cuda.cuh File Reference

#include "../core/common.hpp"
#include "cuda.inl"

Go to the source code of this file.

Namespaces
namespace	CellularAutomata
	Main namespace.

Functions
void	CellularAutomata::cuda::allocateCUDA (void **arrayPtr, size_t bytes)
	Allocated memory on a CUDA device.
void	CellularAutomata::cuda::allocateHost (void **arrayPtr, size_t bytes)
	Allocated memory on the host device (CPU)
void	CellularAutomata::cuda::freeCUDA (void *array)
	Frees memory on a CUDA device.
void	CellularAutomata::cuda::freeHost (void *array)
	Frees memory on the host device (CPU)
void	CellularAutomata::cuda::copyCUDA (void *src, Device from, void *dst, Device to, size_t bytes)
	Memcopy between devices and host.
template<typename T , typename Activation >
void	CellularAutomata::cuda::epoch (T *input, T *kernel, T *output, Activation fn, const unsigned int h, const unsigned int w, const unsigned int s, const bool r)
	Call the CUDA kernel to run a single epoch on a given state using a kernel.

Function Documentation

allocateCUDA()

void CellularAutomata::cuda::allocateCUDA	(	void **	arrayPtr,
		size_t	bytes
	)

Allocated memory on a CUDA device.

Parameters

arrayPtr	pointer to deviceArray
bytes	allocation size in bytes

Exceptions

CudaRuntime

allocateHost()

void CellularAutomata::cuda::allocateHost	(	void **	arrayPtr,
		size_t	bytes
	)

Allocated memory on the host device (CPU)

Note

Allocates "page-locked memory" which is slower in allocation but faster in transmission between devices and host. Should be used for memory which is often transfered

Parameters

arrayPtr	pointer to hostArray
bytes	allocation size in bytes

Exceptions

CudaRuntime

copyCUDA()

void CellularAutomata::cuda::copyCUDA	(	void *	src,
		Device	from,
		void *	dst,
		Device	to,
		size_t	bytes
	)

Memcopy between devices and host.

Attention

Size is assumed to be equal

Note

Host memory can be allocated using regular allocations

Parameters

src	input array to copy from
from	computational device where input lives
dst	output array to copy to
to	computational device where output lives
bytes	allocation size in bytes

Exceptions

CudaRuntime

epoch()

template<typename T , typename Activation >

void CellularAutomata::cuda::epoch	(	T *	input,
		T *	kernel,
		T *	output,
		Activation	fn,
		const unsigned int	h,
		const unsigned int	w,
		const unsigned int	s,
		const bool	r
	)

Call the CUDA kernel to run a single epoch on a given state using a kernel.

Template Parameters

T	type
Activation	functor

Parameters

input	current state array
kernel	kernel for convolution
output	next state array
fn	activation function
h	height of state array
w	width of state array
s	size of kernel
r	grid recursion enabled

Exceptions

CudaRuntime

freeCUDA()

void CellularAutomata::cuda::freeCUDA

(

void *

array

)

Frees memory on a CUDA device.

Parameters

array

deviceArray to free

Exceptions

CudaRuntime

freeHost()

void CellularAutomata::cuda::freeHost

(

void *

array

)

Frees memory on the host device (CPU)

Attention

Do NOT use on regular allocations

Note

Frees "page-locked memory" allocated with CUDA functions

Parameters

array

hostArray to free

Exceptions

CudaRuntime