Struct GraphBuffer 

pub struct GraphBuffer<T: DeviceRepr> { /* private fields */ }

Device memory whose allocation is retained by CUDA graph objects.

GraphBuffer values are created through Graph::create_buffer, Graph::zeroes_buffer, or Graph::buffer_from_slice. Graph and executable-graph APIs that accept GraphBuffer retain the underlying allocation so graph replay cannot outlive the device pointers baked into CUDA graph nodes.

Implementations

impl<T> GraphBuffer<T>

where T: DeviceRepr + Send + Sync,

pub fn len(&self) -> usize

pub fn is_empty(&self) -> bool

pub fn byte_len(&self) -> usize

pub fn context(&self) -> Option<&Context>

pub fn as_ptr(&self) -> *const T

pub fn as_mut_ptr(&mut self) -> *mut T

pub fn copy_from_host(&mut self, host_slice: &[T]) -> Result<()>

Copies a host slice into this graph-retained device buffer.

This updates the stable allocation used by graph-buffer node APIs. The caller is still responsible for ordering this copy against graph launches that read or write the same allocation.

Errors

Returns an error if host_slice does not have the same length as this buffer or if CUDA rejects the copy.

pub fn copy_to_host(&self, host_slice: &mut [T]) -> Result<()>

Copies this graph-retained device buffer into a host slice.

Errors

Returns an error if host_slice does not have the same length as this buffer or if CUDA rejects the copy.

pub fn copy_from_buffer(&mut self, src: &Self) -> Result<()>

Copies another graph-retained buffer into this buffer.

Errors

Returns an error if the buffers have different lengths or if CUDA rejects the copy.

pub fn copy_to_host_vec(&self) -> Result<Vec<T>>

Trait Implementations

impl<T: Debug + DeviceRepr> Debug for GraphBuffer<T>

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.