hamr::buffer_handle< T > Class Template Reference

#include <hamr_buffer_handle.h>

Collaboration diagram for hamr::buffer_handle< T >:

Public Member Functions
	buffer_handle ()
	construct an empty, and unusable object More...
	buffer_handle (const std::shared_ptr< T > &src, size_t size, int read_only, int host_accessible, int cuda_accessible, size_t stream)
	construct from existing data More...
	buffer_handle (const std::shared_ptr< const T > &src, size_t size, int host_accessible, int cuda_accessible, size_t stream)
	construct from existing read only data More...
	~buffer_handle ()
	destruct More...
	buffer_handle (const buffer_handle< T > &other)
	copy construct More...
	buffer_handle (buffer_handle< T > &&other)
	move construct More...
buffer_handle< T > &	operator= (const buffer_handle< T > &)=default
	copy assign More...
buffer_handle< T > &	operator= (buffer_handle< T > &&)=default
	move assign More...
PyObject *	get_cuda_array_interface ()
PyObject *	get_numpy_array_interface ()
PyObject *	get_array_interface ()
void	to_stream (std::ostream &os) const

Public Attributes
std::shared_ptr< T >	m_data
size_t	m_size
int	m_read_only
int	m_host_accessible
int	m_cuda_accessible
size_t	m_stream

Detailed Description

template<typename T>
class hamr::buffer_handle< T >

A resource management class that is used to keep data shared with Python codes from a ::hamr_buffer alive while it is being accessed by the Python codes. The class also implements the Numpy array interface protocol and the Numba CUDA array interface protocol enabling seamless access by those libraries.

Constructor & Destructor Documentation

buffer_handle() [1/5]

template<typename T >

hamr::buffer_handle< T >::buffer_handle ( )

inline

construct an empty, and unusable object

buffer_handle() [2/5]

template<typename T >

hamr::buffer_handle< T >::buffer_handle	(	const std::shared_ptr< T > &	src,
		size_t	size,
		int	read_only,
		int	host_accessible,
		int	cuda_accessible,
		size_t	stream
	)

construct from existing data

buffer_handle() [3/5]

template<typename T >

hamr::buffer_handle< T >::buffer_handle	(	const std::shared_ptr< const T > &	src,
		size_t	size,
		int	host_accessible,
		int	cuda_accessible,
		size_t	stream
	)

construct from existing read only data

~buffer_handle()

template<typename T >

hamr::buffer_handle< T >::~buffer_handle

destruct

buffer_handle() [4/5]

template<typename T >

hamr::buffer_handle< T >::buffer_handle

(

const buffer_handle< T > &

other

)

copy construct

buffer_handle() [5/5]

template<typename T >

hamr::buffer_handle< T >::buffer_handle

(

buffer_handle< T > &&

other

)

move construct

Member Function Documentation

get_array_interface()

template<typename T >

PyObject * hamr::buffer_handle< T >::get_array_interface

returns a dictionary as decribed in the Numpy array interface and Numba CUDA array interface protocols. The caller must ensure the data is accessible in the desired technology before use. See ::get_numpy_array_interface and ::get_cuda_array_interface which implement check ensuring that this is true.

get_cuda_array_interface()

template<typename T >

PyObject * hamr::buffer_handle< T >::get_cuda_array_interface

returns a dictionary as described in the Numba CUDA array interface protocol if the data is accessible from CUDA. Otherwise, an AttributeError is rasied.

get_numpy_array_interface()

template<typename T >

PyObject * hamr::buffer_handle< T >::get_numpy_array_interface

returns a dictionary as described in the Numpy array_interface protocol if the data is accessible from CUDA. Otherwise, an AttributeError is rasied.

operator=() [1/2]

template<typename T >

buffer_handle<T>& hamr::buffer_handle< T >::operator= ( buffer_handle< T > &&  )

default

move assign

operator=() [2/2]

template<typename T >

buffer_handle<T>& hamr::buffer_handle< T >::operator= ( const buffer_handle< T > &  )

default

copy assign

---

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

• stable/hamr_buffer_handle.h