plssvm::cuda::detail Namespace Reference

Namespace containing CUDA backend specific implementation details. Should not directly be used by users. More...

Classes
class	device_ptr
	Small wrapper class around a CUDA device pointer together with commonly used device functions. More...

Functions
template<typename value_type , typename size_type >
__global__ void	fill_array (value_type *data, const value_type value, const size_type pos, const size_type count)
	Fill count values of the array data with the value starting at position pos. More...
void	gpu_assert (cudaError_t code)
	Check the CUDA error code. If code signals an error, throw a plssvm::cuda::backend_exception. More...
int	get_device_count ()
	Returns the number of available CUDA devices. More...
void	set_device (int device)
	Set the device to the active CUDA device. More...
void	peek_at_last_error ()
	Returns the last error from a CUDA runtime call.
void	device_synchronize (int device)
	Wait for the compute device to finish. More...

Detailed Description

Namespace containing CUDA backend specific implementation details. Should not directly be used by users.

Function Documentation

fill_array()

template<typename value_type , typename size_type >

__global__ void plssvm::cuda::detail::fill_array	(	value_type *	data,
		const value_type	value,
		const size_type	pos,
		const size_type	count
	)

Fill count values of the array data with the value starting at position pos.

Template Parameters

value_type	the type of the array and fill value
size_type	the size type

Parameters

[out]	data	the array to fill
[in]	value	the value to fill the array with
[in]	pos	the position to start filling the array
[in]	count	the number of values to fill

gpu_assert()

void plssvm::cuda::detail::gpu_assert

(

cudaError_t

code

)

Check the CUDA error code. If code signals an error, throw a plssvm::cuda::backend_exception.

The exception contains the following message: "CUDA assert 'CUDA_ERROR_NAME' (CUDA_ERROR_CODE): CUDA_ERROR_STRING".

Parameters

[in]

code

the CUDA error code to check

Exceptions

plssvm::cuda::backend_exception

if the error code signals a failure

get_device_count()

int plssvm::cuda::detail::get_device_count

(

)

Returns the number of available CUDA devices.

Returns

the number of devices ([[nodiscard]])

set_device()

void plssvm::cuda::detail::set_device

(

int

device

)

Set the device to the active CUDA device.

Parameters

[in]

device

the now active device

Exceptions

plssvm::cuda::backend_exception

if the given device ID is smaller than 0 or greater or equal than the available number of devices

device_synchronize()

void plssvm::cuda::detail::device_synchronize

(

int

device

)

Wait for the compute device to finish.

Calls plssvm::cuda::detail::peek_at_last_error() before synchronizing.

Parameters

[in]

device

the CUDA device to synchronize

Exceptions

plssvm::cuda::backend_exception

if the given device ID is smaller than 0 or greater or equal than the available number of devices