LoopVectorizationCostModel - estimates the expected speedups due to vectorization. More...

Classes
struct	CallWideningDecision

Public Types
enum	InstWidening { CM_Unknown , CM_Widen , CM_Widen_Reverse , CM_Interleave , CM_GatherScatter , CM_Scalarize , CM_VectorCall , CM_IntrinsicCall }
	Decision that was taken during cost calculation for memory instruction. More...

Public Member Functions
	LoopVectorizationCostModel (ScalarEpilogueLowering SEL, Loop L, PredicatedScalarEvolution &PSE, LoopInfo LI, LoopVectorizationLegality Legal, const TargetTransformInfo &TTI, const TargetLibraryInfo TLI, DemandedBits DB, AssumptionCache AC, OptimizationRemarkEmitter ORE, const Function F, const LoopVectorizeHints Hints, InterleavedAccessInfo &IAI, ProfileSummaryInfo PSI, BlockFrequencyInfo *BFI)
FixedScalableVFPair	computeMaxVF (ElementCount UserVF, unsigned UserIC)
bool	runtimeChecksRequired ()
bool	selectUserVectorizationFactor (ElementCount UserVF)
	Setup cost-based decisions for user vectorization factor.
bool	useMaxBandwidth (TargetTransformInfo::RegisterKind RegKind)
bool	shouldConsiderRegPressureForVF (ElementCount VF)
std::pair< unsigned, unsigned >	getSmallestAndWidestTypes ()
void	setCostBasedWideningDecision (ElementCount VF)
	Memory access instruction may be vectorized in more than one way.
void	setVectorizedCallDecision (ElementCount VF)
	A call may be vectorized in different ways depending on whether we have vectorized variants available and whether the target supports masking.
void	collectValuesToIgnore ()
	Collect values we want to ignore in the cost model.
void	collectElementTypesForWidening ()
	Collect all element types in the loop for which widening is needed.
void	collectInLoopReductions ()
	Split reductions into those that happen in the loop, and those that happen outside.
bool	useOrderedReductions (const RecurrenceDescriptor &RdxDesc) const
	Returns true if we should use strict in-order reductions for the given RdxDesc.
const MapVector< Instruction *, uint64_t > &	getMinimalBitwidths () const
bool	isProfitableToScalarize (Instruction *I, ElementCount VF) const
bool	isUniformAfterVectorization (Instruction *I, ElementCount VF) const
	Returns true if `I` is known to be uniform after vectorization.
bool	isScalarAfterVectorization (Instruction *I, ElementCount VF) const
	Returns true if `I` is known to be scalar after vectorization.
bool	canTruncateToMinimalBitwidth (Instruction *I, ElementCount VF) const
void	setWideningDecision (Instruction *I, ElementCount VF, InstWidening W, InstructionCost Cost)
	Save vectorization decision `W` and `Cost` taken by the cost model for instruction `I` and vector width `VF`.
void	setWideningDecision (const InterleaveGroup< Instruction > *Grp, ElementCount VF, InstWidening W, InstructionCost Cost)
	Save vectorization decision `W` and `Cost` taken by the cost model for interleaving group `Grp` and vector width `VF`.
InstWidening	getWideningDecision (Instruction *I, ElementCount VF) const
	Return the cost model decision for the given instruction `I` and vector width `VF`.
InstructionCost	getWideningCost (Instruction *I, ElementCount VF)
	Return the vectorization cost for the given instruction `I` and vector width `VF`.
void	setCallWideningDecision (CallInst CI, ElementCount VF, InstWidening Kind, Function Variant, Intrinsic::ID IID, std::optional< unsigned > MaskPos, InstructionCost Cost)
CallWideningDecision	getCallWideningDecision (CallInst *CI, ElementCount VF) const
bool	isOptimizableIVTruncate (Instruction *I, ElementCount VF)
	Return True if instruction `I` is an optimizable truncate whose operand is an induction variable.
void	collectInstsToScalarize (ElementCount VF)
	Collects the instructions to scalarize for each predicated instruction in the loop.
void	collectNonVectorizedAndSetWideningDecisions (ElementCount VF)
	Collect values that will not be widened, including Uniforms, Scalars, and Instructions to Scalarize for the given `VF`.
bool	isLegalMaskedStore (Type DataType, Value Ptr, Align Alignment, unsigned AddressSpace) const
	Returns true if the target machine supports masked store operation for the given `DataType` and kind of access to `Ptr`.
bool	isLegalMaskedLoad (Type DataType, Value Ptr, Align Alignment, unsigned AddressSpace) const
	Returns true if the target machine supports masked load operation for the given `DataType` and kind of access to `Ptr`.
bool	isLegalGatherOrScatter (Value *V, ElementCount VF)
	Returns true if the target machine can represent `V` as a masked gather or scatter operation.
bool	canVectorizeReductions (ElementCount VF) const
	Returns true if the target machine supports all of the reduction variables found for the given VF.
bool	isDivRemScalarWithPredication (InstructionCost ScalarCost, InstructionCost SafeDivisorCost) const
	Given costs for both strategies, return true if the scalar predication lowering should be used for div/rem.
bool	isScalarWithPredication (Instruction *I, ElementCount VF) const
	Returns true if `I` is an instruction which requires predication and for which our chosen predication strategy is scalarization (i.e.
bool	isPredicatedInst (Instruction *I) const
	Returns true if `I` is an instruction that needs to be predicated at runtime.
std::pair< InstructionCost, InstructionCost >	getDivRemSpeculationCost (Instruction *I, ElementCount VF) const
	Return the costs for our two available strategies for lowering a div/rem operation which requires speculating at least one lane.
bool	memoryInstructionCanBeWidened (Instruction *I, ElementCount VF)
	Returns true if `I` is a memory instruction with consecutive memory access that can be widened.
bool	interleavedAccessCanBeWidened (Instruction *I, ElementCount VF) const
	Returns true if `I` is a memory instruction in an interleaved-group of memory accesses that can be vectorized with wide vector loads/stores and shuffles.
bool	isAccessInterleaved (Instruction *Instr) const
	Check if `Instr` belongs to any interleaved access group.
const InterleaveGroup< Instruction > *	getInterleavedAccessGroup (Instruction *Instr) const
	Get the interleaved access group that `Instr` belongs to.
bool	requiresScalarEpilogue (bool IsVectorizing) const
	Returns true if we're required to use a scalar epilogue for at least the final iteration of the original loop.
bool	isScalarEpilogueAllowed () const
	Returns true if a scalar epilogue is not allowed due to optsize or a loop hint annotation.
TailFoldingStyle	getTailFoldingStyle (bool IVUpdateMayOverflow=true) const
	Returns the TailFoldingStyle that is best for the current loop.
void	setTailFoldingStyles (bool IsScalableVF, unsigned UserIC)
	Selects and saves TailFoldingStyle for 2 options - if IV update may overflow or not.
bool	foldTailByMasking () const
	Returns true if all loop blocks should be masked to fold tail loop.
std::optional< unsigned >	getMaxSafeElements () const
	Return maximum safe number of elements to be processed per vector iteration, which do not prevent store-load forwarding and are safe with regard to the memory dependencies.
bool	blockNeedsPredicationForAnyReason (BasicBlock *BB) const
	Returns true if the instructions in this block requires predication for any reason, e.g.
bool	foldTailWithEVL () const
	Returns true if VP intrinsics with explicit vector length support should be generated in the tail folded loop.
bool	isInLoopReduction (PHINode *Phi) const
	Returns true if the Phi is part of an inloop reduction.
bool	usePredicatedReductionSelect () const
	Returns true if the predicated reduction select should be used to set the incoming value for the reduction phi.
InstructionCost	getVectorIntrinsicCost (CallInst *CI, ElementCount VF) const
	Estimate cost of an intrinsic call instruction CI if it were vectorized with factor VF.
InstructionCost	getVectorCallCost (CallInst *CI, ElementCount VF) const
	Estimate cost of a call instruction CI if it were vectorized with factor VF.
void	invalidateCostModelingDecisions ()
	Invalidates decisions already taken by the cost model.
InstructionCost	expectedCost (ElementCount VF)
	Returns the expected execution cost.
bool	hasPredStores () const
bool	isEpilogueVectorizationProfitable (const ElementCount VF, const unsigned IC) const
	Returns true if epilogue vectorization is considered profitable, and false otherwise.
InstructionCost	getInstructionCost (Instruction *I, ElementCount VF)
	Returns the execution time cost of an instruction for a given vector width.
std::optional< InstructionCost >	getReductionPatternCost (Instruction I, ElementCount VF, Type VectorTy) const
	Return the cost of instructions in an inloop reduction pattern, if I is part of that pattern.
bool	shouldConsiderInvariant (Value *Op)
	Returns true if `Op` should be considered invariant and if it is trivially hoistable.
std::optional< unsigned >	getVScaleForTuning () const
	Return the value of vscale used for tuning the cost model.

Public Attributes
Loop *	TheLoop
	The loop that we evaluate.
PredicatedScalarEvolution &	PSE
	Predicated scalar evolution analysis.
LoopInfo *	LI
	Loop Info analysis.
LoopVectorizationLegality *	Legal
	Vectorization legality.
const TargetTransformInfo &	TTI
	Vector target information.
const TargetLibraryInfo *	TLI
	Target Library Info.
DemandedBits *	DB
	Demanded bits analysis.
AssumptionCache *	AC
	Assumption cache.
OptimizationRemarkEmitter *	ORE
	Interface to emit optimization remarks.
const Function *	TheFunction
const LoopVectorizeHints *	Hints
	Loop Vectorize Hint.
InterleavedAccessInfo &	InterleaveInfo
	The interleave access information contains groups of interleaved accesses with the same stride and close to each other.
SmallPtrSet< const Value *, 16 >	ValuesToIgnore
	Values to ignore in the cost model.
SmallPtrSet< const Value *, 16 >	VecValuesToIgnore
	Values to ignore in the cost model when VF > 1.
SmallPtrSet< Type *, 16 >	ElementTypesInLoop
	All element types found in the loop.
TTI::TargetCostKind	CostKind
	The kind of cost that we are calculating.
bool	OptForSize
	Whether this loop should be optimized for size based on function attribute or profile information.
FixedScalableVFPair	MaxPermissibleVFWithoutMaxBW
	The highest VF possible for this loop, without using MaxBandwidth.

Friends
class	LoopVectorizationPlanner

Detailed Description

LoopVectorizationCostModel - estimates the expected speedups due to vectorization.

In many cases vectorization is not profitable. This can happen because of a number of reasons. In this class we mainly attempt to predict the expected speedup/slowdowns due to the supported instruction set. We use the TargetTransformInfo to query the different backends for the cost of different operations.

Definition at line 884 of file LoopVectorize.cpp.

Member Enumeration Documentation

◆ InstWidening

enum llvm::LoopVectorizationCostModel::InstWidening

Decision that was taken during cost calculation for memory instruction.

Enumerator
CM_Unknown
CM_Widen
CM_Widen_Reverse
CM_Interleave
CM_GatherScatter
CM_Scalarize
CM_VectorCall
CM_IntrinsicCall

Definition at line 1037 of file LoopVectorize.cpp.

Constructor & Destructor Documentation

◆ LoopVectorizationCostModel()

llvm::LoopVectorizationCostModel::LoopVectorizationCostModel	(	ScalarEpilogueLowering	SEL,
		Loop *	L,
		PredicatedScalarEvolution &	PSE,
		LoopInfo *	LI,
		LoopVectorizationLegality *	Legal,
		const TargetTransformInfo &	TTI,
		const TargetLibraryInfo *	TLI,
		DemandedBits *	DB,
		AssumptionCache *	AC,
		OptimizationRemarkEmitter *	ORE,
		const Function *	F,
		const LoopVectorizeHints *	Hints,
		InterleavedAccessInfo &	IAI,
		ProfileSummaryInfo *	PSI,
		BlockFrequencyInfo *	BFI )

inline

Definition at line 888 of file LoopVectorize.cpp.

References AC, CostKind, DB, F, ForceTargetSupportsScalableVectors, Hints, InterleaveInfo, llvm::IRPass, Legal, LI, OptForSize, ORE, PSE, llvm::shouldOptimizeForSize(), llvm::TargetTransformInfo::TCK_CodeSize, llvm::TargetTransformInfo::TCK_RecipThroughput, TheFunction, TheLoop, TLI, and TTI.

Member Function Documentation

◆ blockNeedsPredicationForAnyReason()

bool llvm::LoopVectorizationCostModel::blockNeedsPredicationForAnyReason ( BasicBlock * BB ) const

inline

Returns true if the instructions in this block requires predication for any reason, e.g.

because tail folding now requires a predicate or because the block in the original loop was predicated.

Definition at line 1378 of file LoopVectorize.cpp.

References foldTailByMasking(), and Legal.

Referenced by collectInstsToScalarize(), and interleavedAccessCanBeWidened().

◆ canTruncateToMinimalBitwidth()

bool llvm::LoopVectorizationCostModel::canTruncateToMinimalBitwidth	(	Instruction *	I,
		ElementCount	VF ) const

inline

Returns: True if instruction I can be truncated to a smaller bitwidth for vectorization factor VF.

Definition at line 1030 of file LoopVectorize.cpp.

References I, isProfitableToScalarize(), isScalarAfterVectorization(), and llvm::ElementCount::isVector().

Referenced by getInstructionCost().

◆ canVectorizeReductions()

bool llvm::LoopVectorizationCostModel::canVectorizeReductions ( ElementCount VF ) const

inline

Returns true if the target machine supports all of the reduction variables found for the given VF.

Definition at line 1214 of file LoopVectorize.cpp.

References llvm::all_of(), and Legal.

◆ collectElementTypesForWidening()

void LoopVectorizationCostModel::collectElementTypesForWidening ( )

Collect all element types in the loop for which widening is needed.

Definition at line 4466 of file LoopVectorize.cpp.

References assert(), llvm::dyn_cast(), ElementTypesInLoop, llvm::RecurrenceDescriptor::getRecurrenceKind(), llvm::RecurrenceDescriptor::getRecurrenceType(), I, llvm::isa(), Legal, PreferInLoopReductions, T, TheLoop, TTI, useOrderedReductions(), and ValuesToIgnore.

Referenced by processLoopInVPlanNativePath().

◆ collectInLoopReductions()

void LoopVectorizationCostModel::collectInLoopReductions ( )

Split reductions into those that happen in the loop, and those that happen outside.

In loop reductions are collected into InLoopReductions.

Definition at line 6493 of file LoopVectorize.cpp.

References llvm::dbgs(), llvm::SmallVectorTemplateCommon< T, typename >::empty(), llvm::RecurrenceDescriptor::getRecurrenceKind(), llvm::RecurrenceDescriptor::getRecurrenceType(), llvm::RecurrenceDescriptor::getReductionOpChain(), I, Legal, LLVM_DEBUG, PreferInLoopReductions, TheLoop, TTI, and useOrderedReductions().

◆ collectInstsToScalarize()

void LoopVectorizationCostModel::collectInstsToScalarize ( ElementCount VF )

Collects the instructions to scalarize for each predicated instruction in the loop.

Definition at line 4856 of file LoopVectorize.cpp.

References assert(), blockNeedsPredicationForAnyReason(), CM_Scalarize, llvm::dyn_cast(), I, llvm::MapVector< KeyT, ValueT, MapType, VectorType >::insert(), llvm::details::FixedOrScalableQuantity< LeafTy, ValueTy >::isScalable(), isScalarAfterVectorization(), isScalarWithPredication(), llvm::ElementCount::isVector(), llvm::predecessors(), and TheLoop.

Referenced by collectNonVectorizedAndSetWideningDecisions().

◆ collectNonVectorizedAndSetWideningDecisions()

void llvm::LoopVectorizationCostModel::collectNonVectorizedAndSetWideningDecisions ( ElementCount VF )

inline

Collect values that will not be widened, including Uniforms, Scalars, and Instructions to Scalarize for the given VF.

The sets depend on CM decision for Load/Store instructions that may be vectorized as interleave, gather-scatter or scalarized. Also make a decision on what to do about call instructions in the loop at that VF – scalarize, call a known vector routine, or call a vector intrinsic.

Definition at line 1170 of file LoopVectorize.cpp.

References collectInstsToScalarize(), llvm::ElementCount::isScalar(), setCostBasedWideningDecision(), and setVectorizedCallDecision().

Referenced by selectUserVectorizationFactor().

◆ collectValuesToIgnore()

void LoopVectorizationCostModel::collectValuesToIgnore ( )

Collect values we want to ignore in the cost model.

Definition at line 6340 of file LoopVectorize.cpp.

References _, AbstractManglingParser< Derived, Alloc >::Ops, AC, llvm::all_of(), llvm::any_of(), llvm::append_range(), llvm::LoopBlocksDFS::beginRPO(), llvm::cast(), llvm::CodeMetrics::collectEphemeralValues(), llvm::drop_end(), llvm::dyn_cast(), llvm::dyn_cast_or_null(), llvm::LoopBlocksDFS::endRPO(), llvm::InductionDescriptor::getCastInsts(), llvm::RecurrenceDescriptor::getCastInsts(), getInterleavedAccessGroup(), llvm::getLoadStorePointerOperand(), getParent(), llvm::BasicBlock::getSingleSuccessor(), I, llvm::isa(), isAccessInterleaved(), IsEmptyBlock(), Legal, LI, llvm::make_range(), llvm::LoopBlocksDFS::perform(), llvm::BasicBlock::phis(), llvm::SmallVectorTemplateBase< T, bool >::push_back(), requiresScalarEpilogue(), llvm::reverse(), llvm::SmallVectorTemplateCommon< T, typename >::size(), TheLoop, TLI, ValuesToIgnore, VecValuesToIgnore, and llvm::wouldInstructionBeTriviallyDead().

◆ computeMaxVF()

FixedScalableVFPair LoopVectorizationCostModel::computeMaxVF	(	ElementCount	UserVF,
		unsigned	UserIC )

Returns: An upper bound for the vectorization factors (both fixed and scalable). If the factors are 0, vectorization and interleaving should be avoided up front.

Definition at line 3487 of file LoopVectorize.cpp.

◆ expectedCost()

InstructionCost LoopVectorizationCostModel::expectedCost ( ElementCount VF )

Returns the expected execution cost.

The unit of the cost does not matter because we use the 'cost' units to compare different vector widths. The cost that is returned is not normalized by the factor width.

Definition at line 5033 of file LoopVectorize.cpp.

References addFullyUnrolledInstructionsToIgnore(), llvm::CallingConv::C, CostKind, llvm::SmallPtrSetImpl< PtrType >::count(), llvm::dbgs(), foldTailByMasking(), llvm::ForceTargetInstructionCost, getInstructionCost(), llvm::getPredBlockCostDivisor(), getSmallConstantTripCount(), I, InstructionCost, llvm::ElementCount::isScalar(), llvm::ElementCount::isVector(), Legal, LLVM_DEBUG, PSE, TheLoop, ValuesToIgnore, and VecValuesToIgnore.

Referenced by selectUserVectorizationFactor().

◆ foldTailByMasking()

bool llvm::LoopVectorizationCostModel::foldTailByMasking ( ) const

inline

Returns true if all loop blocks should be masked to fold tail loop.

Definition at line 1360 of file LoopVectorize.cpp.

References getTailFoldingStyle(), and llvm::None.

Referenced by blockNeedsPredicationForAnyReason(), computeMaxVF(), expectedCost(), isPredicatedInst(), and setCostBasedWideningDecision().

◆ foldTailWithEVL()

bool llvm::LoopVectorizationCostModel::foldTailWithEVL ( ) const

inline

Returns true if VP intrinsics with explicit vector length support should be generated in the tail folded loop.

Definition at line 1384 of file LoopVectorize.cpp.

References llvm::DataWithEVL, and getTailFoldingStyle().

Referenced by getInstructionCost(), and usePredicatedReductionSelect().

◆ getCallWideningDecision()

CallWideningDecision llvm::LoopVectorizationCostModel::getCallWideningDecision	(	CallInst *	CI,
		ElementCount	VF ) const

inline

Definition at line 1124 of file LoopVectorize.cpp.

References assert(), CM_Unknown, I, llvm::ElementCount::isScalar(), and llvm::Intrinsic::not_intrinsic.

Referenced by getVectorCallCost(), and isScalarWithPredication().

◆ getDivRemSpeculationCost()

std::pair< InstructionCost, InstructionCost > LoopVectorizationCostModel::getDivRemSpeculationCost	(	Instruction *	I,
		ElementCount	VF ) const

Return the costs for our two available strategies for lowering a div/rem operation which requires speculating at least one lane.

First result is for scalarization (will be invalid for scalable vectors); second is for the safe-divisor strategy.

Definition at line 2863 of file LoopVectorize.cpp.

References assert(), llvm::CmpInst::BAD_ICMP_PREDICATE, CostKind, llvm::details::FixedOrScalableQuantity< LeafTy, ValueTy >::getFixedValue(), llvm::Type::getInt1Ty(), llvm::InstructionCost::getInvalid(), llvm::getPredBlockCostDivisor(), I, llvm::isSafeToSpeculativelyExecute(), llvm::details::FixedOrScalableQuantity< LeafTy, ValueTy >::isScalable(), Operands, llvm::toVectorTy(), and TTI.

Referenced by getInstructionCost(), and isScalarWithPredication().

◆ getInstructionCost()

InstructionCost LoopVectorizationCostModel::getInstructionCost	(	Instruction *	I,
		ElementCount	VF )

Returns the execution time cost of an instruction for a given vector width.

Vector width of one means scalar.

Definition at line 5904 of file LoopVectorize.cpp.

Referenced by expectedCost(), and getInstructionCost().

◆ getInterleavedAccessGroup()

const InterleaveGroup< Instruction > * llvm::LoopVectorizationCostModel::getInterleavedAccessGroup ( Instruction * Instr ) const

inline

Get the interleaved access group that Instr belongs to.

Definition at line 1272 of file LoopVectorize.cpp.

References InterleaveInfo.

Referenced by collectValuesToIgnore(), interleavedAccessCanBeWidened(), and setCostBasedWideningDecision().

◆ getMaxSafeElements()

std::optional< unsigned > llvm::LoopVectorizationCostModel::getMaxSafeElements ( ) const

inline

Return maximum safe number of elements to be processed per vector iteration, which do not prevent store-load forwarding and are safe with regard to the memory dependencies.

Required for EVL-based VPlans to correctly calculate AVL (application vector length) as min(remaining AVL, MaxSafeElements). TODO: need to consider adjusting cost model to use this value as a vectorization factor for EVL-based vectorization.

Definition at line 1373 of file LoopVectorize.cpp.

◆ getMinimalBitwidths()

const MapVector< Instruction *, uint64_t > & llvm::LoopVectorizationCostModel::getMinimalBitwidths ( ) const

inline

Returns: The smallest bitwidth each instruction can be represented with. The vector equivalents of these instructions should be truncated to this type.

Definition at line 975 of file LoopVectorize.cpp.

◆ getReductionPatternCost()

std::optional< InstructionCost > LoopVectorizationCostModel::getReductionPatternCost	(	Instruction *	I,
		ElementCount	VF,
		Type *	VectorTy ) const

Return the cost of instructions in an inloop reduction pattern, if I is part of that pattern.

Definition at line 5293 of file LoopVectorize.cpp.

References llvm::cast(), CostKind, llvm::dyn_cast(), llvm::FMulAdd, llvm::VectorType::get(), llvm::RecurrenceDescriptor::getFastMathFlags(), llvm::Type::getIntegerBitWidth(), llvm::getMinMaxReductionIntrinsicOp(), llvm::Instruction::getOpcode(), llvm::RecurrenceDescriptor::getOpcode(), llvm::User::getOperand(), llvm::RecurrenceDescriptor::getRecurrenceKind(), llvm::RecurrenceDescriptor::getRecurrenceType(), llvm::Value::getType(), llvm::Value::hasOneUser(), I, llvm::isa(), llvm::RecurrenceDescriptor::isMinMaxRecurrenceKind(), llvm::ElementCount::isScalar(), llvm::InstructionCost::isValid(), Legal, llvm::PatternMatch::m_Instruction(), llvm::PatternMatch::m_Mul(), llvm::MIPatternMatch::m_OneUse(), llvm::PatternMatch::m_Value(), llvm::PatternMatch::m_ZExtOrSExt(), llvm::PatternMatch::match(), llvm::TargetTransformInfo::None, TheLoop, TTI, useOrderedReductions(), and llvm::Instruction::user_back().

Referenced by getInstructionCost(), getVectorCallCost(), and setVectorizedCallDecision().

◆ getSmallestAndWidestTypes()

std::pair< unsigned, unsigned > LoopVectorizationCostModel::getSmallestAndWidestTypes ( )

Returns: The size (in bits) of the smallest and widest types in the code that needs to be vectorized. We ignore values that remain scalar such as 64 bit loop indices.

Definition at line 4436 of file LoopVectorize.cpp.

References DL, ElementTypesInLoop, llvm::RecurrenceDescriptor::getMinWidthCastToRecurrenceTypeInBits(), llvm::RecurrenceDescriptor::getRecurrenceType(), llvm::Type::getScalarSizeInBits(), Legal, T, and TheFunction.

Referenced by determineVPlanVF().

◆ getTailFoldingStyle()

TailFoldingStyle llvm::LoopVectorizationCostModel::getTailFoldingStyle ( bool IVUpdateMayOverflow = true ) const

inline

Returns the TailFoldingStyle that is best for the current loop.

Definition at line 1307 of file LoopVectorize.cpp.

References llvm::None.

Referenced by computeMaxVF(), foldTailByMasking(), and foldTailWithEVL().

◆ getVectorCallCost()

InstructionCost LoopVectorizationCostModel::getVectorCallCost	(	CallInst *	CI,
		ElementCount	VF ) const

Estimate cost of a call instruction CI if it were vectorized with factor VF.

Return the cost of the instruction, including scalarization overhead if it's needed.

Definition at line 2490 of file LoopVectorize.cpp.

References llvm::CallBase::args(), llvm::LoopVectorizationCostModel::CallWideningDecision::Cost, CostKind, llvm::CallBase::getCalledFunction(), getCallWideningDecision(), getReductionPatternCost(), llvm::Value::getType(), getVectorIntrinsicCost(), llvm::getVectorIntrinsicIDForCall(), IntrinsicCost, llvm::RecurrenceDescriptor::isFMulAddIntrinsic(), llvm::ElementCount::isScalar(), llvm::SmallVectorTemplateBase< T, bool >::push_back(), TLI, and TTI.

Referenced by getInstructionCost().

◆ getVectorIntrinsicCost()

InstructionCost LoopVectorizationCostModel::getVectorIntrinsicCost	(	CallInst *	CI,
		ElementCount	VF ) const

Estimate cost of an intrinsic call instruction CI if it were vectorized with factor VF.

Return the cost of the instruction, including scalarization overhead if it's needed.

Definition at line 2524 of file LoopVectorize.cpp.

References llvm::CallBase::args(), Arguments, assert(), CostKind, llvm::dyn_cast(), llvm::CallBase::getCalledFunction(), llvm::Function::getFunctionType(), llvm::InstructionCost::getInvalid(), llvm::Value::getType(), llvm::getVectorIntrinsicIDForCall(), maybeVectorizeType(), llvm::FunctionType::param_begin(), llvm::FunctionType::param_end(), TLI, and TTI.

Referenced by getVectorCallCost(), and setVectorizedCallDecision().

◆ getVScaleForTuning()

std::optional< unsigned > llvm::LoopVectorizationCostModel::getVScaleForTuning ( ) const

inline

Return the value of vscale used for tuning the cost model.

Definition at line 1453 of file LoopVectorize.cpp.

Referenced by llvm::LoopVectorizePass::processLoop().

◆ getWideningCost()

InstructionCost llvm::LoopVectorizationCostModel::getWideningCost	(	Instruction *	I,
		ElementCount	VF )

inline

Return the vectorization cost for the given instruction I and vector width VF.

Definition at line 1100 of file LoopVectorize.cpp.

References assert(), I, and llvm::ElementCount::isVector().

Referenced by getInstructionCost().

◆ getWideningDecision()

InstWidening llvm::LoopVectorizationCostModel::getWideningDecision	(	Instruction *	I,
		ElementCount	VF ) const

inline

Return the cost model decision for the given instruction I and vector width VF.

Return CM_Unknown if this instruction did not pass through the cost modeling.

Definition at line 1085 of file LoopVectorize.cpp.

References assert(), CM_Unknown, I, llvm::ElementCount::isVector(), and TheLoop.

Referenced by getInstructionCost(), interleavedAccessCanBeWidened(), and setCostBasedWideningDecision().

◆ hasPredStores()

bool llvm::LoopVectorizationCostModel::hasPredStores ( ) const

inline

Definition at line 1428 of file LoopVectorize.cpp.

◆ interleavedAccessCanBeWidened()

bool LoopVectorizationCostModel::interleavedAccessCanBeWidened	(	Instruction *	I,
		ElementCount	VF ) const

Returns true if I is a memory instruction in an interleaved-group of memory accesses that can be vectorized with wide vector loads/stores and shuffles.

Definition at line 2920 of file LoopVectorize.cpp.

References assert(), blockNeedsPredicationForAnyReason(), CM_Unknown, DL, getInterleavedAccessGroup(), llvm::getLoadStoreAddressSpace(), llvm::getLoadStoreAlignment(), llvm::getLoadStoreType(), getWideningDecision(), hasIrregularType(), I, llvm::isa(), isAccessInterleaved(), llvm::details::FixedOrScalableQuantity< LeafTy, ValueTy >::isScalable(), isScalarEpilogueAllowed(), Legal, TTI, and useMaskedInterleavedAccesses().

Referenced by setCostBasedWideningDecision().

◆ invalidateCostModelingDecisions()

void llvm::LoopVectorizationCostModel::invalidateCostModelingDecisions ( )

inline

Invalidates decisions already taken by the cost model.

Definition at line 1415 of file LoopVectorize.cpp.

◆ isAccessInterleaved()

bool llvm::LoopVectorizationCostModel::isAccessInterleaved ( Instruction * Instr ) const

inline

Check if Instr belongs to any interleaved access group.

Definition at line 1266 of file LoopVectorize.cpp.

References InterleaveInfo.

Referenced by collectValuesToIgnore(), interleavedAccessCanBeWidened(), and setCostBasedWideningDecision().

◆ isDivRemScalarWithPredication()

bool llvm::LoopVectorizationCostModel::isDivRemScalarWithPredication	(	InstructionCost	ScalarCost,
		InstructionCost	SafeDivisorCost ) const

inline

Given costs for both strategies, return true if the scalar predication lowering should be used for div/rem.

This incorporates an override option so it is not simply a cost comparison.

Definition at line 1224 of file LoopVectorize.cpp.

References llvm::cl::BOU_FALSE, llvm::cl::BOU_TRUE, llvm::cl::BOU_UNSET, ForceSafeDivisor, and llvm_unreachable.

Referenced by getInstructionCost(), and isScalarWithPredication().

◆ isEpilogueVectorizationProfitable()

bool LoopVectorizationCostModel::isEpilogueVectorizationProfitable	(	const ElementCount	VF,
		const unsigned	IC ) const

Returns true if epilogue vectorization is considered profitable, and false otherwise.

VF is the vectorization factor chosen for the original loop. Multiplier is an aditional scaling factor applied to VF before comparing to EpilogueVectorizationMinVF.

Definition at line 4295 of file LoopVectorize.cpp.

References EpilogueVectorizationMinVF, estimateElementCount(), llvm::details::FixedOrScalableQuantity< LeafTy, ValueTy >::isFixed(), and TTI.

◆ isInLoopReduction()

bool llvm::LoopVectorizationCostModel::isInLoopReduction ( PHINode * Phi ) const

inline

Returns true if the Phi is part of an inloop reduction.

Definition at line 1389 of file LoopVectorize.cpp.

Referenced by getInstructionCost().

◆ isLegalGatherOrScatter()

bool llvm::LoopVectorizationCostModel::isLegalGatherOrScatter	(	Value *	V,
		ElementCount	VF )

inline

Returns true if the target machine can represent V as a masked gather or scatter operation.

Definition at line 1199 of file LoopVectorize.cpp.

References llvm::VectorType::get(), llvm::getLoadStoreAlignment(), llvm::getLoadStoreType(), llvm::isa(), llvm::ElementCount::isVector(), LI, and TTI.

Referenced by setCostBasedWideningDecision().

◆ isLegalMaskedLoad()

bool llvm::LoopVectorizationCostModel::isLegalMaskedLoad	(	Type *	DataType,
		Value *	Ptr,
		Align	Alignment,
		unsigned	AddressSpace ) const

inline

Returns true if the target machine supports masked load operation for the given DataType and kind of access to Ptr.

Definition at line 1191 of file LoopVectorize.cpp.

References Legal, Ptr, and TTI.

Referenced by isScalarWithPredication().

◆ isLegalMaskedStore()

bool llvm::LoopVectorizationCostModel::isLegalMaskedStore	(	Type *	DataType,
		Value *	Ptr,
		Align	Alignment,
		unsigned	AddressSpace ) const

inline

Returns true if the target machine supports masked store operation for the given DataType and kind of access to Ptr.

Definition at line 1183 of file LoopVectorize.cpp.

References Legal, Ptr, and TTI.

Referenced by isScalarWithPredication().

◆ isOptimizableIVTruncate()

bool llvm::LoopVectorizationCostModel::isOptimizableIVTruncate	(	Instruction *	I,
		ElementCount	VF )

inline

Return True if instruction I is an optimizable truncate whose operand is an induction variable.

Such a truncate will be removed by adding a new induction variable with the destination type.

Definition at line 1136 of file LoopVectorize.cpp.

References llvm::dyn_cast(), I, Legal, llvm::toVectorTy(), and TTI.

Referenced by getInstructionCost().

◆ isPredicatedInst()

bool LoopVectorizationCostModel::isPredicatedInst ( Instruction * I ) const

Returns true if I is an instruction that needs to be predicated at runtime.

The result is independent of the predication mechanism. Superset of instructions that return true for isScalarWithPredication.

Definition at line 2810 of file LoopVectorize.cpp.

References assert(), llvm::cast(), foldTailByMasking(), llvm::getLoadStorePointerOperand(), I, llvm::isa(), llvm::isSafeToSpeculativelyExecute(), Legal, llvm_unreachable, and TheLoop.

Referenced by getInstructionCost(), isScalarWithPredication(), and shouldConsiderInvariant().

◆ isProfitableToScalarize()

bool llvm::LoopVectorizationCostModel::isProfitableToScalarize	(	Instruction *	I,
		ElementCount	VF ) const

inline

Returns: True if it is more profitable to scalarize instruction I for vectorization factor VF.

Definition at line 981 of file LoopVectorize.cpp.

References assert(), I, llvm::ElementCount::isVector(), and TheLoop.

Referenced by canTruncateToMinimalBitwidth(), and getInstructionCost().

◆ isScalarAfterVectorization()

bool llvm::LoopVectorizationCostModel::isScalarAfterVectorization	(	Instruction *	I,
		ElementCount	VF ) const

inline

Returns true if I is known to be scalar after vectorization.

Definition at line 1015 of file LoopVectorize.cpp.

References assert(), I, llvm::ElementCount::isScalar(), and TheLoop.

Referenced by canTruncateToMinimalBitwidth(), collectInstsToScalarize(), and getInstructionCost().

◆ isScalarEpilogueAllowed()

bool llvm::LoopVectorizationCostModel::isScalarEpilogueAllowed ( ) const

inline

Returns true if a scalar epilogue is not allowed due to optsize or a loop hint annotation.

Definition at line 1302 of file LoopVectorize.cpp.

References llvm::CM_ScalarEpilogueAllowed.

Referenced by interleavedAccessCanBeWidened(), and requiresScalarEpilogue().

◆ isScalarWithPredication()

bool LoopVectorizationCostModel::isScalarWithPredication	(	Instruction *	I,
		ElementCount	VF ) const

Returns true if I is an instruction which requires predication and for which our chosen predication strategy is scalarization (i.e.

we don't have an alternate strategy such as masking available). VF is the vectorization factor that will be used to vectorize I.

Definition at line 2768 of file LoopVectorize.cpp.

References llvm::cast(), CM_Scalarize, llvm::VectorType::get(), getCallWideningDecision(), getDivRemSpeculationCost(), llvm::getLoadStoreAddressSpace(), llvm::getLoadStoreAlignment(), llvm::getLoadStorePointerOperand(), llvm::getLoadStoreType(), I, llvm::isa(), isDivRemScalarWithPredication(), isLegalMaskedLoad(), isLegalMaskedStore(), isPredicatedInst(), llvm::ElementCount::isScalar(), llvm::ElementCount::isVector(), llvm::LoopVectorizationCostModel::CallWideningDecision::Kind, Ptr, and TTI.

Referenced by collectInstsToScalarize(), memoryInstructionCanBeWidened(), and setCostBasedWideningDecision().

◆ isUniformAfterVectorization()

bool llvm::LoopVectorizationCostModel::isUniformAfterVectorization	(	Instruction *	I,
		ElementCount	VF ) const

inline

Returns true if I is known to be uniform after vectorization.

Definition at line 995 of file LoopVectorize.cpp.

References assert(), I, llvm::isa(), llvm::ElementCount::isScalar(), and TheLoop.

Referenced by getInstructionCost(), and setVectorizedCallDecision().

◆ memoryInstructionCanBeWidened()

bool LoopVectorizationCostModel::memoryInstructionCanBeWidened	(	Instruction *	I,
		ElementCount	VF )

Returns true if I is a memory instruction with consecutive memory access that can be widened.

Definition at line 3000 of file LoopVectorize.cpp.

References assert(), DL, llvm::getLoadStorePointerOperand(), llvm::getLoadStoreType(), hasIrregularType(), I, llvm::isa(), isScalarWithPredication(), Legal, and Ptr.

Referenced by setCostBasedWideningDecision().

◆ requiresScalarEpilogue()

bool llvm::LoopVectorizationCostModel::requiresScalarEpilogue ( bool IsVectorizing ) const

inline

Returns true if we're required to use a scalar epilogue for at least the final iteration of the original loop.

Definition at line 1278 of file LoopVectorize.cpp.

References llvm::dbgs(), EnableEarlyExitVectorization, InterleaveInfo, isScalarEpilogueAllowed(), Legal, LLVM_DEBUG, and TheLoop.

Referenced by collectValuesToIgnore().

◆ runtimeChecksRequired()

bool LoopVectorizationCostModel::runtimeChecksRequired ( )

Returns: True if runtime checks are required for vectorization, and false otherwise.

Definition at line 3261 of file LoopVectorize.cpp.

References llvm::dbgs(), Legal, LLVM_DEBUG, ORE, PSE, llvm::reportVectorizationFailure(), and TheLoop.

Referenced by computeMaxVF().

◆ selectUserVectorizationFactor()

bool llvm::LoopVectorizationCostModel::selectUserVectorizationFactor ( ElementCount UserVF )

inline

Setup cost-based decisions for user vectorization factor.

Returns: true if the UserVF is a feasible VF to be chosen.

Definition at line 921 of file LoopVectorize.cpp.

References collectNonVectorizedAndSetWideningDecisions(), expectedCost(), and llvm::InstructionCost::isValid().

◆ setCallWideningDecision()

void llvm::LoopVectorizationCostModel::setCallWideningDecision	(	CallInst *	CI,
		ElementCount	VF,
		InstWidening	Kind,
		Function *	Variant,
		Intrinsic::ID	IID,
		std::optional< unsigned >	MaskPos,
		InstructionCost	Cost )

inline

Definition at line 1116 of file LoopVectorize.cpp.

References assert(), and llvm::ElementCount::isScalar().

Referenced by setVectorizedCallDecision().

◆ setCostBasedWideningDecision()

void LoopVectorizationCostModel::setCostBasedWideningDecision ( ElementCount VF )

Memory access instruction may be vectorized in more than one way.

Form of instruction after vectorization depends on cost. This function takes cost-based decisions for Load/Store instructions and collects them in a map. This decisions map is used for building the lists of loop-uniform and loop-scalar instructions. The calculated cost is saved with widening decision in order to avoid redundant calculations.

Definition at line 5541 of file LoopVectorize.cpp.

References llvm::append_range(), assert(), llvm::cast(), CM_GatherScatter, CM_Interleave, CM_Scalarize, CM_Unknown, CM_Widen, CM_Widen_Reverse, llvm::dyn_cast(), llvm::dyn_cast_or_null(), llvm::SmallVectorTemplateCommon< T, typename >::empty(), foldTailByMasking(), llvm::ElementCount::getFixed(), getInterleavedAccessGroup(), llvm::InstructionCost::getInvalid(), llvm::details::FixedOrScalableQuantity< LeafTy, ValueTy >::getKnownMinValue(), llvm::getLoadStorePointerOperand(), llvm::getLoadStoreType(), getWideningDecision(), I, llvm::SmallPtrSetImpl< PtrType >::insert(), interleavedAccessCanBeWidened(), llvm::isa(), isAccessInterleaved(), isLegalGatherOrScatter(), llvm::details::FixedOrScalableQuantity< LeafTy, ValueTy >::isScalable(), llvm::ElementCount::isScalar(), isScalarWithPredication(), Legal, memoryInstructionCanBeWidened(), llvm::SmallVectorImpl< T >::pop_back_val(), Ptr, llvm::SmallVectorTemplateBase< T, bool >::push_back(), setWideningDecision(), TheLoop, and TTI.

Referenced by collectNonVectorizedAndSetWideningDecisions().

◆ setTailFoldingStyles()

void llvm::LoopVectorizationCostModel::setTailFoldingStyles	(	bool	IsScalableVF,
		unsigned	UserIC )

inline

Selects and saves TailFoldingStyle for 2 options - if IV update may overflow or not.

Parameters

IsScalableVF	true if scalable vector factors enabled.
UserIC	User specific interleave count.

Definition at line 1318 of file LoopVectorize.cpp.

References assert(), llvm::CM_ScalarEpilogueAllowed, llvm::CM_ScalarEpilogueNotNeededUsePredicate, llvm::DataWithEVL, llvm::DataWithoutLaneMask, llvm::dbgs(), llvm::EnableVPlanNativePath, ForceTailFoldingStyle, Legal, LLVM_DEBUG, llvm::None, and TTI.

Referenced by computeMaxVF().

◆ setVectorizedCallDecision()

void LoopVectorizationCostModel::setVectorizedCallDecision ( ElementCount VF )

A call may be vectorized in different ways depending on whether we have vectorized variants available and whether the target supports masking.

This function analyzes all calls in the function at the supplied VF, makes a decision based on the costs of available options, and stores that decision in a map for use in planning and plan execution.

Definition at line 5733 of file LoopVectorize.cpp.

References llvm::CallBase::args(), assert(), CM_IntrinsicCall, CM_Scalarize, CM_VectorCall, CostKind, llvm::dyn_cast(), llvm::CallBase::getArgOperand(), llvm::CallBase::getCalledFunction(), llvm::Module::getFunction(), llvm::InstructionCost::getInvalid(), llvm::details::FixedOrScalableQuantity< LeafTy, ValueTy >::getKnownMinValue(), llvm::VFDatabase::getMappings(), llvm::Instruction::getModule(), llvm::VFInfo::getParamIndexForOptionalMask(), getReductionPatternCost(), llvm::ScalarEvolution::getSCEV(), llvm::Value::getType(), getVectorIntrinsicCost(), llvm::getVectorIntrinsicIDForCall(), llvm::GlobalPredicate, I, IntrinsicCost, llvm::details::FixedOrScalableQuantity< LeafTy, ValueTy >::isFixed(), llvm::RecurrenceDescriptor::isFMulAddIntrinsic(), llvm::CallBase::isNoBuiltin(), llvm::ElementCount::isScalar(), isUniformAfterVectorization(), isValid(), llvm::ElementCount::isVector(), Legal, llvm::SCEVPatternMatch::m_SCEV(), llvm::SCEVPatternMatch::m_scev_AffineAddRec(), llvm::SCEVPatternMatch::m_scev_SpecificSInt(), llvm::SCEVPatternMatch::m_SpecificLoop(), llvm::PatternMatch::match(), llvm::Intrinsic::not_intrinsic, llvm::OMP_Linear, llvm::OMP_Uniform, PSE, llvm::SmallVectorTemplateBase< T, bool >::push_back(), setCallWideningDecision(), TheLoop, TLI, llvm::toVectorizedTy(), TTI, and llvm::Vector.

Referenced by collectNonVectorizedAndSetWideningDecisions().

◆ setWideningDecision() [1/2]

void llvm::LoopVectorizationCostModel::setWideningDecision	(	const InterleaveGroup< Instruction > *	Grp,
		ElementCount	VF,
		InstWidening	W,
		InstructionCost	Cost )

inline

Save vectorization decision W and Cost taken by the cost model for interleaving group Grp and vector width VF.

Broadcast this decicion to all instructions inside the group. When interleaving, the cost will only be assigned one instruction, the insert position. For other cases, add the appropriate fraction of the total cost to each instruction. This ensures accurate costs are used, even if the insert position instruction is not used.

Definition at line 1058 of file LoopVectorize.cpp.

References assert(), CM_Interleave, llvm::InterleaveGroup< InstTy >::getFactor(), llvm::InterleaveGroup< InstTy >::getInsertPos(), llvm::InterleaveGroup< InstTy >::getMember(), llvm::InterleaveGroup< InstTy >::getNumMembers(), I, and llvm::ElementCount::isVector().