SLPVectorizer.cpp File Reference (original) (raw)

Go to the source code of this file.

Classes
class	llvm::slpvectorizer::BoUpSLP
	Bottom Up SLP Vectorizer. More...

struct	llvm::slpvectorizer::BoUpSLP::EdgeInfo
	This structure holds any data we need about the edges being traversed during buildTree_rec(). More...

class	llvm::slpvectorizer::BoUpSLP::LookAheadHeuristics
	A helper class used for scoring candidates for two consecutive lanes. More...

class	llvm::slpvectorizer::BoUpSLP::VLOperands
	A helper data structure to hold the operands of a vector of instructions. More...

struct	llvm::GraphTraits< BoUpSLP * >

struct	llvm::GraphTraits< BoUpSLP * >::ChildIteratorType
	Add the VectorizableTree to the index iterator to be able to return TreeEntry pointers. More...

class	llvm::GraphTraits< BoUpSLP * >::nodes_iterator
	For the node iterator we just need to turn the TreeEntry iterator into a TreeEntry* iterator so that it dereferences to NodeRef. More...

struct	llvm::DOTGraphTraits< BoUpSLP * >

class	llvm::slpvectorizer::BoUpSLP::ShuffleCostEstimator
	Merges shuffle masks and emits final shuffle instruction, if required. More...

class	llvm::slpvectorizer::BoUpSLP::ShuffleInstructionBuilder
	Merges shuffle masks and emits final shuffle instruction, if required. More...

Namespaces
namespace	llvm
	This is an optimization pass for GlobalISel generic memory operations.

namespace	llvm::slpvectorizer
	A private "module" namespace for types and utilities used by this pass.

Functions
	STATISTIC (NumVectorInstructions, "Number of vector instructions generated")

	DEBUG_COUNTER (VectorizedGraphs, "slp-vectorized", "Controls which SLP graphs should be vectorized.")

static bool	isValidElementType (Type *Ty)
	Predicate for the element types that the SLP vectorizer supports.

static Type *	getValueType (Value *V)
	Returns the type of the given value/instruction V.

static unsigned	getNumElements (Type *Ty)

static FixedVectorType *	getWidenedType (Type *ScalarTy, unsigned VF)

static unsigned	getFullVectorNumberOfElements (const TargetTransformInfo &TTI, Type *Ty, unsigned Sz)
	Returns the number of elements of the given type Ty, not less than Sz, which forms type, which splits by TTI into whole vector types during legalization.

static unsigned	getFloorFullVectorNumberOfElements (const TargetTransformInfo &TTI, Type *Ty, unsigned Sz)
	Returns the number of elements of the given type Ty, not greater than Sz, which forms type, which splits by TTI into whole vector types during legalization.

static void	transformScalarShuffleIndiciesToVector (unsigned VecTyNumElements, SmallVectorImpl< int > &Mask)

static unsigned	getShufflevectorNumGroups (ArrayRef< Value * > VL)

static SmallVector< int >	calculateShufflevectorMask (ArrayRef< Value * > VL)

static bool	isConstant (Value *V)

static bool	isVectorLikeInstWithConstOps (Value *V)
	Checks if V is one of vector-like instructions, i.e.

static unsigned	getPartNumElems (unsigned Size, unsigned NumParts)
	Returns power-of-2 number of elements in a single register (part), given the total number of elements Size and number of registers (parts) NumParts.

static unsigned	getNumElems (unsigned Size, unsigned PartNumElems, unsigned Part)
	Returns correct remaining number of elements, considering total amount Size, (power-of-2 number) of elements in a single register PartNumElems and current register (part) Part.

static std::string	shortBundleName (ArrayRef< Value * > VL, int Idx=-1)
	Print a short descriptor of the instruction bundle suitable for debug output.

static bool	allSameBlock (ArrayRef< Value * > VL)

static bool	allConstant (ArrayRef< Value * > VL)

static bool	isSplat (ArrayRef< Value * > VL)

static bool	isCommutative (Instruction *I)

template<typename T >
static std::optional< unsigned >	getInsertExtractIndex (const Value *Inst, unsigned Offset)

static std::optional< unsigned >	getElementIndex (const Value *Inst, unsigned Offset=0)

static SmallBitVector	buildUseMask (int VF, ArrayRef< int > Mask, UseMask MaskArg)
	Prepares a use bitset for the given mask either for the first argument or for the second.

template<bool IsPoisonOnly = false>
static SmallBitVector	isUndefVector (const Value *V, const SmallBitVector &UseMask={})
	Checks if the given value is actually an undefined constant vector.

static std::optional< TargetTransformInfo::ShuffleKind >	isFixedVectorShuffle (ArrayRef< Value * > VL, SmallVectorImpl< int > &Mask, AssumptionCache *AC)
	Checks if the vector of instructions can be represented as a shuffle, like: x0 = extractelement <4 x i8> x, i32 0 x3 = extractelement <4 x i8> x, i32 3 y1 = extractelement <4 x i8> y, i32 1 y2 = extractelement <4 x i8> y, i32 2 x0x0 = mul i8 x0, x0 x3x3 = mul i8 x3, x3 y1y1 = mul i8 y1, y1 y2y2 = mul i8 y2, y2 ins1 = insertelement <4 x i8> poison, i8 x0x0, i32 0 ins2 = insertelement <4 x i8> ins1, i8 x3x3, i32 1 ins3 = insertelement <4 x i8> ins2, i8 y1y1, i32 2 ins4 = insertelement <4 x i8> ins3, i8 y2y2, i32 3 ret <4 x i8> ins4 can be transformed into: %1 = shufflevector <4 x i8> x, <4 x i8> y, <4 x i32> <i32 0, i32 3, i32 5, i32 6> %2 = mul <4 x i8> %1, %1 ret <4 x i8> %2 Mask will return the Shuffle Mask equivalent to the extracted elements.

static std::optional< unsigned >	getExtractIndex (Instruction *E)

static bool	isValidForAlternation (unsigned Opcode)

static InstructionsState	getSameOpcode (ArrayRef< Value * > VL, const TargetLibraryInfo &TLI)

static bool	areCompatibleCmpOps (Value BaseOp0, Value BaseOp1, Value Op0, Value Op1, const TargetLibraryInfo &TLI)
	Checks if the provided operands of 2 cmp instructions are compatible, i.e.

static bool	isCmpSameOrSwapped (const CmpInst BaseCI, const CmpInst CI, const TargetLibraryInfo &TLI)

static bool	allSameType (ArrayRef< Value * > VL)

static bool	doesInTreeUserNeedToExtract (Value Scalar, Instruction UserInst, TargetLibraryInfo TLI, const TargetTransformInfo TTI)

static MemoryLocation	getLocation (Instruction *I)

static bool	isSimple (Instruction *I)

static void	addMask (SmallVectorImpl< int > &Mask, ArrayRef< int > SubMask, bool ExtendingManyInputs=false)
	Shuffles Mask in accordance with the given SubMask.

static void	fixupOrderingIndices (MutableArrayRef< unsigned > Order)
	Order may have elements assigned special value (size) which is out of bounds.

static SmallBitVector	getAltInstrMask (ArrayRef< Value * > VL, unsigned Opcode0, unsigned Opcode1)

static void	llvm::inversePermutation (ArrayRef< unsigned > Indices, SmallVectorImpl< int > &Mask)

static void	llvm::reorderScalars (SmallVectorImpl< Value * > &Scalars, ArrayRef< int > Mask)
	Reorders the list of scalars in accordance with the given Mask.

static bool	llvm::areAllOperandsNonInsts (Value *V)
	Checks if the provided value does not require scheduling.

static bool	llvm::isUsedOutsideBlock (Value *V)
	Checks if the provided value does not require scheduling.

static bool	llvm::doesNotNeedToBeScheduled (Value *V)
	Checks if the specified value does not require scheduling.

static bool	llvm::doesNotNeedToSchedule (ArrayRef< Value * > VL)
	Checks if the specified array of instructions does not require scheduling.

static bool	llvm::hasFullVectorsOrPowerOf2 (const TargetTransformInfo &TTI, Type *Ty, unsigned Sz)
	Returns true if widened type of Ty elements with size Sz represents full vector type, i.e.

static void	reorderReuses (SmallVectorImpl< int > &Reuses, ArrayRef< int > Mask)
	Reorders the given Reuses mask according to the given Mask.

static void	reorderOrder (SmallVectorImpl< unsigned > &Order, ArrayRef< int > Mask, bool BottomOrder=false)
	Reorders the given Order according to the given Mask.

static bool	arePointersCompatible (Value Ptr1, Value Ptr2, const TargetLibraryInfo &TLI, bool CompareOpcodes=true)

template<typename T >
static Align	computeCommonAlignment (ArrayRef< Value * > VL)
	Calculates minimal alignment as a common alignment.

static bool	isReverseOrder (ArrayRef< unsigned > Order)
	Check if Order represents reverse order.

static std::optional< Value * >	calculateRtStride (ArrayRef< Value * > PointerOps, Type ElemTy, const DataLayout &DL, ScalarEvolution &SE, SmallVectorImpl< unsigned > &SortedIndices, Instruction Inst=nullptr)
	Checks if the provided list of pointers Pointers represents the strided pointers for type ElemTy.

static std::pair< InstructionCost, InstructionCost >	getGEPCosts (const TargetTransformInfo &TTI, ArrayRef< Value * > Ptrs, Value BasePtr, unsigned Opcode, TTI::TargetCostKind CostKind, Type ScalarTy, VectorType *VecTy)
	Calculate the scalar and the vector costs from vectorizing set of GEPs.

static InstructionCost	getShuffleCost (const TargetTransformInfo &TTI, TTI::ShuffleKind Kind, VectorType Tp, ArrayRef< int > Mask={}, TTI::TargetCostKind CostKind=TTI::TCK_RecipThroughput, int Index=0, VectorType SubTp=nullptr, ArrayRef< const Value * > Args={})
	Returns the cost of the shuffle instructions with the given Kind, vector type Tp and optional Mask.

static Value *	createInsertVector (IRBuilderBase &Builder, Value Vec, Value V, unsigned Index, function_ref< Value (Value , Value *, ArrayRef< int >)> Generator={})
	Correctly creates insert_subvector, checking that the index is multiple of the subvectors length.

static Value *	createExtractVector (IRBuilderBase &Builder, Value *Vec, unsigned SubVecVF, unsigned Index)
	Correctly creates extract_subvector, checking that the index is multiple of the subvectors length.

static bool	clusterSortPtrAccesses (ArrayRef< Value * > VL, ArrayRef< BasicBlock * > BBs, Type *ElemTy, const DataLayout &DL, ScalarEvolution &SE, SmallVectorImpl< unsigned > &SortedIndices)

static bool	areTwoInsertFromSameBuildVector (InsertElementInst VU, InsertElementInst V, function_ref< Value (InsertElementInst )> GetBaseOperand)
	Check if two insertelement instructions are from the same buildvector.

static bool	isRepeatedNonIdentityClusteredMask (ArrayRef< int > Mask, unsigned Sz)
	Checks if the given mask is a "clustered" mask with the same clusters of size Sz, which are not identity submasks.

static void	combineOrders (MutableArrayRef< unsigned > Order, ArrayRef< unsigned > SecondaryOrder)

static LLVM_DUMP_METHOD void	dumpOrder (const BoUpSLP::OrdersType &Order)

static void	gatherPossiblyVectorizableLoads (const BoUpSLP &R, ArrayRef< Value * > VL, const DataLayout &DL, ScalarEvolution &SE, const TargetTransformInfo &TTI, SmallVectorImpl< SmallVector< std::pair< LoadInst *, int > > > &GatheredLoads, bool AddNew=true)
	Tries to find subvector of loads and builds new vector of only loads if can be profitable.

static bool	needToScheduleSingleInstruction (ArrayRef< Value * > VL)

static std::pair< size_t, size_t >	generateKeySubkey (Value V, const TargetLibraryInfo TLI, function_ref< hash_code(size_t, LoadInst *)> LoadsSubkeyGenerator, bool AllowAlternate)
	Generates key/subkey pair for the given value to provide effective sorting of the values and better detection of the vectorizable values sequences.

static bool	isAlternateInstruction (const Instruction I, const Instruction MainOp, const Instruction *AltOp, const TargetLibraryInfo &TLI)
	Checks if the specified instruction I is an alternate operation for the given MainOp and AltOp instructions.

static std::pair< InstructionCost, InstructionCost >	getVectorCallCosts (CallInst CI, FixedVectorType VecTy, TargetTransformInfo TTI, TargetLibraryInfo TLI, ArrayRef< Type * > ArgTys)

static SmallVector< Type * >	buildIntrinsicArgTypes (const CallInst CI, const Intrinsic::ID ID, const unsigned VF, unsigned MinBW, const TargetTransformInfo TTI)
	Builds the arguments types vector for the given call instruction with the given ID for the specified vector factor.

static bool	isLoadCombineCandidateImpl (Value Root, unsigned NumElts, TargetTransformInfo TTI, bool MustMatchOrInst)

static bool	isFirstInsertElement (const InsertElementInst IE1, const InsertElementInst IE2)
	Checks if the IE1 instructions is followed by IE2 instruction in the buildvector sequence.

template<typename T >
static T *	performExtractsShuffleAction (MutableArrayRef< std::pair< T , SmallVector< int > > > ShuffleMask, Value Base, function_ref< unsigned(T )> GetVF, function_ref< std::pair< T , bool >(T , ArrayRef< int >, bool)> ResizeAction, function_ref< T (ArrayRef< int >, ArrayRef< T * >)> Action)
	Does the analysis of the provided shuffle masks and performs the requested actions on the vectors with the given shuffle masks.

static Instruction *	propagateMetadata (Instruction Inst, ArrayRef< Value > VL)

static RecurKind	getRdxKind (Value *V)
	Gets recurrence kind from the specified value.

static bool	checkTreeSizes (ArrayRef< std::pair< unsigned, unsigned > > Sizes, bool First)
	Checks if the quadratic mean deviation is less than 90% of the mean size.

static std::optional< unsigned >	getAggregateSize (Instruction *InsertInst)

static void	findBuildAggregate_rec (Instruction LastInsertInst, TargetTransformInfo TTI, SmallVectorImpl< Value * > &BuildVectorOpds, SmallVectorImpl< Value * > &InsertElts, unsigned OperandOffset, const BoUpSLP &R)

static bool	findBuildAggregate (Instruction LastInsertInst, TargetTransformInfo TTI, SmallVectorImpl< Value * > &BuildVectorOpds, SmallVectorImpl< Value * > &InsertElts, const BoUpSLP &R)
	Recognize construction of vectors like ra = insertelement <4 x float> poison, float s0, i32 0 rb = insertelement <4 x float> ra, float s1, i32 1 rc = insertelement <4 x float> rb, float s2, i32 2 rd = insertelement <4 x float> rc, float s3, i32 3 starting from the last insertelement or insertvalue instruction.

static Instruction *	getReductionInstr (const DominatorTree DT, PHINode P, BasicBlock ParentBB, LoopInfo LI)
	Try and get a reduction instruction from a phi node.

static bool	matchRdxBop (Instruction I, Value &V0, Value *&V1)

static Instruction *	tryGetSecondaryReductionRoot (PHINode Phi, Instruction Root)
	We could have an initial reduction that is not an add.

static Instruction *	getNonPhiOperand (Instruction I, PHINode Phi)
	Returns the first operand of I that does not match Phi.

static bool	isReductionCandidate (Instruction *I)
	\Returns true if I is a candidate instruction for reduction vectorization.

template<typename T >
static bool	tryToVectorizeSequence (SmallVectorImpl< T * > &Incoming, function_ref< bool(T , T )> Comparator, function_ref< bool(T , T )> AreCompatible, function_ref< bool(ArrayRef< T * >, bool)> TryToVectorizeHelper, bool MaxVFOnly, BoUpSLP &R)

template<bool IsCompatibility>
static bool	compareCmp (Value V, Value V2, TargetLibraryInfo &TLI, const DominatorTree &DT)
	Compare two cmp instructions.

Variables
static cl::opt< bool >	RunSLPVectorization ("vectorize-slp", cl::init(true), cl::Hidden, cl::desc("Run the SLP vectorization passes"))

static cl::opt< bool >	SLPReVec ("slp-revec", cl::init(false), cl::Hidden, cl::desc("Enable vectorization for wider vector utilization"))

static cl::opt< int >	SLPCostThreshold ("slp-threshold", cl::init(0), cl::Hidden, cl::desc("Only vectorize if you gain more than this " "number "))

static cl::opt< bool >	SLPSkipEarlyProfitabilityCheck ("slp-skip-early-profitability-check", cl::init(false), cl::Hidden, cl::desc("When true, SLP vectorizer bypasses profitability checks based on " "heuristics and makes vectorization decision via cost modeling."))

static cl::opt< bool >	ShouldVectorizeHor ("slp-vectorize-hor", cl::init(true), cl::Hidden, cl::desc("Attempt to vectorize horizontal reductions"))

static cl::opt< bool >	ShouldStartVectorizeHorAtStore ("slp-vectorize-hor-store", cl::init(false), cl::Hidden, cl::desc("Attempt to vectorize horizontal reductions feeding into a store"))

static cl::opt< int >	MaxVectorRegSizeOption ("slp-max-reg-size", cl::init(128), cl::Hidden, cl::desc("Attempt to vectorize for this register size in bits"))

static cl::opt< unsigned >	MaxVFOption ("slp-max-vf", cl::init(0), cl::Hidden, cl::desc("Maximum SLP vectorization factor (0=unlimited)"))

static cl::opt< int >	ScheduleRegionSizeBudget ("slp-schedule-budget", cl::init(100000), cl::Hidden, cl::desc("Limit the size of the SLP scheduling region per block"))
	Limits the size of scheduling regions in a block.

static cl::opt< int >	MinVectorRegSizeOption ("slp-min-reg-size", cl::init(128), cl::Hidden, cl::desc("Attempt to vectorize for this register size in bits"))

static cl::opt< unsigned >	RecursionMaxDepth ("slp-recursion-max-depth", cl::init(12), cl::Hidden, cl::desc("Limit the recursion depth when building a vectorizable tree"))

static cl::opt< unsigned >	MinTreeSize ("slp-min-tree-size", cl::init(3), cl::Hidden, cl::desc("Only vectorize small trees if they are fully vectorizable"))

static cl::opt< int >	LookAheadMaxDepth ("slp-max-look-ahead-depth", cl::init(2), cl::Hidden, cl::desc("The maximum look-ahead depth for operand reordering scores"))

static cl::opt< int >	RootLookAheadMaxDepth ("slp-max-root-look-ahead-depth", cl::init(2), cl::Hidden, cl::desc("The maximum look-ahead depth for searching best rooting option"))

static cl::opt< unsigned >	MinProfitableStridedLoads ("slp-min-strided-loads", cl::init(2), cl::Hidden, cl::desc("The minimum number of loads, which should be considered strided, " "if the stride is > 1 or is runtime value"))

static cl::opt< unsigned >	MaxProfitableLoadStride ("slp-max-stride", cl::init(8), cl::Hidden, cl::desc("The maximum stride, considered to be profitable."))

static cl::opt< bool >	ViewSLPTree ("view-slp-tree", cl::Hidden, cl::desc("Display the SLP trees with Graphviz"))

static cl::opt< bool >	VectorizeNonPowerOf2 ("slp-vectorize-non-power-of-2", cl::init(false), cl::Hidden, cl::desc("Try to vectorize with non-power-of-2 number of elements."))

static const unsigned	AliasedCheckLimit = 10

static constexpr int	UsesLimit = 64

static const unsigned	MaxMemDepDistance = 160

static const int	MinScheduleRegionSize = 16
	If the ScheduleRegionSizeBudget is exhausted, we allow small scheduling regions to be handled.

static const unsigned	MaxPHINumOperands = 128
	Maximum allowed number of operands in the PHI nodes.

◆ DEBUG_TYPE

◆ SV_NAME

#define SV_NAME "slp-vectorizer"

◆ addMask()

◆ allConstant()

◆ allSameBlock()

◆ allSameType()

◆ areCompatibleCmpOps()

◆ arePointersCompatible()

◆ areTwoInsertFromSameBuildVector()

◆ buildIntrinsicArgTypes()

◆ buildUseMask()

◆ calculateRtStride()

Checks if the provided list of pointers Pointers represents the strided pointers for type ElemTy.

If they are not, std::nullopt is returned. Otherwise, if Inst is not specified, just initialized optional value is returned to show that the pointers represent strided pointers. If Inst specified, the runtime stride is materialized before the given Inst.

Returns

std::nullopt if the pointers are not pointers with the runtime stride, nullptr or actual stride value, otherwise.

Definition at line 4820 of file SLPVectorizer.cpp.

References llvm::SmallVectorImpl< T >::clear(), DL, llvm::SCEVExpander::expandCodeFor(), llvm::ScalarEvolution::getAddExpr(), llvm::ScalarEvolution::getConstant(), llvm::ScalarEvolution::getMinusSCEV(), llvm::ScalarEvolution::getMulExpr(), llvm::ScalarEvolution::getSCEV(), llvm::SCEV::getType(), llvm::ScalarEvolution::getUDivExactExpr(), llvm::SCEV::isNonConstantNegative(), llvm::SCEV::isZero(), Ptr, llvm::SmallVectorTemplateBase< T, bool >::push_back(), llvm::SmallVectorImpl< T >::resize(), llvm::ArrayRef< T >::size(), llvm::SmallVectorBase< Size_T >::size(), and Size.

Referenced by llvm::slpvectorizer::BoUpSLP::canVectorizeLoads().

◆ calculateShufflevectorMask()

Returns

a shufflevector mask which is used to vectorize shufflevectors e.g., %5 = shufflevector <8 x i16> %3, <8 x i16> poison, <4 x i32> <i32 0, i32 1, i32 2, i32 3> %6 = shufflevector <8 x i16> %3, <8 x i16> poison, <4 x i32> <i32 4, i32 5, i32 6, i32 7> %7 = shufflevector <8 x i16> %4, <8 x i16> poison, <4 x i32> <i32 0, i32 1, i32 2, i32 3> %8 = shufflevector <8 x i16> %4, <8 x i16> poison, <4 x i32> <i32 4, i32 5, i32 6, i32 7> the result is <0, 1, 2, 3, 12, 13, 14, 15, 16, 17, 18, 19, 28, 29, 30, 31>

Definition at line 391 of file SLPVectorizer.cpp.

References assert(), llvm::ArrayRef< T >::front(), getShufflevectorNumGroups(), and llvm::PoisonMaskElem.

◆ checkTreeSizes()

Checks if the quadratic mean deviation is less than 90% of the mean size.

Definition at line 18683 of file SLPVectorizer.cpp.

◆ clusterSortPtrAccesses()

Definition at line 5355 of file SLPVectorizer.cpp.

References llvm::all_of(), llvm::any_of(), assert(), llvm::sampleprof::Base, llvm::SmallVectorImpl< T >::clear(), llvm::SmallPtrSetImpl< PtrType >::contains(), llvm::Depth, llvm::ArrayRef< T >::drop_front(), llvm::enumerate(), llvm::MapVector< KeyT, ValueT, MapType, VectorType >::find(), llvm::MapVector< KeyT, ValueT, MapType, VectorType >::front(), llvm::ArrayRef< T >::front(), llvm::getUnderlyingObject(), llvm::SmallPtrSetImpl< PtrType >::insert(), llvm::isPointerTy(), P, Ptr, llvm::SmallVectorTemplateBase< T, bool >::push_back(), RecursionMaxDepth, llvm::ArrayRef< T >::size(), llvm::MapVector< KeyT, ValueT, MapType, VectorType >::size(), llvm::SmallVectorBase< Size_T >::size(), llvm::stable_sort(), llvm::MapVector< KeyT, ValueT, MapType, VectorType >::try_emplace(), X, and Y.

Referenced by llvm::slpvectorizer::BoUpSLP::findPartiallyOrderedLoads().

◆ combineOrders()

◆ compareCmp()

template<bool IsCompatibility>

Compare two cmp instructions.

If IsCompatibility is true, function returns true if 2 cmps have same/swapped predicates and mos compatible corresponding operands. If IsCompatibility is false, function implements strict weak ordering relation between two cmp instructions, returning true if the first instruction is "less" than the second, i.e. its predicate is less than the predicate of the second or the operands IDs are less than the operands IDs of the second cmp instruction.

Definition at line 21318 of file SLPVectorizer.cpp.

References assert(), llvm::DomTreeNodeBase< NodeT >::getDFSNumIn(), llvm::DominatorTreeBase< NodeT, IsPostDom >::getNode(), llvm::User::getOperand(), getSameOpcode(), llvm::Type::getScalarSizeInBits(), llvm::CmpInst::getSwappedPredicate(), llvm::Value::getType(), llvm::Type::getTypeID(), llvm::Value::getValueID(), I, and isValidElementType().

◆ computeCommonAlignment()

◆ createExtractVector()

◆ createInsertVector()

Correctly creates insert_subvector, checking that the index is multiple of the subvectors length.

Otherwise, generates shuffle using Generator or using default shuffle.

Definition at line 4965 of file SLPVectorizer.cpp.

◆ DEBUG_COUNTER()

DEBUG_COUNTER	(	VectorizedGraphs	,
"slp-vectorized"	,
"Controls which SLP graphs should be vectorized."
)

◆ doesInTreeUserNeedToExtract()

◆ dumpOrder()

◆ findBuildAggregate()

Recognize construction of vectors like ra = insertelement <4 x float> poison, float s0, i32 0 rb = insertelement <4 x float> ra, float s1, i32 1 rc = insertelement <4 x float> rb, float s2, i32 2 rd = insertelement <4 x float> rc, float s3, i32 3 starting from the last insertelement or insertvalue instruction.

Also recognize homogeneous aggregates like {<2 x float>, <2 x float>}, {{float, float}, {float, float}}, [2 x {float, float}] and so on. See llvm/test/Transforms/SLPVectorizer/X86/pr42022.ll for examples.

Assume LastInsertInst is of InsertElementInst or InsertValueInst type.

Returns

true if it matches.

Definition at line 20894 of file SLPVectorizer.cpp.

References assert(), llvm::SmallVectorBase< Size_T >::empty(), llvm::erase(), findBuildAggregate_rec(), getAggregateSize(), llvm::SmallVectorImpl< T >::resize(), and llvm::SmallVectorBase< Size_T >::size().

◆ findBuildAggregate_rec()

◆ fixupOrderingIndices()

Order may have elements assigned special value (size) which is out of bounds.

Such indices only appear on places which correspond to undef values (see canReuseExtract for details) and used in order to avoid undef values have effect on operands ordering. The first loop below simply finds all unused indices and then the next loop nest assigns these indices for undef values positions. As an example below Order has two undef positions and they have assigned values 3 and 7 respectively: before: 6 9 5 4 9 2 1 0 after: 6 3 5 4 7 2 1 0

Definition at line 1185 of file SLPVectorizer.cpp.

References assert(), llvm::SmallBitVector::count(), llvm::SmallBitVector::find_first(), llvm::SmallBitVector::find_next(), I, Idx, llvm::SmallBitVector::none(), llvm::SmallBitVector::reset(), llvm::SmallBitVector::set(), and llvm::ArrayRef< T >::size().

Referenced by llvm::slpvectorizer::BoUpSLP::getReorderingData(), llvm::slpvectorizer::BoUpSLP::reorderBottomToTop(), reorderOrder(), and llvm::slpvectorizer::BoUpSLP::reorderTopToBottom().

◆ gatherPossiblyVectorizableLoads()

Tries to find subvector of loads and builds new vector of only loads if can be profitable.

Definition at line 6775 of file SLPVectorizer.cpp.

References llvm::any_of(), assert(), llvm::ArrayRef< T >::begin(), llvm::bit_ceil(), llvm::SetVector< T, Vector, Set, N >::contains(), llvm::SmallSet< T, N, C >::contains(), llvm::SmallPtrSetImpl< PtrType >::contains(), llvm::Data, DL, llvm::SmallVectorImpl< T >::emplace_back(), llvm::ArrayRef< T >::empty(), llvm::SetVector< T, Vector, Set, N >::empty(), llvm::ArrayRef< T >::end(), llvm::enumerate(), llvm::find_if(), llvm::ArrayRef< T >::front(), llvm::ilist_detail::node_parent_access< NodeTy, ParentTy >::getParent(), llvm::LoadInst::getPointerOperand(), llvm::getPointersDiff(), llvm::Value::getType(), llvm::getUnderlyingObject(), getValueType(), llvm::has_single_bit(), Idx, llvm::SmallSet< T, N, C >::insert(), llvm::SetVector< T, Vector, Set, N >::insert(), llvm::SmallPtrSetImpl< PtrType >::insert(), isValidElementType(), llvm::Offset, P, RecursionMaxDepth, llvm::ArrayRef< T >::size(), and llvm::zip().

Referenced by llvm::slpvectorizer::BoUpSLP::transformNodes().

◆ generateKeySubkey()

Generates key/subkey pair for the given value to provide effective sorting of the values and better detection of the vectorizable values sequences.

The keys/subkeys can be used for better sorting of the values themselves (keys) and in values subgroups (subkeys).

Definition at line 7403 of file SLPVectorizer.cpp.

References llvm::SmallBitVector::all(), generateKeySubkey(), llvm::CmpInst::getInversePredicate(), llvm::VFDatabase::getMappings(), llvm::CmpInst::getSwappedPredicate(), llvm::getVectorIntrinsicIDForCall(), llvm::hash_combine(), llvm::hash_value(), I, llvm::Instruction::isIntDivRem(), llvm::isTriviallyVectorizable(), isUndefVector(), isValidForAlternation(), and isVectorLikeInstWithConstOps().

Referenced by generateKeySubkey().

◆ getAggregateSize()

◆ getAltInstrMask()

◆ getElementIndex()

Returns

inserting or extracting index of InsertElement, ExtractElement or InsertValue instruction, using Offset as base offset for index.

std::nullopt if the index is not an immediate.

Definition at line 568 of file SLPVectorizer.cpp.

References I, IV, and llvm::Offset.

◆ getExtractIndex()

◆ getFloorFullVectorNumberOfElements()

◆ getFullVectorNumberOfElements()

◆ getGEPCosts()

Calculate the scalar and the vector costs from vectorizing set of GEPs.

Definition at line 9503 of file SLPVectorizer.cpp.

References llvm::all_of(), CostKind, llvm::ArrayRef< T >::end(), llvm::find_if(), llvm::TargetTransformInfo::getGEPCost(), llvm::TargetTransformInfo::getPointersChainCost(), Ptr, llvm::SmallVectorTemplateBase< T, bool >::push_back(), llvm::SmallVectorBase< Size_T >::size(), and llvm::TargetTransformInfo::TCC_Free.

Referenced by llvm::slpvectorizer::BoUpSLP::canVectorizeLoads().

◆ getInsertExtractIndex()

◆ getLocation()

◆ getNonPhiOperand()

Returns the first operand of I that does not match Phi.

If operand is not an instruction it returns nullptr.

Definition at line 21018 of file SLPVectorizer.cpp.

References I, and matchRdxBop().

◆ getNumElements()

Returns

the number of elements for Ty.

Definition at line 254 of file SLPVectorizer.cpp.

References assert().

Referenced by llvm::MachineIRBuilder::buildSplatBuildVector(), cacheDIVar(), llvm::slpvectorizer::BoUpSLP::canVectorizeLoads(), llvm::slpvectorizer::BoUpSLP::computeMinimumValueSizes(), containsUndefinedElement(), CreateGCRelocates(), expandAbs(), expandCrossIntrinsic(), expandExpIntrinsic(), expandLogIntrinsic(), extractVector(), getAltInstrMask(), llvm::X86TTIImpl::getScalarizationOverhead(), llvm::BasicTTIImplBase< T >::getTypeBasedIntrinsicInstrCost(), llvm::SystemZTTIImpl::getVectorTruncCost(), getWidenedType(), insertVector(), llvm::X86TTIImpl::isLegalMaskedExpandLoad(), llvm::X86TTIImpl::isLegalMaskedLoad(), llvm::ShuffleVectorInst::isValidOperands(), isVectorPromotionViableForSlice(), isZero(), llvm::RewriteStatepointsForGC::runOnFunction(), llvm::InstCombinerImpl::SimplifyDemandedVectorElts(), simplifyX86pack(), and upgradeX86IntrinsicCall().

◆ getNumElems()

◆ getPartNumElems()

◆ getRdxKind()

◆ getReductionInstr()

◆ getSameOpcode()

Returns

analysis of the Instructions in VL described in InstructionsState, the Opcode that we suppose the whole list could be vectorized even if its structure is diverse.

Definition at line 909 of file SLPVectorizer.cpp.

References llvm::all_of(), assert(), llvm::CmpInst::BAD_ICMP_PREDICATE, llvm::SetVector< T, Vector, Set, N >::contains(), llvm::SmallVectorBase< Size_T >::empty(), llvm::ArrayRef< T >::end(), llvm::find_if(), llvm::SmallVectorTemplateCommon< T, typename >::front(), llvm::CallBase::getBundleOperandsStartIndex(), llvm::CallBase::getCalledFunction(), llvm::VFDatabase::getMappings(), llvm::Instruction::getOpcode(), llvm::User::getOperand(), llvm::CmpInst::getSwappedPredicate(), llvm::Value::getType(), llvm::getVectorIntrinsicIDForCall(), llvm::CallBase::hasOperandBundles(), I, llvm::SetVector< T, Vector, Set, N >::insert(), isCmpSameOrSwapped(), llvm::isTriviallyVectorizable(), isValidForAlternation(), isVectorLikeInstWithConstOps(), llvm::User::op_begin(), llvm::ArrayRef< T >::size(), llvm::SetVector< T, Vector, Set, N >::size(), and llvm::SmallVectorBase< Size_T >::size().

Referenced by areCompatibleCmpOps(), arePointersCompatible(), compareCmp(), llvm::slpvectorizer::BoUpSLP::LookAheadHeuristics::getShallowScore(), llvm::slpvectorizer::BoUpSLP::VLOperands::reorder(), and llvm::slpvectorizer::BoUpSLP::transformNodes().

◆ getShuffleCost()

◆ getShufflevectorNumGroups()

Returns

the number of groups of shufflevector A group has the following features

All of value in a group are shufflevector.
The mask of all shufflevector is isExtractSubvectorMask.
The mask of all shufflevector uses all of the elements of the source. e.g., it is 1 group (%0) %1 = shufflevector <16 x i8> %0, <16 x i8> poison, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7> %2 = shufflevector <16 x i8> %0, <16 x i8> poison, <8 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15> it is 2 groups (%3 and %4) %5 = shufflevector <8 x i16> %3, <8 x i16> poison, <4 x i32> <i32 0, i32 1, i32 2, i32 3> %6 = shufflevector <8 x i16> %3, <8 x i16> poison, <4 x i32> <i32 4, i32 5, i32 6, i32 7> %7 = shufflevector <8 x i16> %4, <8 x i16> poison, <4 x i32> <i32 0, i32 1, i32 2, i32 3> %8 = shufflevector <8 x i16> %4, <8 x i16> poison, <4 x i32> <i32 4, i32 5, i32 6, i32 7> it is 0 group %12 = shufflevector <8 x i16> %10, <8 x i16> poison, <4 x i32> <i32 0, i32 1, i32 2, i32 3> %13 = shufflevector <8 x i16> %11, <8 x i16> poison, <4 x i32> <i32 0, i32 1, i32 2, i32 3>

Definition at line 339 of file SLPVectorizer.cpp.

References llvm::SmallBitVector::all(), llvm::all_of(), assert(), llvm::ArrayRef< T >::empty(), llvm::ArrayRef< T >::front(), I, llvm::SmallBitVector::set(), llvm::ArrayRef< T >::size(), and llvm::ArrayRef< T >::slice().

Referenced by calculateShufflevectorMask().

◆ getValueType()

static Type * getValueType ( Value * V)	static

Returns the type of the given value/instruction V.

If it is store, returns the type of its value operand, for Cmp - the types of the compare operands and for insertelement - the type os the inserted operand. Otherwise, just the type of the value is returned.

Definition at line 243 of file SLPVectorizer.cpp.

Referenced by CollectOpsToWiden(), combineArithReduction(), combineConcatVectorOps(), combineMulToPMULDQ(), combineTargetShuffle(), combineToHorizontalAddSub(), CompactSwizzlableVector(), detectPMADDUBSW(), FoldBUILD_VECTOR(), foldCONCAT_VECTORS(), gatherPossiblyVectorizableLoads(), llvm::Function::getFunctionType(), llvm::RISCVTTIImpl::getIntImmCostInst(), getTestBitOperand(), LLVMGlobalGetValueType(), matchPMADDWD(), matchPMADDWD_2(), narrowVectorSelect(), performCONCAT_VECTORSCombine(), and llvm::SelectionDAGISel::SelectInlineAsmMemoryOperands().

◆ getVectorCallCosts()

Definition at line 9025 of file SLPVectorizer.cpp.

References llvm::CallBase::args(), Arguments, llvm::VFShape::get(), llvm::TargetTransformInfo::getCallInstrCost(), llvm::ElementCount::getFixed(), llvm::CallBase::getFunctionType(), llvm::TargetTransformInfo::getIntrinsicInstrCost(), llvm::FixedVectorType::getNumElements(), llvm::getVectorIntrinsicIDForCall(), llvm::VFDatabase::getVectorizedFunction(), llvm::CallBase::isNoBuiltin(), and llvm::TargetTransformInfo::TCK_RecipThroughput.

◆ getWidenedType()

Returns

the vector type of ScalarTy based on vectorization factor.

Definition at line 263 of file SLPVectorizer.cpp.

References llvm::FixedVectorType::get(), getNumElements(), and llvm::Type::getScalarType().

◆ isAlternateInstruction()

◆ isCmpSameOrSwapped()

◆ isCommutative()

◆ isConstant()

◆ isFirstInsertElement()

◆ isFixedVectorShuffle()

Checks if the vector of instructions can be represented as a shuffle, like: x0 = extractelement <4 x i8> x, i32 0 x3 = extractelement <4 x i8> x, i32 3 y1 = extractelement <4 x i8> y, i32 1 y2 = extractelement <4 x i8> y, i32 2 x0x0 = mul i8 x0, x0 x3x3 = mul i8 x3, x3 y1y1 = mul i8 y1, y1 y2y2 = mul i8 y2, y2 ins1 = insertelement <4 x i8> poison, i8 x0x0, i32 0 ins2 = insertelement <4 x i8> ins1, i8 x3x3, i32 1 ins3 = insertelement <4 x i8> ins2, i8 y1y1, i32 2 ins4 = insertelement <4 x i8> ins3, i8 y2y2, i32 3 ret <4 x i8> ins4 can be transformed into: %1 = shufflevector <4 x i8> x, <4 x i8> y, <4 x i32> <i32 0, i32 3, i32 5, i32 6> %2 = mul <4 x i8> %1, %1 ret <4 x i8> %2 Mask will return the Shuffle Mask equivalent to the extracted elements.

TODO: Can we split off and reuse the shuffle mask detection from ShuffleVectorInst/getShuffleCost?

Definition at line 706 of file SLPVectorizer.cpp.

References llvm::any_of(), llvm::ArrayRef< T >::begin(), llvm::ArrayRef< T >::end(), llvm::find_if(), I, Idx, llvm::isGuaranteedNotToBePoison(), isUndefVector(), llvm::PoisonMaskElem, Select, llvm::ArrayRef< T >::size(), Size, llvm::TargetTransformInfo::SK_PermuteSingleSrc, llvm::TargetTransformInfo::SK_PermuteTwoSrc, llvm::TargetTransformInfo::SK_Select, and Unknown.

◆ isLoadCombineCandidateImpl()

Definition at line 12009 of file SLPVectorizer.cpp.

References llvm::dbgs(), llvm::IntegerType::get(), llvm::Value::getContext(), llvm::Type::getIntegerBitWidth(), llvm::TargetTransformInfo::isTypeLegal(), LLVM_DEBUG, llvm::PatternMatch::m_APInt(), llvm::PatternMatch::m_Or(), llvm::PatternMatch::m_Shl(), llvm::PatternMatch::m_Value(), llvm::PatternMatch::m_ZExt(), llvm::PatternMatch::match(), and llvm::APInt::urem().

Referenced by llvm::slpvectorizer::BoUpSLP::isLoadCombineCandidate(), and llvm::slpvectorizer::BoUpSLP::isLoadCombineReductionCandidate().

◆ isReductionCandidate()

◆ isRepeatedNonIdentityClusteredMask()

◆ isReverseOrder()

◆ isSimple()

Returns

True if the instruction is not a volatile or atomic load/store.

Definition at line 1137 of file SLPVectorizer.cpp.

References I, and MI.

Referenced by llvm::DOTGraphTraits< DOTFuncInfo * >::getNodeLabel(), llvm::DOTGraphTraits< MachineBlockFrequencyInfo * >::getNodeLabel(), llvm::DOTGraphTraits< const MachineFunction * >::getNodeLabel(), llvm::DOTGraphTraits< DOTMachineFuncInfo * >::getNodeLabel(), llvm::DOTGraphTraits< DomTreeNode * >::getNodeLabel(), llvm::DOTGraphTraits< RegionNode * >::getNodeLabel(), llvm::X86TargetLowering::shouldReduceLoadWidth(), tryToFoldExtOfMaskedLoad(), llvm::MBFIWrapper::view(), and llvm::MachineBlockFrequencyInfo::view().

◆ isSplat()

Returns

True if all of the values in VL are identical or some of them are UndefValue.

Definition at line 493 of file SLPVectorizer.cpp.

◆ isUndefVector()

template<bool IsPoisonOnly = false>

◆ isValidElementType()

static bool isValidElementType ( Type * Ty)	static

Predicate for the element types that the SLP vectorizer supports.

The most important thing to filter here are types which are invalid in LLVM vectors. We also filter target specific types which have absolutely no meaningful vectorization path such as x86_fp80 and ppc_f128. This just avoids spending time checking the cost model and realizing that they will be inevitably scalarized.

Definition at line 231 of file SLPVectorizer.cpp.

References llvm::Type::getScalarType(), llvm::Type::isPPC_FP128Ty(), llvm::Type::isX86_FP80Ty(), and SLPReVec.

Referenced by llvm::slpvectorizer::BoUpSLP::canMapToVector(), compareCmp(), llvm::slpvectorizer::BoUpSLP::findReusedOrderedScalars(), gatherPossiblyVectorizableLoads(), getFloorFullVectorNumberOfElements(), getFullVectorNumberOfElements(), llvm::slpvectorizer::BoUpSLP::LookAheadHeuristics::getShallowScore(), and llvm::hasFullVectorsOrPowerOf2().

◆ isValidForAlternation()

◆ isVectorLikeInstWithConstOps()

static bool isVectorLikeInstWithConstOps ( Value * V)	static

◆ matchRdxBop()

◆ needToScheduleSingleInstruction()

◆ performExtractsShuffleAction()

Does the analysis of the provided shuffle masks and performs the requested actions on the vectors with the given shuffle masks.

It tries to do it in several steps.

If the Base vector is not undef vector, resizing the very first mask to have common VF and perform action for 2 input vectors (including non-undef Base). Other shuffle masks are combined with the resulting after the 1 stage and processed as a shuffle of 2 elements.
If the Base is undef vector and have only 1 shuffle mask, perform the action only for 1 vector with the given mask, if it is not the identity mask.
If > 2 masks are used, perform the remaining shuffle actions for 2 vectors, combing the masks properly between the steps.

Definition at line 12352 of file SLPVectorizer.cpp.

References llvm::SmallBitVector::all(), assert(), llvm::sampleprof::Base, buildUseMask(), I, Idx, isUndefVector(), llvm::PoisonMaskElem, and llvm::SmallBitVector::test().

◆ propagateMetadata()

◆ reorderOrder()

Reorders the given Order according to the given Mask.

Order - is the original order of the scalars. Procedure transforms the provided order in accordance with the given Mask. If the resulting Order is just an identity order, Order is cleared.

Definition at line 4563 of file SLPVectorizer.cpp.

References llvm::all_of(), assert(), llvm::SmallVectorImpl< T >::assign(), llvm::SmallVectorTemplateCommon< T, typename >::begin(), llvm::SmallVectorImpl< T >::clear(), llvm::Data, llvm::SmallVectorBase< Size_T >::empty(), llvm::SmallVectorTemplateCommon< T, typename >::end(), llvm::enumerate(), fixupOrderingIndices(), I, llvm::inversePermutation(), llvm::ShuffleVectorInst::isIdentityMask(), llvm::PoisonMaskElem, reorderReuses(), llvm::SmallVectorImpl< T >::resize(), and llvm::SmallVectorImpl< T >::swap().

Referenced by llvm::slpvectorizer::BoUpSLP::getReorderingData(), llvm::slpvectorizer::BoUpSLP::reorderBottomToTop(), and llvm::slpvectorizer::BoUpSLP::reorderTopToBottom().

◆ reorderReuses()

Reorders the given Reuses mask according to the given Mask.

Reuses contains original mask for the scalars reused in the node. Procedure transform this mask in accordance with the given Mask.

Definition at line 4549 of file SLPVectorizer.cpp.

References assert(), llvm::SmallVectorTemplateCommon< T, typename >::begin(), E, llvm::SmallVectorTemplateCommon< T, typename >::end(), I, llvm::PoisonMaskElem, llvm::SmallVectorBase< Size_T >::size(), and llvm::SmallVectorImpl< T >::swap().

Referenced by llvm::slpvectorizer::BoUpSLP::reorderBottomToTop(), and reorderOrder().

◆ shortBundleName()

static std::string shortBundleName ( ArrayRef< Value * > VL, int Idx = -1 )	static

◆ STATISTIC()

◆ transformScalarShuffleIndiciesToVector()

static void transformScalarShuffleIndiciesToVector ( unsigned VecTyNumElements, SmallVectorImpl< int > & Mask )	static

◆ tryGetSecondaryReductionRoot()

We could have an initial reduction that is not an add.

r *= v1 + v2 + v3 + v4 In such a case start looking for a tree rooted in the first '+'. \Returns the new root if found, which may be nullptr if not an instruction.

Definition at line 21000 of file SLPVectorizer.cpp.

References assert(), llvm::User::getOperand(), LHS, and RHS.

◆ tryToVectorizeSequence()

\Returns the minimum number of elements that we will attempt to vectorize.

Definition at line 21199 of file SLPVectorizer.cpp.

References llvm::SmallVectorTemplateCommon< T, typename >::begin(), llvm::SmallVectorImpl< T >::clear(), llvm::dbgs(), llvm::SmallVectorBase< Size_T >::empty(), llvm::SmallVectorTemplateCommon< T, typename >::end(), End, llvm::SmallVectorTemplateCommon< T, typename >::front(), I, LLVM_DEBUG, llvm::SmallVectorTemplateBase< T, bool >::push_back(), llvm::SmallVectorBase< Size_T >::size(), llvm::stable_sort(), and llvm::SmallVectorImpl< T >::swap().

◆ AliasedCheckLimit

◆ LookAheadMaxDepth

cl::opt< int > LookAheadMaxDepth("slp-max-look-ahead-depth", cl::init(2), cl::Hidden, cl::desc("The maximum look-ahead depth for operand reordering scores")) ( "slp-max-look-ahead-depth" , cl::init(2) , cl::Hidden , cl::desc("The maximum look-ahead depth for operand reordering scores") )	static

◆ MaxMemDepDistance

◆ MaxPHINumOperands

Maximum allowed number of operands in the PHI nodes.

Definition at line 222 of file SLPVectorizer.cpp.

◆ MaxProfitableLoadStride

cl::opt< unsigned > MaxProfitableLoadStride("slp-max-stride", cl::init(8), cl::Hidden, cl::desc("The maximum stride, considered to be profitable.")) ( "slp-max-stride" , cl::init(8) , cl::Hidden , cl::desc("The maximum stride, considered to be profitable.") )	static

◆ MaxVectorRegSizeOption

cl::opt< int > MaxVectorRegSizeOption("slp-max-reg-size", cl::init(128), cl::Hidden, cl::desc("Attempt to vectorize for this register size in bits")) ( "slp-max-reg-size" , cl::init(128) , cl::Hidden , cl::desc("Attempt to vectorize for this register size in bits") )	static

◆ MaxVFOption

cl::opt< unsigned > MaxVFOption("slp-max-vf", cl::init(0), cl::Hidden, cl::desc("Maximum SLP vectorization factor (0=unlimited)")) ( "slp-max-vf" , cl::init(0) , cl::Hidden , cl::desc("Maximum SLP vectorization factor (0=unlimited)") )	static

◆ MinProfitableStridedLoads

cl::opt< unsigned > MinProfitableStridedLoads("slp-min-strided-loads", cl::init(2), cl::Hidden, cl::desc("The minimum number of loads, which should be considered strided, " "if the stride is > 1 or is runtime value")) ( "slp-min-strided-loads" , cl::init(2) , cl::Hidden , cl::desc("The minimum number of loads, which should be considered strided, " "if the stride is > 1 or is runtime value") )	static

◆ MinScheduleRegionSize

const int MinScheduleRegionSize = 16	static

If the ScheduleRegionSizeBudget is exhausted, we allow small scheduling regions to be handled.

Definition at line 219 of file SLPVectorizer.cpp.

◆ MinTreeSize

cl::opt< unsigned > MinTreeSize("slp-min-tree-size", cl::init(3), cl::Hidden, cl::desc("Only vectorize small trees if they are fully vectorizable")) ( "slp-min-tree-size" , cl::init(3) , cl::Hidden , cl::desc("Only vectorize small trees if they are fully vectorizable") )	static

◆ MinVectorRegSizeOption

cl::opt< int > MinVectorRegSizeOption("slp-min-reg-size", cl::init(128), cl::Hidden, cl::desc("Attempt to vectorize for this register size in bits")) ( "slp-min-reg-size" , cl::init(128) , cl::Hidden , cl::desc("Attempt to vectorize for this register size in bits") )	static

◆ RecursionMaxDepth

◆ RootLookAheadMaxDepth

cl::opt< int > RootLookAheadMaxDepth("slp-max-root-look-ahead-depth", cl::init(2), cl::Hidden, cl::desc("The maximum look-ahead depth for searching best rooting option")) ( "slp-max-root-look-ahead-depth" , cl::init(2) , cl::Hidden , cl::desc("The maximum look-ahead depth for searching best rooting option") )	static

◆ RunSLPVectorization

cl::opt< bool > RunSLPVectorization("vectorize-slp", cl::init(true), cl::Hidden, cl::desc("Run the SLP vectorization passes")) ( "vectorize-slp" , cl::init(true) , cl::Hidden , cl::desc("Run the SLP vectorization passes") )	static

◆ ScheduleRegionSizeBudget

cl::opt< int > ScheduleRegionSizeBudget("slp-schedule-budget", cl::init(100000), cl::Hidden, cl::desc("Limit the size of the SLP scheduling region per block")) ( "slp-schedule-budget" , cl::init(100000) , cl::Hidden , cl::desc("Limit the size of the SLP scheduling region per block") )	static

Limits the size of scheduling regions in a block.

It avoid long compile times for very large blocks where vector instructions are spread over a wide range. This limit is way higher than needed by real-world functions.

◆ ShouldStartVectorizeHorAtStore

cl::opt< bool > ShouldStartVectorizeHorAtStore("slp-vectorize-hor-store", cl::init(false), cl::Hidden, cl::desc( "Attempt to vectorize horizontal reductions feeding into a store")) ( "slp-vectorize-hor-store" , cl::init(false) , cl::Hidden , cl::desc( "Attempt to vectorize horizontal reductions feeding into a store") )	static

◆ ShouldVectorizeHor

cl::opt< bool > ShouldVectorizeHor("slp-vectorize-hor", cl::init(true), cl::Hidden, cl::desc("Attempt to vectorize horizontal reductions")) ( "slp-vectorize-hor" , cl::init(true) , cl::Hidden , cl::desc("Attempt to vectorize horizontal reductions") )	static

◆ SLPCostThreshold

cl::opt< int > SLPCostThreshold("slp-threshold", cl::init(0), cl::Hidden, cl::desc("Only vectorize if you gain more than this " "number ")) ( "slp-threshold" , cl::init(0) , cl::Hidden , cl::desc("Only vectorize if you gain more than this " "number ") )	static

◆ SLPReVec

cl::opt< bool > SLPReVec("slp-revec", cl::init(false), cl::Hidden, cl::desc("Enable vectorization for wider vector utilization")) ( "slp-revec" , cl::init(false) , cl::Hidden , cl::desc("Enable vectorization for wider vector utilization") )	static

◆ SLPSkipEarlyProfitabilityCheck

cl::opt< bool > SLPSkipEarlyProfitabilityCheck("slp-skip-early-profitability-check", cl::init(false), cl::Hidden, cl::desc("When true, SLP vectorizer bypasses profitability checks based on " "heuristics and makes vectorization decision via cost modeling.")) ( "slp-skip-early-profitability-check" , cl::init(false) , cl::Hidden , cl::desc("When true, SLP vectorizer bypasses profitability checks based on " "heuristics and makes vectorization decision via cost modeling.") )

static

◆ UsesLimit

constexpr int UsesLimit = 64	staticconstexpr

◆ VectorizeNonPowerOf2

cl::opt< bool > VectorizeNonPowerOf2("slp-vectorize-non-power-of-2", cl::init(false), cl::Hidden, cl::desc("Try to vectorize with non-power-of-2 number of elements.")) ( "slp-vectorize-non-power-of-2" , cl::init(false) , cl::Hidden , cl::desc("Try to vectorize with non-power-of-2 number of elements.") )	static

◆ ViewSLPTree

cl::opt< bool > ViewSLPTree("view-slp-tree", cl::Hidden, cl::desc("Display the SLP trees with Graphviz")) ( "view-slp-tree" , cl::Hidden , cl::desc("Display the SLP trees with Graphviz") )	static