Struct CudaDevice 

pub struct CudaDevice { /* private fields */ }

Re-export CUDA backend (requires cuda feature + NVIDIA CUDA Toolkit). A CUDA device handle. Contains the cudarc device and a cuBLAS handle for matrix multiplication. Clonable (uses Arc internally).

Implementations

impl CudaDevice

pub fn new(ordinal: usize) -> Result<CudaDevice, Error>

Create a new CUDA device for the given GPU ordinal (0, 1, …). Compiles all Shrew CUDA kernels on first creation.

pub fn device(&self) -> &Arc<CudaDevice>

Get the underlying cudarc device handle.

pub fn blas(&self) -> &CudaBlas

Get the cuBLAS handle.

pub fn pool(&self) -> &CudaMemPool

Get the memory pool.

pub fn empty_cache(&self)

Release all cached GPU memory back to the CUDA driver.

pub fn pool_stats(&self) -> PoolStats

Return pool statistics (cached bytes, hits, misses, etc.).

pub fn reclaim(&self, storage: CudaStorage)

Reclaim a CudaStorage buffer into the pool for future reuse.

pub fn pool_alloc_f32(&self, n: usize) -> Result<CudaSlice<f32>, Error>

Allocate n elements from the pool (content undefined).

pub fn pool_alloc_f64(&self, n: usize) -> Result<CudaSlice<f64>, Error>

pub fn pool_alloc_u16(&self, n: usize) -> Result<CudaSlice<u16>, Error>

pub fn pool_alloc_u8(&self, n: usize) -> Result<CudaSlice<u8>, Error>

pub fn pool_alloc_u32(&self, n: usize) -> Result<CudaSlice<u32>, Error>

pub fn pool_alloc_i64(&self, n: usize) -> Result<CudaSlice<i64>, Error>

pub fn pool_alloc_zeros_f32(&self, n: usize) -> Result<CudaSlice<f32>, Error>

Allocate n elements from the pool, zeroed.

pub fn pool_alloc_zeros_f64(&self, n: usize) -> Result<CudaSlice<f64>, Error>

pub fn pool_alloc_zeros_u16(&self, n: usize) -> Result<CudaSlice<u16>, Error>

pub fn pool_alloc_zeros_u8(&self, n: usize) -> Result<CudaSlice<u8>, Error>

pub fn pool_alloc_zeros_u32(&self, n: usize) -> Result<CudaSlice<u32>, Error>

pub fn pool_alloc_zeros_i64(&self, n: usize) -> Result<CudaSlice<i64>, Error>

Trait Implementations

impl BackendDevice for CudaDevice

fn name(&self) -> String

A human-readable name for this device (e.g., “cpu”, “cuda:0”).

impl Clone for CudaDevice

fn clone(&self) -> CudaDevice

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Debug for CudaDevice

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

Formats the value using the given formatter.

impl Send for CudaDevice

impl Sync for CudaDevice