Struct CudaBatchDecoder 

pub struct CudaBatchDecoder { /* private fields */ }

Persistent CUDA batch decoder that reuses one CudaSession.

Implementations

impl CudaBatchDecoder

pub fn new() -> Self

Create a decoder with a lazily initialized CUDA session and strict shared batch options.

pub fn with_options(options: BatchDecodeOptions) -> Self

Create a decoder with explicit shared preparation options.

pub fn with_session(session: CudaSession) -> Self

Create a decoder around an existing CUDA session.

pub const fn with_session_and_options( session: CudaSession, options: BatchDecodeOptions, ) -> Self

Create a decoder around an existing session and preparation policy.

pub const fn options(&self) -> BatchDecodeOptions

Shared preparation options used by Self::prepare and Self::decode_batch.

pub const fn session(&self) -> &CudaSession

Borrow the persistent session for diagnostics.

pub fn decode_pool_diagnostics( &self, ) -> Result<CudaDecodePoolDiagnostics, Error>

Snapshot the persistent session’s two private decode buffer pools.

pub fn diagnostics(&self) -> Result<CudaSessionDiagnostics, Error>

Snapshot CUDA transfer/event counters and retained decode-pool memory.

pub fn session_mut(&mut self) -> &mut CudaSession

Mutably borrow the persistent session for advanced configuration.

pub fn prepare( &self, inputs: Vec<EncodedImage>, ) -> Result<PreparedBatch, BatchInfrastructureError>

Inspect and group owned inputs without copying their compressed bytes.

pub fn prepare_prepared_images( &self, images: Vec<PreparedImage>, ) -> Result<PreparedBatch, BatchInfrastructureError>

Regroup caller-supplied prepared images without reparsing codestream bytes.

Returned source indices are positions in images; each image retains its original PreparedImage::source_index for provenance.

pub fn decode_batch( &mut self, inputs: Vec<EncodedImage>, ) -> Result<CudaBatchDecodeResult, CudaBatchError>

Prepare and strictly decode one owned batch to CUDA-resident groups.

pub fn decode_prepared_images( &mut self, images: Vec<PreparedImage>, ) -> Result<CudaBatchDecodeResult, CudaBatchError>

Regroup and decode prepared images without reparsing their encoded bytes.

pub fn decode_prepared( &mut self, prepared: &PreparedBatch, ) -> Result<CudaBatchDecodeResult, CudaBatchError>

Strictly decode a reusable shared codec batch to CUDA-resident groups.

Explicit CUDA execution never falls back to CPU-decoded pixels. A CUDA execution error discards the entire affected dense group.

pub unsafe fn decode_batch_into( &mut self, group: &PreparedBatchGroup, destination: &mut CudaExternalDeviceBufferViewMut<'_>, ) -> Result<CudaExternalBatchGroup, CudaBatchError>

Decode one prepared exact-native Gray/RGB/RGBA group directly into a validated caller-owned CUDA allocation range.

The destination must belong to this decoder’s CUDA context and cover the tightly concatenated group output. Decoded pixels are never staged through host memory or copied through an intermediate device output.

Safety

The destination allocation must remain live until this method returns success. If CUDA completion cannot be proven and an error is returned, the caller must quarantine the allocation rather than free or reuse it.

pub unsafe fn submit_batch_into( &mut self, group: &PreparedBatchGroup, destination: &mut CudaExternalDeviceBufferViewMut<'_>, ) -> Result<SubmittedCudaExternalBatch, CudaBatchError>

Submit one prepared exact-native Gray/RGB/RGBA group into CUDA storage without a host completion wait.

Callers integrating another CUDA runtime must use j2k_cuda_runtime::CudaContext::with_primary_stream_ordering around this submission, then retain the returned value alongside the tensor so codec-internal resources outlive the ordered final store.

Safety

The destination allocation must remain live and may only be consumed on a CUDA stream ordered after codec completion until this value is waited or dropped. No unordered host or device access may overlap the decode. Stream completion alone does not validate entropy status: callers must not expose the destination as decoded pixels until SubmittedCudaExternalBatch::wait succeeds. If CUDA completion cannot be proven, the external allocation must be quarantined rather than freed or reused.

impl CudaBatchDecoder

pub fn submit_batch( &mut self, inputs: Vec<EncodedImage>, ) -> Result<SubmittedCudaResidentBatch, CudaBatchError>

Prepare and asynchronously submit a codec-owned CUDA resident batch.

pub fn submit_prepared( &mut self, prepared: &PreparedBatch, ) -> Result<SubmittedCudaResidentBatch, CudaBatchError>

Asynchronously submit reusable prepared inputs to codec-owned CUDA output.

Recoverable execution failures remain group-local and do not suppress later groups. No decoded host staging or final device copy is used.

Trait Implementations

impl BatchDecoder for CudaBatchDecoder

type Output = CudaBatchDecodeResult

Backend-specific successful output.

type Error = CudaBatchError

Backend-specific infrastructure or execution error.

fn decode_prepared( &mut self, prepared: &PreparedBatch, ) -> Result<Self::Output, Self::Error>

Decode a reusable prepared batch without consuming its encoded inputs or plans.

fn options(&self) -> BatchDecodeOptions

Preparation and output policy retained by this persistent session.

fn prepare_batch( &self, inputs: Vec<EncodedImage>, ) -> Result<PreparedBatch, Self::Error>

Inspect and group owned inputs without consuming their encoded byte owners.

fn prepare_prepared_images( &self, images: Vec<PreparedImage>, ) -> Result<PreparedBatch, Self::Error>

Regroup caller-supplied prepared images without reparsing their encoded bytes.

fn decode_batch( &mut self, inputs: Vec<EncodedImage>, ) -> Result<Self::Output, Self::Error>

Prepare and decode one owned batch through the common session boundary.

fn decode_prepared_images( &mut self, images: Vec<PreparedImage>, ) -> Result<Self::Output, Self::Error>

Regroup and decode caller-supplied prepared images without reparsing them.

impl Clone for CudaBatchDecoder

fn clone(&self) -> CudaBatchDecoder

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for CudaBatchDecoder

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

Formats the value using the given formatter.

impl Default for CudaBatchDecoder

fn default() -> CudaBatchDecoder

Returns the “default value” for a type.