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

Collaboration diagram for llvm::LoopVectorizationCostModel:

[legend]

Classes
struct	RegisterUsage
	A struct that represents some properties of the register usage of a loop. More...

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

Public Member Functions
	LoopVectorizationCostModel (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)

Optional< unsigned >	computeMaxVF (bool OptForSize)

VectorizationFactor	selectVectorizationFactor (unsigned MaxVF)

void	selectUserVectorizationFactor (unsigned UserVF)
	Setup cost-based decisions for user vectorization factor. More...

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

unsigned	selectInterleaveCount (bool OptForSize, unsigned VF, unsigned LoopCost)

void	setCostBasedWideningDecision (unsigned VF)
	Memory access instruction may be vectorized in more than one way. More...

SmallVector< RegisterUsage, 8 >	calculateRegisterUsage (ArrayRef< unsigned > VFs)

void	collectValuesToIgnore ()
	Collect values we want to ignore in the cost model. More...

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

bool	isProfitableToScalarize (Instruction *I, unsigned VF) const

bool	isUniformAfterVectorization (Instruction *I, unsigned VF) const
	Returns true if `I` is known to be uniform after vectorization. More...

bool	isScalarAfterVectorization (Instruction *I, unsigned VF) const
	Returns true if `I` is known to be scalar after vectorization. More...

bool	canTruncateToMinimalBitwidth (Instruction *I, unsigned VF) const

void	setWideningDecision (Instruction *I, unsigned VF, InstWidening W, unsigned Cost)
	Save vectorization decision `W` and `Cost` taken by the cost model for instruction `I` and vector width `VF`. More...

void	setWideningDecision (const InterleaveGroup< Instruction > *Grp, unsigned VF, InstWidening W, unsigned Cost)
	Save vectorization decision `W` and `Cost` taken by the cost model for interleaving group `Grp` and vector width `VF`. More...

InstWidening	getWideningDecision (Instruction *I, unsigned VF)
	Return the cost model decision for the given instruction `I` and vector width `VF`. More...

unsigned	getWideningCost (Instruction *I, unsigned VF)
	Return the vectorization cost for the given instruction `I` and vector width `VF`. More...

bool	isOptimizableIVTruncate (Instruction *I, unsigned VF)
	Return True if instruction `I` is an optimizable truncate whose operand is an induction variable. More...

void	collectInstsToScalarize (unsigned VF)
	Collects the instructions to scalarize for each predicated instruction in the loop. More...

void	collectUniformsAndScalars (unsigned VF)
	Collect Uniform and Scalar values for the given `VF`. More...

bool	isLegalMaskedStore (Type DataType, Value Ptr)
	Returns true if the target machine supports masked store operation for the given `DataType` and kind of access to `Ptr`. More...

bool	isLegalMaskedLoad (Type DataType, Value Ptr)
	Returns true if the target machine supports masked load operation for the given `DataType` and kind of access to `Ptr`. More...

bool	isLegalMaskedScatter (Type *DataType)
	Returns true if the target machine supports masked scatter operation for the given `DataType`. More...

bool	isLegalMaskedGather (Type *DataType)
	Returns true if the target machine supports masked gather operation for the given `DataType`. More...

bool	isLegalGatherOrScatter (Value *V)
	Returns true if the target machine can represent `V` as a masked gather or scatter operation. More...

bool	isScalarWithPredication (Instruction *I, unsigned VF=1)
	Returns true if `I` is an instruction that will be scalarized with predication. More...

bool	isPredicatedInst (Instruction *I)

bool	memoryInstructionCanBeWidened (Instruction *I, unsigned VF=1)
	Returns true if `I` is a memory instruction with consecutive memory access that can be widened. More...

bool	interleavedAccessCanBeWidened (Instruction *I, unsigned VF=1)
	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. More...

bool	isAccessInterleaved (Instruction *Instr)
	Check if `Instr` belongs to any interleaved access group. More...

const InterleaveGroup< Instruction > *	getInterleavedAccessGroup (Instruction *Instr)
	Get the interleaved access group that `Instr` belongs to. More...

bool	requiresScalarEpilogue () const
	Returns true if an interleaved group requires a scalar iteration to handle accesses with gaps, and there is nothing preventing us from creating a scalar epilogue. More...

bool	isScalarEpilogueAllowed () const
	Returns true if a scalar epilogue is not allowed due to optsize. More...

bool	foldTailByMasking () const
	Returns true if all loop blocks should be masked to fold tail loop. More...

bool	blockNeedsPredication (BasicBlock *BB)

Public Attributes
Loop *	TheLoop
	The loop that we evaluate. More...

PredicatedScalarEvolution &	PSE
	Predicated scalar evolution analysis. More...

LoopInfo *	LI
	Loop Info analysis. More...

LoopVectorizationLegality *	Legal
	Vectorization legality. More...

const TargetTransformInfo &	TTI
	Vector target information. More...

const TargetLibraryInfo *	TLI
	Target Library Info. More...

DemandedBits *	DB
	Demanded bits analysis. More...

AssumptionCache *	AC
	Assumption cache. More...

OptimizationRemarkEmitter *	ORE
	Interface to emit optimization remarks. More...

const Function *	TheFunction

const LoopVectorizeHints *	Hints
	Loop Vectorize Hint. More...

InterleavedAccessInfo &	InterleaveInfo
	The interleave access information contains groups of interleaved accesses with the same stride and close to each other. More...

SmallPtrSet< const Value *, 16 >	ValuesToIgnore
	Values to ignore in the cost model. More...

SmallPtrSet< const Value *, 16 >	VecValuesToIgnore
	Values to ignore in the cost model when VF > 1. More...

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 825 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

Definition at line 956 of file LoopVectorize.cpp.

Constructor & Destructor Documentation

◆ LoopVectorizationCostModel()

llvm::LoopVectorizationCostModel::LoopVectorizationCostModel	(	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
	)

inline

Definition at line 827 of file LoopVectorize.cpp.

Member Function Documentation

◆ blockNeedsPredication()

bool llvm::LoopVectorizationCostModel::blockNeedsPredication ( BasicBlock * BB )

inline

Definition at line 1148 of file LoopVectorize.cpp.

References llvm::LoopVectorizationLegality::blockNeedsPredication(), llvm::createLVMissedAnalysis(), and llvm::InnerLoopVectorizer::Legal.

◆ calculateRegisterUsage()

SmallVector< LoopVectorizationCostModel::RegisterUsage, 8 > LoopVectorizationCostModel::calculateRegisterUsage ( ArrayRef< unsigned > VFs )

Returns: Returns information about the register usages of the loop for the given vectorization factors.

Definition at line 5061 of file LoopVectorize.cpp.

References assert(), llvm::LoopBlocksDFS::beginRPO(), llvm::dbgs(), DFS(), llvm::dyn_cast(), llvm::SmallPtrSetImpl< PtrType >::end(), llvm::LoopBlocksDFS::endRPO(), llvm::SmallPtrSetImpl< PtrType >::erase(), llvm::SmallPtrSetImpl< PtrType >::find(), llvm::LoopVectorizationLegality::getMaxSafeDepDistBytes(), llvm::TargetTransformInfo::getRegisterBitWidth(), I, llvm::SmallPtrSetImpl< PtrType >::insert(), llvm::InnerLoopVectorizer::Legal, llvm::InnerLoopVectorizer::LI, List, LLVM_DEBUG, llvm::LoopVectorizationCostModel::RegisterUsage::LoopInvariantRegs, llvm::make_range(), llvm::max(), llvm::LoopVectorizationCostModel::RegisterUsage::MaxLocalUsers, NumberOfStoresToPredicate, llvm::LoopBlocksDFS::perform(), llvm::SmallVectorTemplateBase< T >::push_back(), llvm::RegUsage, llvm::SmallVectorBase::size(), llvm::SmallPtrSetImplBase::size(), llvm::ArrayRef< T >::size(), llvm::InnerLoopVectorizer::TTI, and llvm::InnerLoopVectorizer::VF.

◆ canTruncateToMinimalBitwidth()

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

inline

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

Definition at line 949 of file LoopVectorize.cpp.

◆ collectInstsToScalarize()

void LoopVectorizationCostModel::collectInstsToScalarize ( unsigned VF )

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

Definition at line 5234 of file LoopVectorize.cpp.

References assert(), llvm::DenseMapBase< DerivedT, KeyT, ValueT, KeyInfoT, BucketT >::begin(), C, llvm::InnerLoopVectorizer::Cost, llvm::dbgs(), llvm::SmallVectorBase::empty(), llvm::DenseMapBase< DerivedT, KeyT, ValueT, KeyInfoT, BucketT >::end(), llvm::DenseMapBase< DenseMap< KeyT, ValueT, KeyInfoT, BucketT >, KeyT, ValueT, KeyInfoT, BucketT >::end(), llvm::DenseMapBase< DenseMap< KeyT, ValueT, KeyInfoT, BucketT >, KeyT, ValueT, KeyInfoT, BucketT >::find(), ForceTargetInstructionCost, llvm::TargetTransformInfo::getCFInstrCost(), llvm::Instruction::getParent(), getReciprocalPredBlockProb(), llvm::TargetTransformInfo::getScalarizationOverhead(), llvm::Value::getType(), llvm::Value::hasOneUse(), I, llvm::DenseMapBase< DerivedT, KeyT, ValueT, KeyInfoT, BucketT >::insert(), llvm::VectorType::isValidElementType(), llvm::Type::isVoidTy(), LLVM_DEBUG, llvm::User::operands(), llvm::SmallVectorImpl< T >::pop_back_val(), llvm::SmallVectorTemplateBase< T >::push_back(), ToVectorTy(), llvm::InnerLoopVectorizer::TTI, and llvm::InnerLoopVectorizer::VF.

◆ collectUniformsAndScalars()

void llvm::LoopVectorizationCostModel::collectUniformsAndScalars ( unsigned VF )

inline

Collect Uniform and Scalar values for the given VF.

The sets depend on CM decision for Load/Store instructions that may be vectorized as interleave, gather-scatter or scalarized.

Definition at line 1051 of file LoopVectorize.cpp.

◆ collectValuesToIgnore()

void LoopVectorizationCostModel::collectValuesToIgnore ( )

Collect values we want to ignore in the cost model.

Definition at line 6049 of file LoopVectorize.cpp.

References llvm::InnerLoopVectorizer::AC, llvm::SmallVectorTemplateCommon< T >::begin(), llvm::SmallPtrSetImpl< PtrType >::begin(), llvm::CodeMetrics::collectEphemeralValues(), llvm::SmallVectorTemplateCommon< T >::end(), llvm::SmallPtrSetImpl< PtrType >::end(), llvm::RecurrenceDescriptor::getCastInsts(), llvm::InductionDescriptor::getCastInsts(), llvm::LoopVectorizationLegality::getInductionVars(), llvm::LoopVectorizationLegality::getReductionVars(), llvm::InnerLoopVectorizer::Induction, llvm::InnerLoopVectorizer::Legal, and Reduction.

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

◆ computeMaxVF()

Optional< unsigned > LoopVectorizationCostModel::computeMaxVF ( bool OptForSize )

Returns: An upper bound for the vectorization factor, or None if vectorization should be avoided up front.

Definition at line 4624 of file LoopVectorize.cpp.

References assert(), llvm::LoopVectorizationLegality::canFoldTailByMasking(), llvm::computeMinimumValueSizes(), llvm::dbgs(), llvm::OptimizationRemarkEmitter::emit(), llvm::DenseMapBase< DerivedT, KeyT, ValueT, KeyInfoT, BucketT >::empty(), llvm::LoopVectorizationLegality::getLAI(), llvm::LoopVectorizationLegality::getMaxSafeRegisterWidth(), llvm::TargetTransformInfo::getMinimumVF(), llvm::TargetTransformInfo::getNumberOfRegisters(), llvm::SCEVUnionPredicate::getPredicates(), llvm::TargetTransformInfo::getRegisterBitWidth(), llvm::LoopVectorizationLegality::getRuntimePointerChecking(), llvm::PredicatedScalarEvolution::getSE(), llvm::ScalarEvolution::getSmallConstantTripCount(), llvm::LoopAccessInfo::getSymbolicStrides(), llvm::PredicatedScalarEvolution::getUnionPredicate(), llvm::TargetTransformInfo::hasBranchDivergence(), llvm::isPowerOf2_32(), llvm::InnerLoopVectorizer::Legal, LLVM_DEBUG, MaximizeBandwidth, llvm::RuntimePointerChecking::Need, llvm::None, llvm::InnerLoopVectorizer::ORE, llvm::InnerLoopVectorizer::PSE, llvm::SmallVectorTemplateBase< T, bool >::push_back(), llvm::TargetTransformInfo::shouldMaximizeVectorBandwidth(), llvm::InnerLoopVectorizer::TTI, useMaskedInterleavedAccesses(), and llvm::AArch64CC::VS.

◆ foldTailByMasking()

bool llvm::LoopVectorizationCostModel::foldTailByMasking ( ) const

inline

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

Definition at line 1146 of file LoopVectorize.cpp.

Referenced by llvm::InnerLoopVectorizer::createVectorizedLoopSkeleton(), llvm::InnerLoopVectorizer::emitMemRuntimeChecks(), llvm::InnerLoopVectorizer::emitMinimumIterationCountCheck(), llvm::InnerLoopVectorizer::emitSCEVChecks(), and llvm::InnerLoopVectorizer::getOrCreateVectorTripCount().

◆ getInterleavedAccessGroup()

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

inline

Get the interleaved access group that Instr belongs to.

Definition at line 1131 of file LoopVectorize.cpp.

Referenced by llvm::InnerLoopVectorizer::vectorizeInterleaveGroup().

◆ 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 895 of file LoopVectorize.cpp.

Referenced by llvm::InnerLoopVectorizer::truncateToMinimalBitwidths().

◆ 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 4852 of file LoopVectorize.cpp.

References llvm::RecurrenceDescriptor::getRecurrenceType(), llvm::LoopVectorizationLegality::getReductionVars(), llvm::Type::getScalarType(), llvm::DataLayout::getTypeSizeInBits(), I, llvm::Type::isPointerTy(), llvm::LoopVectorizationLegality::isReductionVariable(), llvm::InnerLoopVectorizer::Legal, llvm::max(), and llvm::ARM_MB::ST.

◆ getWideningCost()

unsigned llvm::LoopVectorizationCostModel::getWideningCost	(	Instruction *	I,
		unsigned	VF
	)

inline

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

Definition at line 1010 of file LoopVectorize.cpp.

References assert().

◆ getWideningDecision()

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

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 993 of file LoopVectorize.cpp.

References assert(), and EnableVPlanNativePath.

Referenced by llvm::InnerLoopVectorizer::vectorizeMemoryInstruction().

◆ interleavedAccessCanBeWidened()

bool LoopVectorizationCostModel::interleavedAccessCanBeWidened	(	Instruction *	I,
		unsigned	VF = `1`
	)

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 4390 of file LoopVectorize.cpp.

References assert(), llvm::LoopVectorizationLegality::blockNeedsPredication(), getMemInstValueType(), llvm::Instruction::getParent(), I, llvm::TargetTransformInfo::isLegalMaskedLoad(), llvm::TargetTransformInfo::isLegalMaskedStore(), llvm::LoopVectorizationLegality::isMaskRequired(), llvm::InnerLoopVectorizer::Legal, llvm::InnerLoopVectorizer::TTI, and useMaskedInterleavedAccesses().

◆ isAccessInterleaved()

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

inline

Check if Instr belongs to any interleaved access group.

Definition at line 1125 of file LoopVectorize.cpp.

◆ isLegalGatherOrScatter()

bool llvm::LoopVectorizationCostModel::isLegalGatherOrScatter ( Value * V )

inline

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

Definition at line 1086 of file LoopVectorize.cpp.

References getMemInstValueType(), I, llvm::InnerLoopVectorizer::LI, and SI.

◆ isLegalMaskedGather()

bool llvm::LoopVectorizationCostModel::isLegalMaskedGather ( Type * DataType )

inline

Returns true if the target machine supports masked gather operation for the given DataType.

Definition at line 1080 of file LoopVectorize.cpp.

References llvm::TargetTransformInfo::isLegalMaskedGather(), and llvm::InnerLoopVectorizer::TTI.

◆ isLegalMaskedLoad()

bool llvm::LoopVectorizationCostModel::isLegalMaskedLoad	(	Type *	DataType,
		Value *	Ptr
	)

inline

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

Definition at line 1068 of file LoopVectorize.cpp.

References llvm::LoopVectorizationLegality::isConsecutivePtr(), llvm::TargetTransformInfo::isLegalMaskedLoad(), llvm::InnerLoopVectorizer::Legal, and llvm::InnerLoopVectorizer::TTI.

◆ isLegalMaskedScatter()

bool llvm::LoopVectorizationCostModel::isLegalMaskedScatter ( Type * DataType )

inline

Returns true if the target machine supports masked scatter operation for the given DataType.

Definition at line 1074 of file LoopVectorize.cpp.

References llvm::TargetTransformInfo::isLegalMaskedScatter(), and llvm::InnerLoopVectorizer::TTI.

◆ isLegalMaskedStore()

bool llvm::LoopVectorizationCostModel::isLegalMaskedStore	(	Type *	DataType,
		Value *	Ptr
	)

inline

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

Definition at line 1062 of file LoopVectorize.cpp.

References llvm::LoopVectorizationLegality::isConsecutivePtr(), llvm::TargetTransformInfo::isLegalMaskedStore(), llvm::InnerLoopVectorizer::Legal, and llvm::InnerLoopVectorizer::TTI.

◆ isOptimizableIVTruncate()

bool llvm::LoopVectorizationCostModel::isOptimizableIVTruncate	(	Instruction *	I,
		unsigned	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 1021 of file LoopVectorize.cpp.

References llvm::dyn_cast(), llvm::LoopVectorizationLegality::getPrimaryInduction(), I, llvm::LoopVectorizationLegality::isInductionPhi(), llvm::TargetTransformInfo::isTruncateFree(), llvm::InnerLoopVectorizer::Legal, ToVectorTy(), and llvm::InnerLoopVectorizer::TTI.

◆ isPredicatedInst()

bool llvm::LoopVectorizationCostModel::isPredicatedInst ( Instruction * I )

inline

Definition at line 1105 of file LoopVectorize.cpp.

References llvm::Instruction::getParent(), I, llvm::LoopVectorizationLegality::isMaskRequired(), and llvm::InnerLoopVectorizer::Legal.

◆ isProfitableToScalarize()

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

inline

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

Definition at line 901 of file LoopVectorize.cpp.

References assert(), and EnableVPlanNativePath.

Referenced by llvm::InnerLoopVectorizer::shouldScalarizeInstruction().

◆ isScalarAfterVectorization()

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

inline

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

Definition at line 932 of file LoopVectorize.cpp.

References assert(), and EnableVPlanNativePath.

Referenced by llvm::InnerLoopVectorizer::shouldScalarizeInstruction().

◆ isScalarEpilogueAllowed()

bool llvm::LoopVectorizationCostModel::isScalarEpilogueAllowed ( ) const

inline

Returns true if a scalar epilogue is not allowed due to optsize.

Definition at line 1143 of file LoopVectorize.cpp.

Referenced by llvm::InnerLoopVectorizer::vectorizeInterleaveGroup().

◆ isScalarWithPredication()

bool LoopVectorizationCostModel::isScalarWithPredication	(	Instruction *	I,
		unsigned	VF = `1`
	)

Returns true if I is an instruction that will be scalarized with predication.

Such instructions include conditional stores and instructions that may divide by zero. If a non-zero VF has been calculated, we check if I will be scalarized predication for that VF.

Definition at line 4357 of file LoopVectorize.cpp.

References assert(), llvm::getLoadStorePointerOperand(), getMemInstValueType(), llvm::Instruction::getOpcode(), llvm::Instruction::getParent(), I, llvm::LoopVectorizationLegality::isMaskRequired(), llvm::InnerLoopVectorizer::Legal, llvm::SPII::Load, mayDivideByZero(), and llvm::SPII::Store.

◆ isUniformAfterVectorization()

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

inline

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

Definition at line 916 of file LoopVectorize.cpp.

References assert(), and EnableVPlanNativePath.

Referenced by llvm::InnerLoopVectorizer::buildScalarSteps(), llvm::InnerLoopVectorizer::fixLCSSAPHIs(), llvm::InnerLoopVectorizer::getOrCreateScalarValue(), llvm::InnerLoopVectorizer::getOrCreateVectorValue(), and llvm::InnerLoopVectorizer::widenPHIInstruction().

◆ memoryInstructionCanBeWidened()

bool LoopVectorizationCostModel::memoryInstructionCanBeWidened	(	Instruction *	I,
		unsigned	VF = `1`
	)

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

Definition at line 4419 of file LoopVectorize.cpp.

References llvm::all_of(), assert(), llvm::SetVector< T, Vector, Set >::begin(), llvm::SetVector< T, Vector, Set >::count(), llvm::dbgs(), llvm::dyn_cast(), llvm::SetVector< T, Vector, Set >::end(), llvm::SmallPtrSetImpl< PtrType >::end(), llvm::SmallPtrSetImpl< PtrType >::find(), llvm::Module::getDataLayout(), llvm::LoopVectorizationLegality::getInductionVars(), llvm::getLoadStorePointerOperand(), llvm::Instruction::getModule(), llvm::User::getOperand(), llvm::BasicBlock::getTerminator(), llvm::Value::getType(), hasIrregularType(), I, llvm::InnerLoopVectorizer::Induction, llvm::SetVector< T, Vector, Set >::insert(), llvm::SetVector< T, SmallVector< T, N >, SmallDenseSet< T, N > >::insert(), llvm::SmallPtrSetImpl< PtrType >::insert(), llvm::LoopVectorizationLegality::isConsecutivePtr(), llvm::LoopVectorizationLegality::isFirstOrderRecurrence(), llvm::InnerLoopVectorizer::Legal, llvm::InnerLoopVectorizer::LI, LLVM_DEBUG, OP, llvm::User::operand_values(), SI, llvm::SetVector< T, Vector, Set >::size(), and llvm::InnerLoopVectorizer::VF.

◆ requiresScalarEpilogue()

bool llvm::LoopVectorizationCostModel::requiresScalarEpilogue ( ) const

inline

Returns true if an interleaved group requires a scalar iteration to handle accesses with gaps, and there is nothing preventing us from creating a scalar epilogue.

Definition at line 1138 of file LoopVectorize.cpp.

Referenced by llvm::InnerLoopVectorizer::emitMinimumIterationCountCheck(), and llvm::InnerLoopVectorizer::getOrCreateVectorTripCount().

◆ selectInterleaveCount()

unsigned LoopVectorizationCostModel::selectInterleaveCount	(	bool	OptForSize,
		unsigned	VF,
		unsigned	LoopCost
	)

Returns: The desired interleave count. If interleave count has been specified by metadata it will be returned. Otherwise, the interleave count is computed and returned. VF and LoopCost are the selected vectorization factor and the cost of the selected VF.

Definition at line 4907 of file LoopVectorize.cpp.

References llvm::dbgs(), llvm::DenseMapBase< DerivedT, KeyT, ValueT, KeyInfoT, BucketT >::empty(), llvm::TargetTransformInfo::enableAggressiveInterleaving(), EnableIndVarRegisterHeur, EnableLoadStoreRuntimeInterleave, F(), ForceTargetMaxScalarInterleaveFactor, ForceTargetMaxVectorInterleaveFactor, ForceTargetNumScalarRegs, ForceTargetNumVectorRegs, llvm::TargetTransformInfo::getMaxInterleaveFactor(), llvm::LoopVectorizationLegality::getMaxSafeDepDistBytes(), llvm::TargetTransformInfo::getNumberOfRegisters(), llvm::LoopVectorizationLegality::getNumLoads(), llvm::LoopVectorizationLegality::getNumStores(), llvm::LoopVectorizationLegality::getReductionVars(), llvm::LoopVectorizationLegality::getRuntimePointerChecking(), llvm::PredicatedScalarEvolution::getSE(), llvm::ScalarEvolution::getSmallConstantTripCount(), llvm::InnerLoopVectorizer::Legal, LLVM_DEBUG, llvm::LoopVectorizationCostModel::RegisterUsage::LoopInvariantRegs, llvm::max(), llvm::LoopVectorizationCostModel::RegisterUsage::MaxLocalUsers, MaxNestedScalarReductionIC, llvm::RuntimePointerChecking::Need, llvm::PowerOf2Floor(), llvm::InnerLoopVectorizer::PSE, SmallLoopCost, and llvm::InnerLoopVectorizer::TTI.

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

◆ selectUserVectorizationFactor()

void llvm::LoopVectorizationCostModel::selectUserVectorizationFactor ( unsigned UserVF )

inline

Setup cost-based decisions for user vectorization factor.

Definition at line 849 of file LoopVectorize.cpp.

References llvm::InnerLoopVectorizer::VF.

◆ selectVectorizationFactor()

VectorizationFactor LoopVectorizationCostModel::selectVectorizationFactor ( unsigned MaxVF )

Returns: The most profitable vectorization factor and the cost of that VF. This method checks every power of two up to MaxVF. If UserVF is not ZERO then this vectorization factor will be selected if vectorization is possible.

Definition at line 4800 of file LoopVectorize.cpp.

References C, llvm::InnerLoopVectorizer::Cost, llvm::dbgs(), llvm::OptimizationRemarkEmitter::emit(), EnableCondStoresVectorization, llvm::LoopVectorizeHints::FK_Enabled, LLVM_DEBUG, llvm::max(), and llvm::InnerLoopVectorizer::ORE.

◆ setCostBasedWideningDecision()

void LoopVectorizationCostModel::setCostBasedWideningDecision ( unsigned 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 5639 of file LoopVectorize.cpp.

◆ setWideningDecision() [1/2]

void llvm::LoopVectorizationCostModel::setWideningDecision	(	Instruction *	I,
		unsigned	VF,
		InstWidening	W,
		unsigned	Cost
	)

inline

Save vectorization decision W and Cost taken by the cost model for instruction I and vector width VF.

Definition at line 967 of file LoopVectorize.cpp.

References assert().

◆ setWideningDecision() [2/2]

void llvm::LoopVectorizationCostModel::setWideningDecision	(	const InterleaveGroup< Instruction > *	Grp,
		unsigned	VF,
		InstWidening	W,
		unsigned	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. But the cost will be assigned to one instruction only.

Definition at line 975 of file LoopVectorize.cpp.

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

Member Data Documentation

◆ AC

AssumptionCache* llvm::LoopVectorizationCostModel::AC

Assumption cache.

Definition at line 1321 of file LoopVectorize.cpp.

◆ DB

DemandedBits* llvm::LoopVectorizationCostModel::DB

Demanded bits analysis.

Definition at line 1318 of file LoopVectorize.cpp.

◆ Hints

const LoopVectorizeHints* llvm::LoopVectorizationCostModel::Hints

Loop Vectorize Hint.

Definition at line 1329 of file LoopVectorize.cpp.

◆ InterleaveInfo

InterleavedAccessInfo& llvm::LoopVectorizationCostModel::InterleaveInfo

The interleave access information contains groups of interleaved accesses with the same stride and close to each other.

Definition at line 1333 of file LoopVectorize.cpp.

◆ Legal

LoopVectorizationLegality* llvm::LoopVectorizationCostModel::Legal

Vectorization legality.

Definition at line 1309 of file LoopVectorize.cpp.

◆ LI

LoopInfo* llvm::LoopVectorizationCostModel::LI

Loop Info analysis.

Definition at line 1306 of file LoopVectorize.cpp.

◆ ORE

OptimizationRemarkEmitter* llvm::LoopVectorizationCostModel::ORE

Interface to emit optimization remarks.

Definition at line 1324 of file LoopVectorize.cpp.

◆ PSE

PredicatedScalarEvolution& llvm::LoopVectorizationCostModel::PSE

Predicated scalar evolution analysis.

Definition at line 1303 of file LoopVectorize.cpp.

◆ TheFunction

const Function* llvm::LoopVectorizationCostModel::TheFunction

Definition at line 1326 of file LoopVectorize.cpp.

◆ TheLoop

Loop* llvm::LoopVectorizationCostModel::TheLoop

The loop that we evaluate.

Definition at line 1300 of file LoopVectorize.cpp.

◆ TLI

const TargetLibraryInfo* llvm::LoopVectorizationCostModel::TLI

Target Library Info.

Definition at line 1315 of file LoopVectorize.cpp.

◆ TTI

const TargetTransformInfo& llvm::LoopVectorizationCostModel::TTI

Vector target information.

Definition at line 1312 of file LoopVectorize.cpp.

◆ ValuesToIgnore

SmallPtrSet<const Value *, 16> llvm::LoopVectorizationCostModel::ValuesToIgnore

Values to ignore in the cost model.

Definition at line 1336 of file LoopVectorize.cpp.

◆ VecValuesToIgnore

SmallPtrSet<const Value *, 16> llvm::LoopVectorizationCostModel::VecValuesToIgnore

Values to ignore in the cost model when VF > 1.

Definition at line 1339 of file LoopVectorize.cpp.

The documentation for this class was generated from the following file:

lib/Transforms/Vectorize/LoopVectorize.cpp

Classes

Public Types

Public Member Functions

Public Attributes

Detailed Description

Member Enumeration Documentation

◆ InstWidening

Constructor & Destructor Documentation

◆ LoopVectorizationCostModel()

Member Function Documentation

◆ blockNeedsPredication()

◆ calculateRegisterUsage()

◆ canTruncateToMinimalBitwidth()

◆ collectInstsToScalarize()

◆ collectUniformsAndScalars()

◆ collectValuesToIgnore()

◆ computeMaxVF()

◆ foldTailByMasking()

◆ getInterleavedAccessGroup()

◆ getMinimalBitwidths()

◆ getSmallestAndWidestTypes()

◆ getWideningCost()

◆ getWideningDecision()

◆ interleavedAccessCanBeWidened()

◆ isAccessInterleaved()

◆ isLegalGatherOrScatter()

◆ isLegalMaskedGather()

◆ isLegalMaskedLoad()

◆ isLegalMaskedScatter()

◆ isLegalMaskedStore()

◆ isOptimizableIVTruncate()

◆ isPredicatedInst()

◆ isProfitableToScalarize()

◆ isScalarAfterVectorization()

◆ isScalarEpilogueAllowed()

◆ isScalarWithPredication()

◆ isUniformAfterVectorization()

◆ memoryInstructionCanBeWidened()

◆ requiresScalarEpilogue()

◆ selectInterleaveCount()

◆ selectUserVectorizationFactor()

◆ selectVectorizationFactor()

◆ setCostBasedWideningDecision()

◆ setWideningDecision() [1/2]

◆ setWideningDecision() [2/2]

Member Data Documentation

◆ AC

◆ DB

◆ Hints

◆ InterleaveInfo

◆ Legal

◆ LI

◆ ORE

◆ PSE

◆ TheFunction

◆ TheLoop

◆ TLI

◆ TTI

◆ ValuesToIgnore

◆ VecValuesToIgnore