RooBatchCompute::CudaInterface Namespace Reference

Classes
class	Array
	A templated class for managing an array of data using a specified memory type. More...
class	CudaEvent
class	CudaStream

Typedefs
template<class T >
using	DeviceArray = Array< T, DeviceMemory >
	An array of specific type that is allocated on the device with cudaMalloc and freed with cudaFree.
template<class T >
using	PinnedHostArray = Array< T, PinnedHostMemory >
	A pinned array of specific type that allocated on the host with cudaMallocHost and freed with cudaFreeHost.

Functions
template<class T >
void	copyDeviceToDevice (const T *src, T *dest, std::size_t n, CudaStream *=nullptr)
	Copies data from the CUDA device to the CUDA device.
template<class T >
void	copyDeviceToHost (const T *src, T *dest, std::size_t n, CudaStream *=nullptr)
	Copies data from the CUDA device to the host.
template<class T >
void	copyHostToDevice (const T *src, T *dest, std::size_t n, CudaStream *=nullptr)
	Copies data from the host to the CUDA device.
float	cudaEventElapsedTime (CudaEvent &begin, CudaEvent &end)
	Calculates the elapsed time between two CUDA events.
void	cudaEventRecord (CudaEvent &event, CudaStream &stream)
	Records a CUDA event.

Typedef Documentation

DeviceArray

template<class T >

using RooBatchCompute::CudaInterface::DeviceArray = typedef Array<T, DeviceMemory>

An array of specific type that is allocated on the device with cudaMalloc and freed with cudaFree.

Definition at line 209 of file CudaInterface.h.

PinnedHostArray

template<class T >

using RooBatchCompute::CudaInterface::PinnedHostArray = typedef Array<T, PinnedHostMemory>

A pinned array of specific type that allocated on the host with cudaMallocHost and freed with cudaFreeHost.

The memory is "pinned", i.e. page-locked and accessible to the device for fast copying.

See also

The documentation of cudaMallocHost on developer.download.nvidia.com.

Definition at line 218 of file CudaInterface.h.

Function Documentation

copyDeviceToDevice()

template<class T >

void RooBatchCompute::CudaInterface::copyDeviceToDevice	(	const T *	src,
		T *	dest,
		std::size_t	n,
		CudaStream *	= nullptr
	)

Copies data from the CUDA device to the CUDA device.

Parameters

[in]	src	Pointer to the source memory on the device.
[in]	dest	Pointer to the destination memory on the device.
[in]	nBytes	Number of bytes to copy.
[in]	stream	CudaStream for asynchronous memory transfer (optional).

Definition at line 120 of file CudaInterface.h.

copyDeviceToHost()

template<class T >

void RooBatchCompute::CudaInterface::copyDeviceToHost	(	const T *	src,
		T *	dest,
		std::size_t	n,
		CudaStream *	= nullptr
	)

Copies data from the CUDA device to the host.

Parameters

[in]	src	Pointer to the source memory on the device.
[in]	dest	Pointer to the destination memory on the host.
[in]	nBytes	Number of bytes to copy.
[in]	stream	CudaStream for asynchronous memory transfer (optional).

Definition at line 106 of file CudaInterface.h.

copyHostToDevice()

template<class T >

void RooBatchCompute::CudaInterface::copyHostToDevice	(	const T *	src,
		T *	dest,
		std::size_t	n,
		CudaStream *	= nullptr
	)

Copies data from the host to the CUDA device.

Parameters

[in]	src	Pointer to the source memory on the host.
[in]	dest	Pointer to the destination memory on the device.
[in]	nBytes	Number of bytes to copy.
[in]	stream	CudaStream for asynchronous memory transfer (optional).

Definition at line 92 of file CudaInterface.h.

cudaEventElapsedTime()

float RooBatchCompute::CudaInterface::cudaEventElapsedTime	(	CudaEvent &	begin,
		CudaEvent &	end
	)

Calculates the elapsed time between two CUDA events.

Parameters

[in]	begin	CudaEvent representing the start event.
[in]	end	CudaEvent representing the end event.

Returns

Elapsed time in milliseconds.

Definition at line 146 of file CudaInterface.cu.

cudaEventRecord()

void RooBatchCompute::CudaInterface::cudaEventRecord	(	CudaEvent &	event,
		CudaStream &	stream
	)

Records a CUDA event.

Parameters

[in]	event	CudaEvent object representing the event to be recorded.
[in]	stream	CudaStream in which to record the event.

Definition at line 96 of file CudaInterface.cu.