HostTensorDescriptor Struct Reference

HostTensorDescriptor Struct Reference#

Composable Kernel: ck_tile::HostTensorDescriptor Struct Reference

Descriptor for tensors in host memory. More...

#include <host_tensor.hpp>

Public Member Functions
	HostTensorDescriptor ()=default
void	CalculateStrides ()
template<typename X, typename = std::enable_if_t<std::is_convertible_v<X, std::size_t>>>
	HostTensorDescriptor (const std::initializer_list< X > &lens)
template<typename Lengths, typename = std::enable_if_t< std::is_convertible_v<ck_tile::ranges::range_value_t<Lengths>, std::size_t>>>
	HostTensorDescriptor (const Lengths &lens)
template<typename X, typename Y, typename = std::enable_if_t<std::is_convertible_v<X, std::size_t> && std::is_convertible_v<Y, std::size_t>>>
	HostTensorDescriptor (const std::initializer_list< X > &lens, const std::initializer_list< Y > &strides)
template<typename Lengths, typename Strides, typename = std::enable_if_t< std::is_convertible_v<ck_tile::ranges::range_value_t<Lengths>, std::size_t> && std::is_convertible_v<ck_tile::ranges::range_value_t<Strides>, std::size_t>>>
	HostTensorDescriptor (const Lengths &lens, const Strides &strides)
std::size_t	get_num_of_dimension () const
std::size_t	get_element_size () const
	Calculates the total number of elements in the tensor.
std::size_t	get_element_space_size () const
	Calculates the total element space required for the tensor in memory.
std::size_t	get_length (std::size_t dim) const
const std::vector< std::size_t > &	get_lengths () const
std::size_t	get_stride (std::size_t dim) const
const std::vector< std::size_t > &	get_strides () const
template<typename... Is>
std::size_t	GetOffsetFromMultiIndex (Is... is) const
	Calculates the linear offset from multi-dimensional indices.
std::size_t	GetOffsetFromMultiIndex (const std::vector< std::size_t > &iss) const
	Calculates the linear memory offset from a multi-dimensional index.

Friends
std::ostream &	operator<< (std::ostream &os, const HostTensorDescriptor &desc)

Detailed Description

Descriptor for tensors in host memory.

HostTensorDescriptor manages the shape (dimensions) and memory layout (strides) of a tensor in host memory. It provides functionality to:

Store tensor dimensions and strides
Calculate default strides for contiguous memory layout
Convert multi-dimensional indices to linear memory offsets
Query tensor metadata (dimensions, element counts, etc.)

The class supports both automatic stride calculation for contiguous memory layout and custom strides for more complex memory patterns.

Constructor & Destructor Documentation

◆ HostTensorDescriptor() [1/5]

ck_tile::HostTensorDescriptor::HostTensorDescriptor ( )

default

◆ HostTensorDescriptor() [2/5]

template<typename X, typename = std::enable_if_t<std::is_convertible_v<X, std::size_t>>>

ck_tile::HostTensorDescriptor::HostTensorDescriptor ( const std::initializer_list< X > & lens )

inline

◆ HostTensorDescriptor() [3/5]

template<typename Lengths, typename = std::enable_if_t< std::is_convertible_v<ck_tile::ranges::range_value_t<Lengths>, std::size_t>>>

ck_tile::HostTensorDescriptor::HostTensorDescriptor ( const Lengths & lens )

inline

◆ HostTensorDescriptor() [4/5]

template<typename X, typename Y, typename = std::enable_if_t<std::is_convertible_v<X, std::size_t> && std::is_convertible_v<Y, std::size_t>>>

ck_tile::HostTensorDescriptor::HostTensorDescriptor	(	const std::initializer_list< X > &	lens,
		const std::initializer_list< Y > &	strides )

inline

◆ HostTensorDescriptor() [5/5]

template<typename Lengths, typename Strides, typename = std::enable_if_t< std::is_convertible_v<ck_tile::ranges::range_value_t<Lengths>, std::size_t> && std::is_convertible_v<ck_tile::ranges::range_value_t<Strides>, std::size_t>>>

ck_tile::HostTensorDescriptor::HostTensorDescriptor	(	const Lengths &	lens,
		const Strides &	strides )

inline

Member Function Documentation

◆ CalculateStrides()

void ck_tile::HostTensorDescriptor::CalculateStrides ( )

inline

◆ get_element_size()

std::size_t ck_tile::HostTensorDescriptor::get_element_size ( ) const

inline

Calculates the total number of elements in the tensor.

Computes the product of all dimension lengths to determine the total element count in the tensor.

Precondition: The lengths array (mLens) and strides array (mStrides) must have the same size.

Returns: The total number of elements in the tensor.

◆ get_element_space_size()

std::size_t ck_tile::HostTensorDescriptor::get_element_space_size ( ) const

inline

Calculates the total element space required for the tensor in memory.

This method computes the minimum size of contiguous memory needed to store all elements of the tensor, taking into account the tensor's dimensions and strides. The calculation is based on the formula: 1 + max((length_i - 1) * stride_i) across all dimensions.

Dimensions with length 0 are skipped in this calculation.

Returns: The size of the tensor's element space (number of elements).

◆ get_length()

std::size_t ck_tile::HostTensorDescriptor::get_length ( std::size_t dim ) const

inline

◆ get_lengths()

const std::vector< std::size_t > & ck_tile::HostTensorDescriptor::get_lengths ( ) const

inline

◆ get_num_of_dimension()

std::size_t ck_tile::HostTensorDescriptor::get_num_of_dimension ( ) const

inline

◆ get_stride()

std::size_t ck_tile::HostTensorDescriptor::get_stride ( std::size_t dim ) const

inline

◆ get_strides()

const std::vector< std::size_t > & ck_tile::HostTensorDescriptor::get_strides ( ) const

inline

◆ GetOffsetFromMultiIndex() [1/2]

std::size_t ck_tile::HostTensorDescriptor::GetOffsetFromMultiIndex ( const std::vector< std::size_t > & iss ) const

inline

Calculates the linear memory offset from a multi-dimensional index.

Computes the linear offset by performing an inner product between the provided multi-dimensional indices and the tensor's strides.

Parameters

iss	Vector containing the multi-dimensional indices

Returns: The calculated linear offset as a size_t

◆ GetOffsetFromMultiIndex() [2/2]

template<typename... Is>

std::size_t ck_tile::HostTensorDescriptor::GetOffsetFromMultiIndex ( Is... is ) const

inline

Calculates the linear offset from multi-dimensional indices.

Converts a set of N-dimensional indices into a single linear offset by computing the inner product of the indices with the tensor's strides.

Template Parameters

Is	Parameter pack of index types (should be convertible to std::size_t)

Parameters

is	Variable number of indices, one for each dimension of the tensor

Returns: std::size_t Linear offset corresponding to the given multi-dimensional indices

Precondition: The number of indices must match the number of dimensions in the tensor

◆ operator<<

std::ostream & operator<<	(	std::ostream &	os,
		const HostTensorDescriptor &	desc )

friend

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

tile/host/host_tensor.hpp

HostTensorDescriptor Struct Reference

HostTensorDescriptor Struct Reference#

Public Member Functions

Friends

Detailed Description

Constructor & Destructor Documentation

◆ HostTensorDescriptor() [1/5]

◆ HostTensorDescriptor() [2/5]

◆ HostTensorDescriptor() [3/5]

◆ HostTensorDescriptor() [4/5]

◆ HostTensorDescriptor() [5/5]

Member Function Documentation

◆ CalculateStrides()

◆ get_element_size()

◆ get_element_space_size()

◆ get_length()

◆ get_lengths()

◆ get_num_of_dimension()

◆ get_stride()

◆ get_strides()

◆ GetOffsetFromMultiIndex() [1/2]

◆ GetOffsetFromMultiIndex() [2/2]

Friends And Related Symbol Documentation

◆ operator<<