Struct CudaBuffer 

pub struct CudaBuffer<T> { /* private fields */ }

Owned GPU memory buffer holding len elements of type T.

When pool_fn is Some, dropping returns the inner CudaSlice to the global pool (crate::pool) instead of freeing GPU memory.

alloc_len is the rounded allocation size used as the pool key. len is the logical element count visible to callers.

Implementations

impl<T> CudaBuffer<T>

pub fn len(&self) -> usize

Number of logical elements in this buffer.

pub fn alloc_len(&self) -> usize

Rounded allocation length used as the pool key.

For pooled buffers, this is round_len(len). For non-pooled buffers, this equals len. Stats (hits, misses, returns) use len consistently within the allocator for user-facing reporting; alloc_len is an internal detail for pool key stability.

pub fn is_empty(&self) -> bool

Whether the buffer is empty.

pub fn device_ordinal(&self) -> usize

The ordinal of the device that owns this memory.

pub fn inner(&self) -> &CudaSlice<T>

Borrow the underlying CudaSlice for use with cudarc APIs.

pub fn inner_mut(&mut self) -> &mut CudaSlice<T>

Mutably borrow the underlying CudaSlice.

Trait Implementations

impl<T> Debug for CudaBuffer<T>

Available on crate feature cuda only.

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

Formats the value using the given formatter.

impl<T> Drop for CudaBuffer<T>

Available on crate feature cuda only.

fn drop(&mut self)

Executes the destructor for this type.