ActsPlugins::Tensor< T > Class Template Reference

This is a very small, limited class that models a 2D tensor of arbitrary type. More...

#include <ActsPlugins/Gnn/Tensor.hpp>

Inheritance diagram for ActsPlugins::Tensor< T >:

Public Types
using	Shape = std::array<std::size_t, 2>
	Type alias for tensor shape as 2D dimensions array.

Public Member Functions
Tensor	clone (const ExecutionContext &to) const
	Clone the tensor, copying the data to the new device.
T *	data ()
	Get the non-const data pointer.
const T *	data () const
	Get the const data pointer.
Device	device () const
	Get the device of the tensor.
std::size_t	nbytes () const
	Get the number of bytes of the tensor.
Shape	shape () const
	Get the shape of the tensor.
std::size_t	size () const
	Get the number of elements in the tensor.

Static Public Member Functions
static Tensor	Create (Shape shape, const ExecutionContext &execContext)
	Create a new tensor with specified shape and execution context.

Detailed Description

template<Acts::Concepts::arithmetic T>
class ActsPlugins::Tensor< T >

This is a very small, limited class that models a 2D tensor of arbitrary type.

It is move-only, and only possible to create via static factory functions to ensure lifetime management. This on purpose does not implement operations such as clone/to-host/to-cuda

Member Typedef Documentation

Shape

template<Acts::Concepts::arithmetic T>

using ActsPlugins::Tensor< T >::Shape = std::array<std::size_t, 2>

Type alias for tensor shape as 2D dimensions array.

Member Function Documentation

clone()

template<Acts::Concepts::arithmetic T>

Tensor ActsPlugins::Tensor< T >::clone

(

const ExecutionContext &

to

)

const

Clone the tensor, copying the data to the new device.

Parameters

to	The {device, stream} to clone to

Note

This is a always a deep copy, even if the source and destination are the same device

Returns

New tensor on target device with copied data

Create()

template<Acts::Concepts::arithmetic T>

Tensor ActsPlugins::Tensor< T >::Create ( Shape shape, const ExecutionContext & execContext )

static

Create a new tensor with specified shape and execution context.

Parameters

shape	2D tensor dimensions [rows, columns]
execContext	Execution context specifying device and optional CUDA stream

Returns

Newly created tensor with allocated memory on specified device

data() [1/2]

template<Acts::Concepts::arithmetic T>

T * ActsPlugins::Tensor< T >::data

(

)

Get the non-const data pointer.

Returns

Mutable pointer to tensor data

data() [2/2]

template<Acts::Concepts::arithmetic T>

const T * ActsPlugins::Tensor< T >::data

(

)

const

Get the const data pointer.

Returns

Const pointer to tensor data

device()

template<Acts::Concepts::arithmetic T>

Device ActsPlugins::Tensor< T >::device

(

)

const

Get the device of the tensor.

nbytes()

template<Acts::Concepts::arithmetic T>

std::size_t ActsPlugins::Tensor< T >::nbytes

(

)

const

Get the number of bytes of the tensor.

Returns

Total memory size of tensor data in bytes

shape()

template<Acts::Concepts::arithmetic T>

Shape ActsPlugins::Tensor< T >::shape

(

)

const

Get the shape of the tensor.

Returns

Array containing tensor dimensions [rows, columns]

size()

template<Acts::Concepts::arithmetic T>

std::size_t ActsPlugins::Tensor< T >::size

(

)

const

Get the number of elements in the tensor.

Returns

Total number of elements in the tensor