#include <streamed_parallel_for.hpp>

Public Member Functions
template<typename H , typename D >
	StreamView (H &view_h, D &view_d, bool stage_in_pinned_memory_in, int n_on_device, int n_partitions_of_device_view)

	~StreamView ()

template<typename ST >
void	copy_to_device (int offset_h, int offset_d, int n, int i_staged_area, ST &gpu_stream)

template<typename ST >
void	copy_to_host (int offset_h, int offset_d, int n, int i_staged_area, ST &gpu_stream)

void	copy_to_pinned (int offset_h, int n, int i_staging_area)

void	copy_from_pinned (int offset_h, int n, int i_staging_area)

Public Attributes
bool	stage_in_pinned_memory

T *	h_view_ptr

T *	d_view_ptr

T *	pinned_send [2]

T *	pinned_return [2]

Detailed Description

template<typename T>
struct Streamed::StreamView< T >

Each AoSoA or View that is to be streamed to device is handled as a StreamView. The StreamView handles the copies to and from device as well as to and from the pinned memory staging area

Constructor & Destructor Documentation

template<typename T>

template<typename H , typename D >

Streamed::StreamView< T >::StreamView	(	H &	view_h,
		D &	view_d,
		bool	stage_in_pinned_memory_in,
		int	n_on_device,
		int	n_partitions_of_device_view
	)

inline

Constructor

Parameters

[in]	view_h	is the host data, either an AoSoA or a View
[in]	view_d	is the device data allocation, either an AoSoA or a View
[in]	stage_in_pinned_memory_in	says whether to use pinned memory
[in]	n_on_device	is the size of the device data allocation
[in]	n_partitions_of_device_view	is the number of partitions to split the device allocation into

Here is the call graph for this function:

template<typename T>

Streamed::StreamView< T >::~StreamView ( )

inline

Custom destructor to deallocate the pinned memory

Member Function Documentation

template<typename T>

void Streamed::StreamView< T >::copy_from_pinned	(	int	offset_h,
		int	n,
		int	i_staging_area
	)

inline

Copies a chunk of data to the host memory from the temporary pinned memory staging area

Parameters

[in]	offset_h	is the offset where the data chunk begins in the full view
[in]	n	is the number of elements to be copied
[in]	i_staging_area	is which pinned host memory to use

Returns: void

Here is the caller graph for this function:

template<typename T>

template<typename ST >

void Streamed::StreamView< T >::copy_to_device	(	int	offset_h,
		int	offset_d,
		int	n,
		int	i_staged_area,
		ST &	gpu_stream
	)

inline

Copies a chunk of data to device by wrapping the specified section of the host and device allocations in unmanaged views. Could potentially use subviews instead.

Parameters

[in]	offset_h	is the offset where the data chunk begins in the full view
[in]	offset_d	is the offset where the data chunk begins in the full view
[in]	n	is the number of elements to be copied
[in]	i_staged_area	is which pinned host memory to use
[in]	gpu_stream	is the GPU stream used to execute the send

Returns: void

Here is the caller graph for this function:

template<typename T>

template<typename ST >

void Streamed::StreamView< T >::copy_to_host	(	int	offset_h,
		int	offset_d,
		int	n,
		int	i_staged_area,
		ST &	gpu_stream
	)

inline

Copies a chunk of data to the host by wrapping the specified section of the host and device allocations in unmanaged views. Could potentially use subviews instead.

Parameters

[in]	offset_h	is the offset where the data chunk begins in the full view
[in]	offset_d	is the offset where the data chunk begins in the full view
[in]	n	is the number of elements to be copied
[in]	i_staged_area	is which pinned host memory to use
[in]	gpu_stream	is the GPU stream used to execute the return

Returns: void

Here is the caller graph for this function:

template<typename T>

void Streamed::StreamView< T >::copy_to_pinned	(	int	offset_h,
		int	n,
		int	i_staging_area
	)

inline

Copies a chunk of data from the host memory to the temporary pinned memory staging area

Parameters

[in]	offset_h	is the offset where the data chunk begins in the full view
[in]	n	is the number of elements to be copied
[in]	i_staging_area	is which pinned host memory to use