#include <buffer.h>

Collaboration diagram for cuda_buffer:

Public Attributes
void *	host

void *	dev

size_t	size

int	symm

int	dev_only

int	registered

struct cuda_buffer *	next

Detailed Description

Device buffer: a lazily grown pair of a pinned host buffer and a device buffer, owned by the device layer. A buffer is linked into a registry on its first reservation, and cuda_buffer_free_all() (called from cuda_finalize) releases all registered buffers and resets them to an empty state, from which they grow again on demand after a subsequent device init

Definition at line 51 of file buffer.h.

Member Data Documentation

◆ dev

void* cuda_buffer::dev

Device buffer

Definition at line 53 of file buffer.h.

◆ dev_only

int cuda_buffer::dev_only

Device buffer only, no pinned host buffer

Definition at line 56 of file buffer.h.

◆ host

void* cuda_buffer::host

Pinned host buffer (NULL if device-only)

Definition at line 52 of file buffer.h.

◆ next

struct cuda_buffer* cuda_buffer::next

Registry link

Definition at line 58 of file buffer.h.

◆ registered

int cuda_buffer::registered

Buffer linked into the registry

Definition at line 57 of file buffer.h.

◆ size

size_t cuda_buffer::size

Capacity in bytes

Definition at line 54 of file buffer.h.

◆ symm

int cuda_buffer::symm

Device buffer on NVSHMEM symmetric heap

Definition at line 55 of file buffer.h.

The documentation for this struct was generated from the following file:

/home/runner/work/neko/neko/src/device/cuda/buffer.h

Public Attributes