GPUMemoryAllocation Namespace Reference

Functions
template<class T , bool ZeroInit = false>
T *	gpu_alloc (size_t size) noexcept
	Memory allocation for variables on the GPU.
template<class T >
void	gpu_free (T *ptr) noexcept
	Free memory allocated on the GPU with gpu_alloc.
template<class T >
T *	gpu_alloc_cpy (const T *src_ptr, size_t size) noexcept
	Memory allocation for variables on the GPU along with initialization from a source host array.
template<class T , bool ZeroInit = false>
T *	pinned_alloc (size_t size) noexcept
	Page-locked ("pinned") host memory allocation.
template<class T >
void	pinned_free (T *ptr) noexcept
	Free memory allocated with pinned_alloc.

Function Documentation

gpu_alloc()

template<class T , bool ZeroInit = false>

T * GPUMemoryAllocation::gpu_alloc ( size_t  size )

inlinenoexcept

Memory allocation for variables on the GPU.

Parameters

[in]

size

in bytes.

Template Parameters

ZeroInit,initialize

memory to 0.

Returns

Pointer to memory, always use gpu_free to deallocate.

gpu_alloc_cpy()

template<class T >

T * GPUMemoryAllocation::gpu_alloc_cpy ( const T *  src_ptr, size_t  size  )

inlinenoexcept

Memory allocation for variables on the GPU along with initialization from a source host array.

Parameters

[in]

size

in bytes.

Returns

Pointer to memory, always use gpu_free to deallocate.

gpu_free()

template<class T >

void GPUMemoryAllocation::gpu_free ( T *  ptr )

inlinenoexcept

Free memory allocated on the GPU with gpu_alloc.

Parameters

[in]

ptr,pointer

to memory we want to release.

pinned_alloc()

template<class T , bool ZeroInit = false>

T * GPUMemoryAllocation::pinned_alloc ( size_t  size )

inlinenoexcept

Page-locked ("pinned") host memory allocation.

Note

Unlike regular (pageable) host memory, cudaMemcpyAsync from/to a pinned buffer is actually asynchronous with respect to the host thread; from pageable memory the driver silently falls back to a synchronous staged copy. Only worth it for host buffers that are the source/destination of an async transfer meant to overlap with other host work.

Parameters

[in]

size

in bytes.

Template Parameters

ZeroInit,initialize

memory to 0.

Returns

Pointer to memory, always use pinned_free to deallocate.

pinned_free()

template<class T >

void GPUMemoryAllocation::pinned_free ( T *  ptr )

inlinenoexcept

Free memory allocated with pinned_alloc.

Parameters

[in]

ptr,pointer

to memory we want to release.