Struct SparseHandle 

pub struct SparseHandle { /* private fields */ }

Central handle for sparse matrix operations.

Every sparse routine requires a SparseHandle. The handle binds operations to a specific CUDA context and stream, maintains a BLAS sub-handle for dense operations (e.g. SpMM output accumulation), caches compiled PTX kernels, and stores the device’s SM version for kernel dispatch.

Thread safety

SparseHandle is Send but not Sync. Each thread should create its own handle (possibly sharing the same Arc<Context>).

Implementations

impl SparseHandle

pub fn new(ctx: &Arc<Context>) -> SparseResult<Self>

Creates a new sparse handle with a freshly-allocated default stream.

The device’s compute capability is queried once and cached.

Errors

Returns SparseError::Cuda if stream creation or device query fails.

pub fn with_stream(ctx: &Arc<Context>, stream: Stream) -> SparseResult<Self>

Creates a new sparse handle bound to an existing stream.

Errors

Same as new except stream creation cannot fail.

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

Returns a reference to the CUDA context.

pub fn stream(&self) -> &Stream

Returns a reference to the stream used for kernel launches.

pub fn blas_handle(&self) -> &BlasHandle

Returns a reference to the internal BLAS handle.

pub fn blas_handle_mut(&mut self) -> &mut BlasHandle

Returns a mutable reference to the internal BLAS handle.

pub fn sm_version(&self) -> SmVersion

Returns the SM version of the bound device.

pub fn ptx_cache(&self) -> &PtxCache

Returns a reference to the PTX cache.

pub fn set_stream(&mut self, stream: Stream)

Replaces the stream used for subsequent sparse operations.