asrl::CudaSynchronizedMemory< T > Class Template Reference

a class that handles synchronization of memory between the host and device More...

#include <CudaSynchronizedMemory.hpp>

Public Types
typedef T	value_type
	The underlying type of the allocated memory.
Public Member Functions
	CudaSynchronizedMemory ()
	CudaSynchronizedMemory (unsigned N, bool pageLocked=false)
	~CudaSynchronizedMemory ()
void	init (unsigned N, bool pageLocked=false)
void	reset ()
T *	begin ()
T *	end ()
T *	h_get ()
T const *	h_get () const
T *	d_get ()
T const *	d_get () const
size_t	size () const
T &	operator[] (size_t i)
T const &	operator[] (size_t i) const
void	pushToDevice (size_t nElements=std::numeric_limits< size_t >::max())
void	pullFromDevice (size_t nElements=std::numeric_limits< size_t >::max())
void	pullFromDeviceAsync (cudaStream_t stream, size_t nElements=std::numeric_limits< size_t >::max())
void	memsetHost (int val)
void	memsetDevice (int val)
void	memset (int val)
bool	isPageLocked ()
Private Member Functions
	CudaSynchronizedMemory (CudaSynchronizedMemory const &rhs)
CudaSynchronizedMemory &	operator= (CudaSynchronizedMemory const &rhs)
void	freeHostMemory ()
void	freeDeviceMemory ()
Private Attributes
T *	m_host
	A reference-counted pointer to the host memory.
T *	m_device
	A reference-counted pointer to the device memory.
size_t	m_size
	The number of elements of type T allocated.
bool	m_pageLocked
	Is the host memory page-locked.

Detailed Description

template<typename T>
class asrl::CudaSynchronizedMemory< T >

a class that handles synchronization of memory between the host and device

A class that simplifies allocation/deallocation and synchronization of a buffer allocated on the host and on the device.

Definition at line 60 of file CudaSynchronizedMemory.hpp.

Member Typedef Documentation

template<typename T>

typedef T asrl::CudaSynchronizedMemory< T >::value_type

The underlying type of the allocated memory.

Definition at line 75 of file CudaSynchronizedMemory.hpp.

Constructor & Destructor Documentation

template<typename T>

asrl::CudaSynchronizedMemory< T >::CudaSynchronizedMemory ( CudaSynchronizedMemory< T > const & rhs ) [private]

The object is non-copyable as we need to be careful about memory leaks.

template<typename T >

asrl::CudaSynchronizedMemory< T >::CudaSynchronizedMemory ( ) [inline]

A constructor that initializes both the host and device pointers with null values

Definition at line 247 of file CudaSynchronizedMemory.hpp.

template<typename T >

asrl::CudaSynchronizedMemory< T >::CudaSynchronizedMemory	(	unsigned	N,
		bool	pageLocked = `false`
	)			`[inline]`

A constructor that allocates memory for N elements of type T on both the host and the device.

Parameters:

	N	The number of elements to allocate
	pageLocked	Should the host memory be page-locked

Definition at line 250 of file CudaSynchronizedMemory.hpp.

template<typename T >

asrl::CudaSynchronizedMemory< T >::~CudaSynchronizedMemory ( ) [inline]

A destructor to clean up the allocated memory

Definition at line 256 of file CudaSynchronizedMemory.hpp.

Member Function Documentation

template<typename T >

T * asrl::CudaSynchronizedMemory< T >::begin ( ) [inline]

Returns:: a pointer to the beginning of the host memory

Definition at line 379 of file CudaSynchronizedMemory.hpp.

template<typename T >

T const * asrl::CudaSynchronizedMemory< T >::d_get ( ) const [inline]

Returns:: the device memory pointer

Definition at line 402 of file CudaSynchronizedMemory.hpp.

template<typename T >

T * asrl::CudaSynchronizedMemory< T >::d_get ( ) [inline]

Returns:: the device memory pointer

Definition at line 399 of file CudaSynchronizedMemory.hpp.

template<typename T >

T * asrl::CudaSynchronizedMemory< T >::end ( ) [inline]

Returns:: a pointer to the end of the host memory

Definition at line 386 of file CudaSynchronizedMemory.hpp.

template<typename T >

void asrl::CudaSynchronizedMemory< T >::freeDeviceMemory ( ) [inline, private]

Definition at line 299 of file CudaSynchronizedMemory.hpp.

template<typename T >

void asrl::CudaSynchronizedMemory< T >::freeHostMemory ( ) [inline, private]

Definition at line 279 of file CudaSynchronizedMemory.hpp.

template<typename T >

T const * asrl::CudaSynchronizedMemory< T >::h_get ( ) const [inline]

Returns:: the host memory pointer

Definition at line 396 of file CudaSynchronizedMemory.hpp.

template<typename T >

T * asrl::CudaSynchronizedMemory< T >::h_get ( ) [inline]

Returns:: the host memory pointer

Definition at line 393 of file CudaSynchronizedMemory.hpp.

template<typename T >

void asrl::CudaSynchronizedMemory< T >::init	(	unsigned	N,
		bool	pageLocked = `false`
	)			`[inline]`

A function that allocates memory for N elements of type T on both the host and the device.

Parameters:

	N	The number of elements to allocate
	pageLocked	Should the host memory be page-locked

Definition at line 339 of file CudaSynchronizedMemory.hpp.

template<typename T >

bool asrl::CudaSynchronizedMemory< T >::isPageLocked ( ) [inline]

Returns:: returns true if the underlying host memory is page-locked

Definition at line 373 of file CudaSynchronizedMemory.hpp.

template<typename T >

void asrl::CudaSynchronizedMemory< T >::memset ( int val ) [inline]

Calls memset on both the host and device memory.

Definition at line 315 of file CudaSynchronizedMemory.hpp.

template<typename T >

void asrl::CudaSynchronizedMemory< T >::memsetDevice ( int val ) [inline]

calls cudaMemset() on the device memory.

Definition at line 330 of file CudaSynchronizedMemory.hpp.

template<typename T >

void asrl::CudaSynchronizedMemory< T >::memsetHost ( int val ) [inline]

calls memset() on the host memory.

Definition at line 323 of file CudaSynchronizedMemory.hpp.

template<typename T>

CudaSynchronizedMemory& asrl::CudaSynchronizedMemory< T >::operator= ( CudaSynchronizedMemory< T > const & rhs ) [private]

The object is non-copyable as we need to be careful about memory leaks.

template<typename T >

T const & asrl::CudaSynchronizedMemory< T >::operator[] ( size_t i ) const [inline]

An convenience operator that dereferences the host array

Parameters:

i

the index to dereference the host array

Returns:: a const reference to the value at host_memory[i]

Definition at line 415 of file CudaSynchronizedMemory.hpp.

template<typename T >

T & asrl::CudaSynchronizedMemory< T >::operator[] ( size_t i ) [inline]

An convenience operator that dereferences the host array

Parameters:

i

the index to dereference the host array

Returns:: a reference to the value at host_memory[i]

Definition at line 408 of file CudaSynchronizedMemory.hpp.

template<typename T >

void asrl::CudaSynchronizedMemory< T >::pullFromDevice ( size_t nElements = std::numeric_limits<size_t>::max() ) [inline]

Pulls bytes from the device to the host.

Parameters:

nElements

The number of elements of the array to pull. If this is greater than size() or -1, all allocated elements are pulled from the device.

Definition at line 433 of file CudaSynchronizedMemory.hpp.

template<typename T >

void asrl::CudaSynchronizedMemory< T >::pullFromDeviceAsync	(	cudaStream_t	stream,
		size_t	nElements = `std::numeric_limits<size_t>::max()`
	)			`[inline]`

Initiate an asynchronous transfer from the host to the device.

Parameters:

	stream	The cuda stream associated with the transfer.
	nElements	The number of elements to pull. If this is greater than size() or -1, all allocated elements are pulled from the device.

Definition at line 443 of file CudaSynchronizedMemory.hpp.

template<typename T >

void asrl::CudaSynchronizedMemory< T >::pushToDevice ( size_t nElements = std::numeric_limits<size_t>::max() ) [inline]