]->getPointerOperand();
644 cast<SCEVAddRecExpr>(SE->
getSCEV(SecondStorePtr));
647 if (FirstStride != SecondStride)
650 Value *SecondStoredVal = SL[
k]->getValueOperand();
651 Value *SecondSplatValue =
nullptr;
652 Constant *SecondPatternValue =
nullptr;
654 if (For == ForMemset::Yes)
659 assert((SecondSplatValue || SecondPatternValue) &&
660 "Expected either splat value or pattern value.");
663 if (For == ForMemset::Yes) {
664 if (isa<UndefValue>(FirstSplatValue))
665 FirstSplatValue = SecondSplatValue;
666 if (FirstSplatValue != SecondSplatValue)
669 if (isa<UndefValue>(FirstPatternValue))
670 FirstPatternValue = SecondPatternValue;
671 if (FirstPatternValue != SecondPatternValue)
676 ConsecutiveChain[SL[i]] = SL[
k];
685 bool Changed =
false;
696 unsigned StoreSize = 0;
699 while (Tails.
count(
I) || Heads.count(
I)) {
700 if (TransformedStores.
count(
I))
704 StoreSize +=
DL->getTypeStoreSize(
I->getValueOperand()->getType());
706 I = ConsecutiveChain[
I];
716 if (StoreSize != Stride && StoreSize != -Stride)
719 bool IsNegStride = StoreSize == -Stride;
723 if (processLoopStridedStore(StorePtr, StoreSizeSCEV,
725 HeadStore, AdjacentStores, StoreEv, BECount,
727 TransformedStores.
insert(AdjacentStores.
begin(), AdjacentStores.
end());
737template <
typename MemInst>
738bool LoopIdiomRecognize::processLoopMemIntrinsic(
740 bool (LoopIdiomRecognize::*Processor)(MemInst *,
const SCEV *),
741 const SCEV *BECount) {
742 bool MadeChange =
false;
746 if (MemInst *
MI = dyn_cast<MemInst>(Inst)) {
748 if (!(this->*Processor)(
MI, BECount))
762bool LoopIdiomRecognize::processLoopMemCpy(
MemCpyInst *MCI,
763 const SCEV *BECount) {
774 if (!Dest || !Source)
789 if ((SizeInBytes >> 32) != 0)
795 dyn_cast<SCEVConstant>(StoreEv->
getOperand(1));
797 dyn_cast<SCEVConstant>(LoadEv->
getOperand(1));
798 if (!ConstStoreStride || !ConstLoadStride)
807 if (SizeInBytes != StoreStrideValue && SizeInBytes != -StoreStrideValue) {
810 <<
ore::NV(
"Inst",
"memcpy") <<
" in "
812 <<
" function will not be hoisted: "
813 <<
ore::NV(
"Reason",
"memcpy size is not equal to stride");
818 int64_t StoreStrideInt = StoreStrideValue.
getSExtValue();
821 if (StoreStrideInt != LoadStrideInt)
824 return processLoopStoreOfLoopLoad(
831bool LoopIdiomRecognize::processLoopMemSet(
MemSetInst *MSI,
832 const SCEV *BECount) {
847 if (!Ev || Ev->
getLoop() != CurLoop)
856 if (!PointerStrideSCEV || !MemsetSizeSCEV)
859 bool IsNegStride =
false;
860 const bool IsConstantSize = isa<ConstantInt>(MSI->
getLength());
862 if (IsConstantSize) {
873 if (SizeInBytes != Stride && SizeInBytes != -Stride)
876 IsNegStride = SizeInBytes == -Stride;
884 if (
Pointer->getType()->getPointerAddressSpace() != 0) {
897 const SCEV *PositiveStrideSCEV =
900 LLVM_DEBUG(
dbgs() <<
" MemsetSizeSCEV: " << *MemsetSizeSCEV <<
"\n"
901 <<
" PositiveStrideSCEV: " << *PositiveStrideSCEV
904 if (PositiveStrideSCEV != MemsetSizeSCEV) {
907 const SCEV *FoldedPositiveStride =
909 const SCEV *FoldedMemsetSize =
913 <<
" FoldedMemsetSize: " << *FoldedMemsetSize <<
"\n"
914 <<
" FoldedPositiveStride: " << *FoldedPositiveStride
917 if (FoldedPositiveStride != FoldedMemsetSize) {
934 BECount, IsNegStride,
true);
942 const SCEV *BECount,
const SCEV *StoreSizeSCEV,
952 const SCEVConstant *BECst = dyn_cast<SCEVConstant>(BECount);
953 const SCEVConstant *ConstSize = dyn_cast<SCEVConstant>(StoreSizeSCEV);
954 if (BECst && ConstSize)
976 Type *IntPtr,
const SCEV *StoreSizeSCEV,
979 if (!StoreSizeSCEV->
isOne()) {
993 const SCEV *TripCountS =
nullptr;
1000 if (
DL->getTypeSizeInBits(BECount->
getType()) <
1001 DL->getTypeSizeInBits(IntPtr) &&
1021 const SCEV *StoreSizeSCEV,
Loop *CurLoop,
1032bool LoopIdiomRecognize::processLoopStridedStore(
1036 const SCEV *BECount,
bool IsNegStride,
bool IsLoopMemset) {
1044 assert((SplatValue || PatternValue) &&
1045 "Expected either splat value or pattern value.");
1056 Type *DestInt8PtrTy =
Builder.getInt8PtrTy(DestAS);
1059 bool Changed =
false;
1067 if (!Expander.isSafeToExpand(Start))
1076 Expander.expandCodeFor(Start, DestInt8PtrTy, Preheader->
getTerminator());
1088 StoreSizeSCEV, *AA, Stores))
1091 if (avoidLIRForMultiBlockLoop(
true, IsLoopMemset))
1096 const SCEV *NumBytesS =
1097 getNumBytes(BECount, IntIdxTy, StoreSizeSCEV, CurLoop,
DL, SE);
1101 if (!Expander.isSafeToExpand(NumBytesS))
1105 Expander.expandCodeFor(NumBytesS, IntIdxTy, Preheader->
getTerminator());
1111 AATags = AATags.
merge(
Store->getAAMetadata());
1112 if (
auto CI = dyn_cast<ConstantInt>(NumBytes))
1113 AATags = AATags.
extendTo(CI->getZExtValue());
1117 NewCall =
Builder.CreateMemSet(
1118 BasePtr, SplatValue, NumBytes,
MaybeAlign(StoreAlignment),
1122 Type *Int8PtrTy = DestInt8PtrTy;
1124 StringRef FuncName =
"memset_pattern16";
1126 Builder.getVoidTy(), Int8PtrTy, Int8PtrTy, IntIdxTy);
1133 PatternValue,
".memset_pattern");
1137 NewCall =
Builder.CreateCall(MSP, {
BasePtr, PatternPtr, NumBytes});
1144 MemoryAccess *NewMemAcc = MSSAU->createMemoryAccessInBB(
1146 MSSAU->insertDef(cast<MemoryDef>(NewMemAcc),
true);
1150 <<
" from store to: " << *Ev <<
" at: " << *TheStore
1156 R <<
"Transformed loop-strided store in "
1158 <<
" function into a call to "
1161 if (!Stores.empty())
1163 for (
auto *
I : Stores) {
1164 R <<
ore::NV(
"FromBlock",
I->getParent()->getName())
1172 for (
auto *
I : Stores) {
1174 MSSAU->removeMemoryAccess(
I,
true);
1178 MSSAU->getMemorySSA()->verifyMemorySSA();
1180 ExpCleaner.markResultUsed();
1187bool LoopIdiomRecognize::processLoopStoreOfLoopLoad(
StoreInst *SI,
1188 const SCEV *BECount) {
1189 assert(
SI->isUnordered() &&
"Expected only non-volatile non-ordered stores.");
1191 Value *StorePtr =
SI->getPointerOperand();
1193 unsigned StoreSize =
DL->getTypeStoreSize(
SI->getValueOperand()->getType());
1196 LoadInst *LI = cast<LoadInst>(
SI->getValueOperand());
1206 return processLoopStoreOfLoopLoad(StorePtr, LoadPtr, StoreSizeSCEV,
1208 StoreEv, LoadEv, BECount);
1212class MemmoveVerifier {
1214 explicit MemmoveVerifier(
const Value &LoadBasePtr,
const Value &StoreBasePtr,
1217 LoadBasePtr.stripPointerCasts(), LoadOff,
DL)),
1219 StoreBasePtr.stripPointerCasts(), StoreOff,
DL)),
1220 IsSameObject(BP1 == BP2) {}
1222 bool loadAndStoreMayFormMemmove(
unsigned StoreSize,
bool IsNegStride,
1224 bool IsMemCpy)
const {
1228 if ((!IsNegStride && LoadOff <= StoreOff) ||
1229 (IsNegStride && LoadOff >= StoreOff))
1235 DL.getTypeSizeInBits(TheLoad.
getType()).getFixedValue() / 8;
1236 if (BP1 != BP2 || LoadSize != int64_t(StoreSize))
1238 if ((!IsNegStride && LoadOff < StoreOff + int64_t(StoreSize)) ||
1239 (IsNegStride && LoadOff + LoadSize > StoreOff))
1247 int64_t LoadOff = 0;
1248 int64_t StoreOff = 0;
1253 const bool IsSameObject;
1257bool LoopIdiomRecognize::processLoopStoreOfLoopLoad(
1266 if (isa<MemCpyInlineInst>(TheStore))
1278 bool Changed =
false;
1281 Type *IntIdxTy =
Builder.getIntNTy(
DL->getIndexSizeInBits(StrAS));
1284 const SCEVConstant *ConstStoreSize = dyn_cast<SCEVConstant>(StoreSizeSCEV);
1287 assert(ConstStoreSize &&
"store size is expected to be a constant");
1290 bool IsNegStride = StoreSize == -Stride;
1303 Value *StoreBasePtr = Expander.expandCodeFor(
1316 IgnoredInsts.
insert(TheStore);
1318 bool IsMemCpy = isa<MemCpyInst>(TheStore);
1319 const StringRef InstRemark = IsMemCpy ?
"memcpy" :
"load and store";
1321 bool LoopAccessStore =
1323 StoreSizeSCEV, *AA, IgnoredInsts);
1324 if (LoopAccessStore) {
1330 IgnoredInsts.
insert(TheLoad);
1332 BECount, StoreSizeSCEV, *AA, IgnoredInsts)) {
1336 <<
ore::NV(
"Inst", InstRemark) <<
" in "
1338 <<
" function will not be hoisted: "
1339 <<
ore::NV(
"Reason",
"The loop may access store location");
1343 IgnoredInsts.
erase(TheLoad);
1356 Value *LoadBasePtr = Expander.expandCodeFor(
1361 MemmoveVerifier
Verifier(*LoadBasePtr, *StoreBasePtr, *
DL);
1362 if (IsMemCpy && !
Verifier.IsSameObject)
1363 IgnoredInsts.
erase(TheStore);
1365 StoreSizeSCEV, *AA, IgnoredInsts)) {
1368 <<
ore::NV(
"Inst", InstRemark) <<
" in "
1370 <<
" function will not be hoisted: "
1371 <<
ore::NV(
"Reason",
"The loop may access load location");
1376 bool UseMemMove = IsMemCpy ?
Verifier.IsSameObject : LoopAccessStore;
1378 if (!
Verifier.loadAndStoreMayFormMemmove(StoreSize, IsNegStride, *TheLoad,
1382 if (avoidLIRForMultiBlockLoop())
1387 const SCEV *NumBytesS =
1388 getNumBytes(BECount, IntIdxTy, StoreSizeSCEV, CurLoop,
DL, SE);
1391 Expander.expandCodeFor(NumBytesS, IntIdxTy, Preheader->
getTerminator());
1395 AATags = AATags.
merge(StoreAATags);
1396 if (
auto CI = dyn_cast<ConstantInt>(NumBytes))
1397 AATags = AATags.
extendTo(CI->getZExtValue());
1407 NewCall =
Builder.CreateMemMove(
1408 StoreBasePtr, StoreAlign, LoadBasePtr, LoadAlign, NumBytes,
1412 Builder.CreateMemCpy(StoreBasePtr, StoreAlign, LoadBasePtr, LoadAlign,
1413 NumBytes,
false, AATags.
TBAA,
1421 assert((StoreAlign && LoadAlign) &&
1422 "Expect unordered load/store to have align.");
1423 if (*StoreAlign < StoreSize || *LoadAlign < StoreSize)
1436 NewCall =
Builder.CreateElementUnorderedAtomicMemCpy(
1437 StoreBasePtr, *StoreAlign, LoadBasePtr, *LoadAlign, NumBytes, StoreSize,
1443 MemoryAccess *NewMemAcc = MSSAU->createMemoryAccessInBB(
1445 MSSAU->insertDef(cast<MemoryDef>(NewMemAcc),
true);
1449 <<
" from load ptr=" << *LoadEv <<
" at: " << *TheLoad
1451 <<
" from store ptr=" << *StoreEv <<
" at: " << *TheStore
1457 <<
"Formed a call to "
1459 <<
"() intrinsic from " <<
ore::NV(
"Inst", InstRemark)
1470 MSSAU->removeMemoryAccess(TheStore,
true);
1473 MSSAU->getMemorySSA()->verifyMemorySSA();
1478 ExpCleaner.markResultUsed();
1485bool LoopIdiomRecognize::avoidLIRForMultiBlockLoop(
bool IsMemset,
1486 bool IsLoopMemset) {
1487 if (ApplyCodeSizeHeuristics && CurLoop->
getNumBlocks() > 1) {
1488 if (CurLoop->
isOutermost() && (!IsMemset || !IsLoopMemset)) {
1490 <<
" : LIR " << (IsMemset ?
"Memset" :
"Memcpy")
1491 <<
" avoided: multi-block top-level loop\n");
1499bool LoopIdiomRecognize::runOnNoncountableLoop() {
1502 <<
"] Noncountable Loop %"
1505 return recognizePopcount() || recognizeAndInsertFFS() ||
1506 recognizeShiftUntilBitTest() || recognizeShiftUntilZero();
1516 bool JmpOnZero =
false) {
1525 if (!CmpZero || !CmpZero->
isZero())
1536 return Cond->getOperand(0);
1545 auto *PhiX = dyn_cast<PHINode>(VarX);
1546 if (PhiX && PhiX->getParent() == LoopEntry &&
1547 (PhiX->getOperand(0) == DefX || PhiX->
getOperand(1) == DefX))
1583 Value *VarX1, *VarX0;
1586 DefX2 = CountInst =
nullptr;
1587 VarX1 = VarX0 =
nullptr;
1588 PhiX = CountPhi =
nullptr;
1594 dyn_cast<BranchInst>(LoopEntry->
getTerminator()), LoopEntry))
1595 DefX2 = dyn_cast<Instruction>(
T);
1602 if (!DefX2 || DefX2->
getOpcode() != Instruction::And)
1607 if ((SubOneOp = dyn_cast<BinaryOperator>(DefX2->
getOperand(0))))
1611 SubOneOp = dyn_cast<BinaryOperator>(DefX2->
getOperand(1));
1613 if (!SubOneOp || SubOneOp->
getOperand(0) != VarX1)
1619 (SubOneOp->
getOpcode() == Instruction::Add &&
1632 CountInst =
nullptr;
1635 if (Inst.
getOpcode() != Instruction::Add)
1639 if (!Inc || !Inc->
isOne())
1647 bool LiveOutLoop =
false;
1649 if ((cast<Instruction>(U))->
getParent() != LoopEntry) {
1669 auto *PreCondBr = dyn_cast<BranchInst>(PreCondBB->
getTerminator());
1674 CntInst = CountInst;
1714 Value *VarX =
nullptr;
1723 dyn_cast<BranchInst>(LoopEntry->
getTerminator()), LoopEntry))
1724 DefX = dyn_cast<Instruction>(
T);
1729 if (!DefX || !DefX->
isShift())
1731 IntrinID = DefX->
getOpcode() == Instruction::Shl ? Intrinsic::cttz :
1734 if (!Shft || !Shft->
isOne())
1759 if (Inst.
getOpcode() != Instruction::Add)
1783bool LoopIdiomRecognize::recognizeAndInsertFFS() {
1795 size_t IdiomCanonicalSize = 6;
1798 CntInst, CntPhi, DefX))
1801 bool IsCntPhiUsedOutsideLoop =
false;
1803 if (!CurLoop->
contains(cast<Instruction>(U))) {
1804 IsCntPhiUsedOutsideLoop =
true;
1807 bool IsCntInstUsedOutsideLoop =
false;
1809 if (!CurLoop->
contains(cast<Instruction>(U))) {
1810 IsCntInstUsedOutsideLoop =
true;
1815 if (IsCntInstUsedOutsideLoop && IsCntPhiUsedOutsideLoop)
1821 bool ZeroCheck =
false;
1830 if (!IsCntPhiUsedOutsideLoop) {
1834 auto *PreCondBI = dyn_cast<BranchInst>(PreCondBB->getTerminator());
1859 std::distance(InstWithoutDebugIt.begin(), InstWithoutDebugIt.end());
1864 if (HeaderSize != IdiomCanonicalSize &&
1868 transformLoopToCountable(IntrinID, PH, CntInst, CntPhi, InitX, DefX,
1870 IsCntPhiUsedOutsideLoop);
1878bool LoopIdiomRecognize::recognizePopcount() {
1892 if (LoopBody->
size() >= 20) {
1902 if (!EntryBI || EntryBI->isConditional())
1910 auto *PreCondBI = dyn_cast<BranchInst>(PreCondBB->getTerminator());
1911 if (!PreCondBI || PreCondBI->isUnconditional())
1920 transformLoopToPopcount(PreCondBB, CntInst, CntPhi, Val);
1926 Value *Ops[] = {Val};
1982void LoopIdiomRecognize::transformLoopToCountable(
1985 bool ZeroCheck,
bool IsCntPhiUsedOutsideLoop) {
1999 if (IsCntPhiUsedOutsideLoop) {
2000 if (DefX->
getOpcode() == Instruction::AShr)
2001 InitXNext =
Builder.CreateAShr(InitX, 1);
2002 else if (DefX->
getOpcode() == Instruction::LShr)
2003 InitXNext =
Builder.CreateLShr(InitX, 1);
2004 else if (DefX->
getOpcode() == Instruction::Shl)
2005 InitXNext =
Builder.CreateShl(InitX, 1);
2015 Value *NewCount = Count;
2016 if (IsCntPhiUsedOutsideLoop)
2019 NewCount =
Builder.CreateZExtOrTrunc(NewCount, CntInst->
getType());
2022 if (cast<ConstantInt>(CntInst->
getOperand(1))->isOne()) {
2025 ConstantInt *InitConst = dyn_cast<ConstantInt>(CntInitVal);
2026 if (!InitConst || !InitConst->
isZero())
2027 NewCount =
Builder.CreateAdd(NewCount, CntInitVal);
2031 NewCount =
Builder.CreateSub(CntInitVal, NewCount);
2044 ICmpInst *LbCond = cast<ICmpInst>(LbBr->getCondition());
2048 Builder.SetInsertPoint(LbCond);
2063 if (IsCntPhiUsedOutsideLoop)
2073void LoopIdiomRecognize::transformLoopToPopcount(
BasicBlock *PreCondBB,
2077 auto *PreCondBr = cast<BranchInst>(PreCondBB->
getTerminator());
2086 Value *PopCnt, *PopCntZext, *NewCount, *TripCnt;
2089 NewCount = PopCntZext =
2090 Builder.CreateZExtOrTrunc(PopCnt, cast<IntegerType>(CntPhi->
getType()));
2092 if (NewCount != PopCnt)
2093 (cast<Instruction>(NewCount))->setDebugLoc(
DL);
2100 ConstantInt *InitConst = dyn_cast<ConstantInt>(CntInitVal);
2101 if (!InitConst || !InitConst->
isZero()) {
2102 NewCount =
Builder.CreateAdd(NewCount, CntInitVal);
2103 (cast<Instruction>(NewCount))->setDebugLoc(
DL);
2112 ICmpInst *PreCond = cast<ICmpInst>(PreCondBr->getCondition());
2114 Value *Opnd0 = PopCntZext;
2119 ICmpInst *NewPreCond = cast<ICmpInst>(
2121 PreCondBr->setCondition(NewPreCond);
2149 ICmpInst *LbCond = cast<ICmpInst>(LbBr->getCondition());
2154 Builder.SetInsertPoint(LbCond);
2157 "tcdec",
false,
true));
2185 : SubPattern(SP), L(L) {}
2187 template <
typename ITy>
bool match(ITy *V) {
2188 return L->isLoopInvariant(V) && SubPattern.match(V);
2193template <
typename Ty>
2224 " Performing shift-until-bittest idiom detection.\n");
2234 assert(LoopPreheaderBB &&
"There is always a loop preheader.");
2236 using namespace PatternMatch;
2241 Value *CmpLHS, *CmpRHS;
2252 auto MatchVariableBitMask = [&]() {
2261 auto MatchConstantBitMask = [&]() {
2267 auto MatchDecomposableConstantBitMask = [&]() {
2275 if (!MatchVariableBitMask() && !MatchConstantBitMask() &&
2276 !MatchDecomposableConstantBitMask()) {
2282 auto *CurrXPN = dyn_cast<PHINode>(CurrX);
2283 if (!CurrXPN || CurrXPN->getParent() != LoopHeaderBB) {
2288 BaseX = CurrXPN->getIncomingValueForBlock(LoopPreheaderBB);
2290 dyn_cast<Instruction>(CurrXPN->getIncomingValueForBlock(LoopHeaderBB));
2293 "Expected BaseX to be avaliable in the preheader!");
2304 "Should only get equality predicates here.");
2314 if (TrueBB != LoopHeaderBB) {
2373bool LoopIdiomRecognize::recognizeShiftUntilBitTest() {
2374 bool MadeChange =
false;
2376 Value *
X, *BitMask, *BitPos, *XCurr;
2381 " shift-until-bittest idiom detection failed.\n");
2391 assert(LoopPreheaderBB &&
"There is always a loop preheader.");
2394 assert(SuccessorBB &&
"There is only a single successor.");
2400 Type *Ty =
X->getType();
2414 " Intrinsic is too costly, not beneficial\n");
2429 BitPos->
getName() +
".lowbitmask");
2431 Builder.CreateOr(LowBitMask, BitMask, BitPos->
getName() +
".mask");
2432 Value *XMasked =
Builder.CreateAnd(
X, Mask,
X->getName() +
".masked");
2434 IntrID, Ty, {XMasked,
Builder.getTrue()},
2435 nullptr, XMasked->
getName() +
".numleadingzeros");
2438 XMasked->
getName() +
".numactivebits",
true,
2440 Value *XMaskedLeadingOnePos =
2442 XMasked->
getName() +
".leadingonepos",
false,
2446 BitPos, XMaskedLeadingOnePos, CurLoop->
getName() +
".backedgetakencount",
2450 Value *LoopTripCount =
2452 CurLoop->
getName() +
".tripcount",
true,
2461 if (
auto *
I = dyn_cast<Instruction>(NewX))
2462 I->copyIRFlags(XNext,
true);
2474 NewXNext =
Builder.CreateShl(
X, LoopTripCount);
2483 if (
auto *
I = dyn_cast<Instruction>(NewXNext))
2484 I->copyIRFlags(XNext,
true);
2503 true, Bitwidth != 2);
2506 auto *IVCheck =
Builder.CreateICmpEQ(IVNext, LoopTripCount,
2507 CurLoop->
getName() +
".ivcheck");
2508 Builder.CreateCondBr(IVCheck, SuccessorBB, LoopHeaderBB);
2513 IV->addIncoming(IVNext, LoopHeaderBB);
2524 ++NumShiftUntilBitTest;
2560 const SCEV *&ExtraOffsetExpr,
2561 bool &InvertedCond) {
2563 " Performing shift-until-zero idiom detection.\n");
2576 assert(LoopPreheaderBB &&
"There is always a loop preheader.");
2578 using namespace PatternMatch;
2587 !
match(ValShiftedIsZero,
2601 IntrinID = ValShifted->
getOpcode() == Instruction::Shl ? Intrinsic::cttz
2610 else if (
match(NBits,
2614 ExtraOffsetExpr = SE->
getSCEV(ExtraOffset);
2621 auto *IVPN = dyn_cast<PHINode>(
IV);
2622 if (!IVPN || IVPN->getParent() != LoopHeaderBB) {
2627 Start = IVPN->getIncomingValueForBlock(LoopPreheaderBB);
2628 IVNext = dyn_cast<Instruction>(IVPN->getIncomingValueForBlock(LoopHeaderBB));
2638 "Should only get equality predicates here.");
2649 if (FalseBB != LoopHeaderBB) {
2660 if (ValShifted->
getOpcode() == Instruction::AShr &&
2724bool LoopIdiomRecognize::recognizeShiftUntilZero() {
2725 bool MadeChange =
false;
2731 const SCEV *ExtraOffsetExpr;
2734 Start, Val, ExtraOffsetExpr, InvertedCond)) {
2736 " shift-until-zero idiom detection failed.\n");
2746 assert(LoopPreheaderBB &&
"There is always a loop preheader.");
2749 assert(SuccessorBB &&
"There is only a single successor.");
2752 Builder.SetCurrentDebugLocation(
IV->getDebugLoc());
2768 " Intrinsic is too costly, not beneficial\n");
2775 bool OffsetIsZero =
false;
2776 if (
auto *ExtraOffsetExprC = dyn_cast<SCEVConstant>(ExtraOffsetExpr))
2777 OffsetIsZero = ExtraOffsetExprC->isZero();
2782 IntrID, Ty, {Val,
Builder.getFalse()},
2783 nullptr, Val->
getName() +
".numleadingzeros");
2786 Val->
getName() +
".numactivebits",
true,
2790 Expander.setInsertPoint(&*
Builder.GetInsertPoint());
2791 Value *ExtraOffset = Expander.expandCodeFor(ExtraOffsetExpr);
2794 ValNumActiveBits, ExtraOffset, ValNumActiveBits->
getName() +
".offset",
2795 OffsetIsZero,
true);
2796 Value *IVFinal =
Builder.CreateIntrinsic(Intrinsic::smax, {Ty},
2797 {ValNumActiveBitsOffset, Start},
2798 nullptr,
"iv.final");
2800 auto *LoopBackedgeTakenCount = cast<Instruction>(
Builder.CreateSub(
2801 IVFinal, Start, CurLoop->
getName() +
".backedgetakencount",
2802 OffsetIsZero,
true));
2806 Value *LoopTripCount =
2808 CurLoop->
getName() +
".tripcount",
true,
2814 IV->replaceUsesOutsideBlock(IVFinal, LoopHeaderBB);
2820 auto *CIV =
Builder.CreatePHI(Ty, 2, CurLoop->
getName() +
".iv");
2826 true, Bitwidth != 2);
2829 auto *CIVCheck =
Builder.CreateICmpEQ(CIVNext, LoopTripCount,
2830 CurLoop->
getName() +
".ivcheck");
2831 auto *NewIVCheck = CIVCheck;
2833 NewIVCheck =
Builder.CreateNot(CIVCheck);
2834 NewIVCheck->takeName(ValShiftedIsZero);
2838 auto *IVDePHId =
Builder.CreateAdd(CIV, Start,
"",
false,
2840 IVDePHId->takeName(
IV);
2844 Builder.CreateCondBr(CIVCheck, SuccessorBB, LoopHeaderBB);
2849 CIV->addIncoming(CIVNext, LoopHeaderBB);
2857 IV->replaceAllUsesWith(IVDePHId);
2858 IV->eraseFromParent();
2867 ++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 const SCEV * getTripCount(const SCEV *BECount, Type *IntPtr, Loop *CurLoop, const DataLayout *DL, ScalarEvolution *SE)
Compute trip count from the backedge taken count.
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.
Type * getType() const
Return the LLVM type of this SCEV expression.
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 * getOne(Type *Ty)
Return a SCEV for the constant 1 of a specific type.
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 * getZeroExtendExpr(const SCEV *Op, Type *Ty, unsigned Depth=0)
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.
bool isLoopEntryGuardedByCond(const Loop *L, ICmpInst::Predicate Pred, const SCEV *LHS, const SCEV *RHS)
Test whether entry to the loop is protected by a conditional between LHS and RHS.
const SCEV * getAddExpr(SmallVectorImpl< const SCEV * > &Ops, SCEV::NoWrapFlags Flags=SCEV::FlagAnyWrap, unsigned Depth=0)
Get a canonical add expression, or something simpler if possible.
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.
constexpr std::nullopt_t None
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)