#include <device_array.hpp>

Inheritance diagram for kfusion::cuda::DeviceArray2D< T >:

Public Types
enum	{ elem_size = sizeof(T) }
	Element size. More...

typedef T	type
	Element type. More...

Public Member Functions
int	cols () const
	Returns number of elements in each row. More...

void	copyTo (DeviceArray2D &other) const
	Performs data copying. If destination size differs it will be reallocated. More...

void	create (int rows, int cols)
	Allocates internal buffer in GPU memory. If internal buffer was created before the function recreates it with new size. If new and old sizes are equal it does nothing. More...

	DeviceArray2D ()
	Empty constructor. More...

	DeviceArray2D (const DeviceArray2D &other)
	Copy constructor. Just increments reference counter. More...

	DeviceArray2D (int rows, int cols)
	Allocates internal buffer in GPU memory. More...

	DeviceArray2D (int rows, int cols, void *data, size_t stepBytes)
	Initializes with user allocated buffer. Reference counting is disabled in this case. More...

template<class A >
void	download (std::vector< T, A > &data, int &cols) const
	Downloads data from internal buffer to CPU memory. More...

void	download (void *host_ptr, size_t host_step) const
	Downloads data from internal buffer to CPU memory. User is resposible for correct host buffer size. More...

size_t	elem_step () const
	Returns step in elements. More...

	operator const T * () const
	Returns const pointer for internal buffer in GPU memory. More...

	operator T* ()
	Returns pointer for internal buffer in GPU memory. More...

DeviceArray2D &	operator= (const DeviceArray2D &other)
	Assigment operator. Just increments reference counter. More...

T *	ptr (int y=0)
	Returns pointer to given row in internal buffer. More...

const T *	ptr (int y=0) const
	Returns const pointer to given row in internal buffer. More...

void	release ()
	Decrements reference counter and releases internal buffer if needed. More...

int	rows () const
	Returns number of rows. More...

void	swap (DeviceArray2D &other_arg)
	Performs swap of data pointed with another device array. More...

template<class A >
void	upload (const std::vector< T, A > &data, int cols)
	Uploads data to internal buffer in GPU memory. It calls create() inside to ensure that intenal buffer size is enough. More...

void	upload (const void *host_ptr, size_t host_step, int rows, int cols)
	Uploads data to internal buffer in GPU memory. It calls create() inside to ensure that intenal buffer size is enough. More...

Public Member Functions inherited from kfusion::cuda::DeviceMemory2D
int	colsBytes () const
	Returns number of bytes in each row. More...

void	copyTo (DeviceMemory2D &other) const
	Performs data copying. If destination size differs it will be reallocated. More...

void	create (int rows_arg, int colsBytes_arg)
	Allocates internal buffer in GPU memory. If internal buffer was created before the function recreates it with new size. If new and old sizes are equal it does nothing. More...

	DeviceMemory2D ()
	Empty constructor. More...

	DeviceMemory2D (const DeviceMemory2D &other_arg)
	Copy constructor. Just increments reference counter. More...

	DeviceMemory2D (int rows_arg, int colsBytes_arg)
	Allocates internal buffer in GPU memory. More...

	DeviceMemory2D (int rows_arg, int colsBytes_arg, void *data_arg, size_t step_arg)
	Initializes with user allocated buffer. Reference counting is disabled in this case. More...

void	download (void *host_ptr_arg, size_t host_step_arg) const
	Downloads data from internal buffer to CPU memory. User is resposible for correct host buffer size. More...

bool	empty () const
	Returns true if unallocated otherwise false. More...

template<class U >
	operator PtrStep< U > () const
	Conversion to PtrStep for passing to kernel functions. More...

template<class U >
	operator PtrStepSz< U > () const
	Conversion to PtrStepSz for passing to kernel functions. More...

DeviceMemory2D &	operator= (const DeviceMemory2D &other_arg)
	Assigment operator. Just increments reference counter. More...

template<class T >
T *	ptr (int y_arg=0)
	Returns pointer to given row in internal buffer. More...

template<class T >
const T *	ptr (int y_arg=0) const
	Returns constant pointer to given row in internal buffer. More...

void	release ()
	Decrements reference counter and releases internal buffer if needed. More...

int	rows () const
	Returns number of rows. More...

size_t	step () const
	Returns stride between two consecutive rows in bytes for internal buffer. Step is stored always and everywhere in bytes!!! More...

void	swap (DeviceMemory2D &other_arg)
	Performs swap of data pointed with another device memory. More...

void	upload (const void *host_ptr_arg, size_t host_step_arg, int rows_arg, int colsBytes_arg)
	Uploads data to internal buffer in GPU memory. It calls create() inside to ensure that intenal buffer size is enough. More...

	~DeviceMemory2D ()
	Destructor. More...

Detailed Description

template<class T>
class kfusion::cuda::DeviceArray2D< T >

DeviceArray2D class

Note: Typed container for pitched GPU memory with reference counting.

Author: Anatoly Baksheev

Definition at line 117 of file device_array.hpp.

Member Typedef Documentation

◆ type

template<class T >

typedef T kfusion::cuda::DeviceArray2D< T >::type

Element type.

Definition at line 121 of file device_array.hpp.

Member Enumeration Documentation

◆ anonymous enum

template<class T >

anonymous enum

Element size.

Enumerator
elem_size

Definition at line 124 of file device_array.hpp.

Constructor & Destructor Documentation

◆ DeviceArray2D() [1/4]

template<class T >

kfusion::cuda::DeviceArray2D< T >::DeviceArray2D

inline

Empty constructor.

Definition at line 267 of file device_array.hpp.

◆ DeviceArray2D() [2/4]

template<class T >

kfusion::cuda::DeviceArray2D< T >::DeviceArray2D	(	int	rows,
		int	cols
	)

inline

Allocates internal buffer in GPU memory.

Parameters

rows	number of rows to allocate
cols	number of elements in each row

Definition at line 268 of file device_array.hpp.

◆ DeviceArray2D() [3/4]

template<class T >

kfusion::cuda::DeviceArray2D< T >::DeviceArray2D	(	int	rows,
		int	cols,
		void *	data,
		size_t	stepBytes
	)

inline

Initializes with user allocated buffer. Reference counting is disabled in this case.

Parameters

rows	number of rows
cols	number of elements in each row
data	pointer to buffer
stepBytes	stride between two consecutive rows in bytes

Definition at line 269 of file device_array.hpp.

◆ DeviceArray2D() [4/4]

template<class T >

kfusion::cuda::DeviceArray2D< T >::DeviceArray2D ( const DeviceArray2D< T > & other )

inline

Copy constructor. Just increments reference counter.

Definition at line 270 of file device_array.hpp.

Member Function Documentation

◆ cols()

template<class T >

int kfusion::cuda::DeviceArray2D< T >::cols

inline

Returns number of elements in each row.

Definition at line 300 of file device_array.hpp.

◆ copyTo()

template<class T >

void kfusion::cuda::DeviceArray2D< T >::copyTo ( DeviceArray2D< T > & other ) const

inline

Performs data copying. If destination size differs it will be reallocated.

Parameters

other destination container

Definition at line 279 of file device_array.hpp.

◆ create()

template<class T >

void kfusion::cuda::DeviceArray2D< T >::create	(	int	rows,
		int	cols
	)

inline

Allocates internal buffer in GPU memory. If internal buffer was created before the function recreates it with new size. If new and old sizes are equal it does nothing.

Parameters

rows	number of rows to allocate
cols	number of elements in each row

Definition at line 274 of file device_array.hpp.

◆ download() [1/2]

template<class T >

template<class A >

void kfusion::cuda::DeviceArray2D< T >::download	(	std::vector< T, A > &	data,
		int &	cols
	)		const

inline

Downloads data from internal buffer to CPU memory.

Parameters

data	host vector to download to
cols	Output stride in elements between two consecutive rows for host vector.

Definition at line 289 of file device_array.hpp.

◆ download() [2/2]

template<class T >

void kfusion::cuda::DeviceArray2D< T >::download	(	void *	host_ptr,
		size_t	host_step
	)		const

inline

Downloads data from internal buffer to CPU memory. User is resposible for correct host buffer size.

Parameters

host_ptr	pointer to host buffer to download
host_step	stride between two consecutive rows in bytes for host buffer

Definition at line 283 of file device_array.hpp.

◆ elem_step()

template<class T >

size_t kfusion::cuda::DeviceArray2D< T >::elem_step

inline

Returns step in elements.

Definition at line 303 of file device_array.hpp.

◆ operator const T *()

template<class T >

kfusion::cuda::DeviceArray2D< T >::operator const T *

inline

Returns const pointer for internal buffer in GPU memory.

Definition at line 298 of file device_array.hpp.

◆ operator T*()

template<class T >

kfusion::cuda::DeviceArray2D< T >::operator T*

inline

Returns pointer for internal buffer in GPU memory.

Definition at line 297 of file device_array.hpp.

◆ operator=()

template<class T >

kfusion::cuda::DeviceArray2D< T > & kfusion::cuda::DeviceArray2D< T >::operator= ( const DeviceArray2D< T > & other )

inline

Assigment operator. Just increments reference counter.

Definition at line 271 of file device_array.hpp.

◆ ptr() [1/2]

template<class T >

T * kfusion::cuda::DeviceArray2D< T >::ptr ( int y = 0 )

inline

Returns pointer to given row in internal buffer.

Parameters

y_arg row index

Definition at line 294 of file device_array.hpp.

◆ ptr() [2/2]

template<class T >

const T * kfusion::cuda::DeviceArray2D< T >::ptr ( int y = 0 ) const

inline

Returns const pointer to given row in internal buffer.

Parameters

y_arg row index

Definition at line 295 of file device_array.hpp.

◆ release()

template<class T >

void kfusion::cuda::DeviceArray2D< T >::release

inline

Decrements reference counter and releases internal buffer if needed.

Definition at line 276 of file device_array.hpp.

◆ rows()

template<class T >

int kfusion::cuda::DeviceArray2D< T >::rows

inline

Returns number of rows.

Definition at line 301 of file device_array.hpp.

◆ swap()

template<class T >

void kfusion::cuda::DeviceArray2D< T >::swap ( DeviceArray2D< T > & other_arg )

Performs swap of data pointed with another device array.

Parameters

other device array to swap with

Definition at line 292 of file device_array.hpp.

◆ upload() [1/2]

template<class T >

template<class A >

void kfusion::cuda::DeviceArray2D< T >::upload	(	const std::vector< T, A > &	data,
		int	cols
	)

inline

Uploads data to internal buffer in GPU memory. It calls create() inside to ensure that intenal buffer size is enough.

Parameters

data	host vector to upload from
cols	stride in elements between two consecutive rows for host buffer

Definition at line 286 of file device_array.hpp.

◆ upload() [2/2]

template<class T >

void kfusion::cuda::DeviceArray2D< T >::upload	(	const void *	host_ptr,
		size_t	host_step,
		int	rows,
		int	cols
	)

inline

Uploads data to internal buffer in GPU memory. It calls create() inside to ensure that intenal buffer size is enough.

Parameters

host_ptr	pointer to host buffer to upload
host_step	stride between two consecutive rows in bytes for host buffer
rows	number of rows to upload
cols	number of elements in each row

Definition at line 281 of file device_array.hpp.

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

device_array.hpp

Public Types

Public Member Functions

Detailed Description

template<class T> class kfusion::cuda::DeviceArray2D< T >

Member Typedef Documentation

◆ type

Member Enumeration Documentation

◆ anonymous enum

Constructor & Destructor Documentation

◆ DeviceArray2D() [1/4]

◆ DeviceArray2D() [2/4]

◆ DeviceArray2D() [3/4]

◆ DeviceArray2D() [4/4]

Member Function Documentation

◆ cols()

◆ copyTo()

◆ create()

◆ download() [1/2]

◆ download() [2/2]

◆ elem_step()

◆ operator const T *()

◆ operator T*()

◆ operator=()

◆ ptr() [1/2]

◆ ptr() [2/2]

◆ release()

◆ rows()

◆ swap()

◆ upload() [1/2]

◆ upload() [2/2]

template<class T>
class kfusion::cuda::DeviceArray2D< T >