Struct CudaDecoderSession 

pub struct CudaDecoderSession { /* private fields */ }

Reusable CUDA runtime session with cached kernels, uploads, streams, and scratch.

Implementations

impl CudaDecoderSession

pub fn is_available() -> bool

Probe for both dynamically loaded CUDA driver and NVRTC libraries.

This performs no CUDA initialization and is safe on CPU-only hosts.

pub fn system_default() -> Result<Self, CudaError>

Create a reusable session on device ordinal zero.

pub fn new(device_ordinal: usize) -> Result<Self, CudaError>

Create a reusable session on a caller-selected CUDA device ordinal.

pub const fn is_usable(&self) -> bool

Whether this initialized session can accept reconstruction work.

pub fn device_ordinal(&self) -> usize

Device ordinal retained by this session.

pub fn stream_count(&self) -> usize

Number of reusable nonblocking streams available to batch work.

pub fn buffer_pool_diagnostics( &self, ) -> Result<CudaBufferPoolDiagnostics, CudaError>

Return retained exact-size scratch allocation counters.

pub fn upload_cache_diagnostics( &self, ) -> Result<CudaUploadCacheDiagnostics, CudaError>

Return immutable coefficient upload-cache counters.

pub fn allocate_destination( &self, layout: SurfaceLayout, ) -> Result<CudaDestination, CudaError>

Allocate an exact device output owned by the caller.

pub fn allocate_batch_destination( &self, layout: DenseCudaBatchLayout, ) -> Result<CudaBatchDestination, CudaError>

Allocate a dense device batch output owned by the caller.

pub fn submit(&self, plan: &CudaDecodePlan) -> Result<CudaSubmission, CudaError>

Submit one validated plan without waiting for completion.

pub fn submit_batch( &self, plans: &[CudaDecodePlan], ) -> Result<CudaBatchSubmission, CudaError>

Submit a bounded batch across the session’s reusable streams.

pub fn submit_dense_batch( &self, plans: &[CudaDecodePlan], ) -> Result<CudaResidentBatchSubmission, CudaError>

Submit a homogeneous batch into one internally owned device allocation.

pub fn submit_into( &self, plan: &CudaDecodePlan, destination: CudaDestination, ) -> Result<CudaDestinationSubmission, CudaError>

Submit into an exclusively retained caller-owned device allocation.

pub fn submit_batch_into( &self, plans: &[CudaDecodePlan], destination: CudaBatchDestination, ) -> Result<CudaBatchDestinationSubmission, CudaError>

Submit a homogeneous batch into caller-owned dense device storage.

pub fn decode_to_host( &self, plan: &CudaDecodePlan, ) -> Result<DecodedImage, CudaError>

Decode one plan and copy its exact native bytes to typed host storage.

pub fn decode_batch_to_host( &self, plans: &[CudaDecodePlan], ) -> Result<Vec<DecodedImage>, CudaError>

Decode a batch to host storage while preserving caller order.

pub fn readback(&self, image: &ResidentCudaImage) -> Result<Vec<u8>, CudaError>

Copy one completed resident image to host bytes.

pub fn readback_batch_image( &self, batch: &CudaResidentBatch, image: usize, ) -> Result<Vec<u8>, CudaError>

Copy one image from a completed dense batch to host bytes.

Trait Implementations

impl Clone for CudaDecoderSession

fn clone(&self) -> Self

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for CudaDecoderSession

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

Formats the value using the given formatter.