Struct CudaEncodeStageAccelerator 

pub struct CudaEncodeStageAccelerator { /* private fields */ }

CUDA implementation of selected JPEG 2000 encode stages.

Implementations

impl CudaEncodeStageAccelerator

pub fn with_profile_collection(collect_profile: bool) -> Self

Create an encode-stage accelerator with optional CUDA stage timing collection.

pub fn for_auto_host_output() -> Self

Create the measured Auto route for host-output HTJ2K encode.

CUDA keeps the DWT and HT code-block stages, while forward RCT and Tier-2 packetization stay on the CPU for the current host-pixel path.

pub fn prefer_cpu_forward_rct(self, prefer_cpu_forward_rct: bool) -> Self

Prefer scalar CPU forward RCT while keeping later CUDA stages enabled.

pub fn prefer_cpu_packetization(self, prefer_cpu_packetization: bool) -> Self

Prefer scalar CPU Tier-2 packetization while keeping CUDA Tier-1/HT block coding enabled.

This is useful for batches of many small tiles where launching a CUDA packetization kernel and copying several tiny descriptor buffers per tile costs more than forming the packet body on the host.

pub fn prefer_cpu_ht_subband(self, prefer_cpu_ht_subband: bool) -> Self

Prefer host sub-band quantization while keeping batched CUDA HT code-block encode enabled.

This avoids launching one CUDA quantize/subband path for every prepared subband in multi-resolution precomputed transcode outputs, where the many tiny launches cost more than CPU quantization.

pub fn prefer_cpu_quantize_subband( self, prefer_cpu_quantize_subband: bool, ) -> Self

Prefer host sub-band quantization while keeping CUDA HT code-block encode enabled.

Multi-resolution transcode workloads can contain thousands of small subbands; for those, CPU quantization plus one batched HT code-block encode per tile is currently faster than launching CUDA quantization for every subband.

pub const fn collected_stage_timings(&self) -> CudaEncodeStageTimings

Return cumulative CUDA encode stage timings collected by this accelerator.

pub fn reset_collected_stage_timings(&mut self)

Clear cumulative CUDA encode stage timings without changing dispatch counters.

pub fn deinterleave_attempts(&self) -> usize

Number of deinterleave attempts observed.

pub fn forward_rct_attempts(&self) -> usize

Number of forward RCT attempts observed.

pub fn forward_ict_attempts(&self) -> usize

Number of forward ICT attempts observed.

pub fn forward_dwt53_attempts(&self) -> usize

Number of forward 5/3 DWT attempts observed.

pub fn forward_dwt97_attempts(&self) -> usize

Number of forward 9/7 DWT attempts observed.

pub fn quantize_subband_attempts(&self) -> usize

Number of sub-band quantization attempts observed.

pub fn tier1_code_block_attempts(&self) -> usize

Number of classic Tier-1 code-block attempts observed.

pub fn ht_code_block_attempts(&self) -> usize

Number of HT code-block attempts observed.

pub fn packetization_attempts(&self) -> usize

Number of packetization attempts observed.

pub fn deinterleave_dispatches(&self) -> usize

Number of deinterleave CUDA dispatches.

pub fn forward_rct_dispatches(&self) -> usize

Number of forward RCT CUDA dispatches.

pub fn forward_ict_dispatches(&self) -> usize

Number of forward ICT CUDA dispatches.

pub fn forward_dwt53_dispatches(&self) -> usize

Number of forward 5/3 DWT CUDA dispatches.

pub fn forward_dwt97_dispatches(&self) -> usize

Number of forward 9/7 DWT CUDA dispatches.

pub fn quantize_subband_dispatches(&self) -> usize

Number of sub-band quantization CUDA dispatches.

pub fn tier1_code_block_dispatches(&self) -> usize

Number of classic Tier-1 CUDA dispatches.

pub fn ht_code_block_dispatches(&self) -> usize

Number of HT code-block CUDA dispatches.

pub fn packetization_dispatches(&self) -> usize

Number of packetization CUDA dispatches.

Trait Implementations

impl Clone for CudaEncodeStageAccelerator

fn clone(&self) -> CudaEncodeStageAccelerator

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for CudaEncodeStageAccelerator

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

Formats the value using the given formatter.

impl Default for CudaEncodeStageAccelerator

fn default() -> CudaEncodeStageAccelerator

Returns the “default value” for a type.

impl J2kEncodeStageAccelerator for CudaEncodeStageAccelerator

fn dispatch_report(&self) -> J2kEncodeDispatchReport

Report cumulative backend dispatches completed by this accelerator.

fn encode_deinterleave( &mut self, job: J2kDeinterleaveToF32Job<'_>, ) -> Result<Option<Vec<Vec<f32>>>, &'static str>

Optionally deinterleave interleaved pixel bytes into f32 component planes.

fn encode_forward_rct( &mut self, job: J2kForwardRctJob<'_>, ) -> Result<bool, &'static str>

Optionally apply forward RCT in place.

fn encode_forward_ict( &mut self, job: J2kForwardIctJob<'_>, ) -> Result<bool, &'static str>

Optionally apply forward ICT in place.

fn encode_forward_dwt53( &mut self, job: J2kForwardDwt53Job<'_>, ) -> Result<Option<J2kForwardDwt53Output>, &'static str>

Optionally run a forward reversible 5/3 DWT.

fn encode_forward_dwt97( &mut self, job: J2kForwardDwt97Job<'_>, ) -> Result<Option<J2kForwardDwt97Output>, &'static str>

Optionally run a forward irreversible 9/7 DWT.

fn encode_quantize_subband( &mut self, job: J2kQuantizeSubbandJob<'_>, ) -> Result<Option<Vec<i32>>, &'static str>

Optionally quantize one sub-band.

fn encode_tier1_code_block( &mut self, _job: J2kTier1CodeBlockEncodeJob<'_>, ) -> Result<Option<EncodedJ2kCodeBlock>, &'static str>

Optionally encode one classic Tier-1 code-block.

fn encode_ht_code_block( &mut self, job: J2kHtCodeBlockEncodeJob<'_>, ) -> Result<Option<EncodedHtJ2kCodeBlock>, &'static str>

Optionally encode one HTJ2K code-block.

fn encode_ht_code_blocks( &mut self, jobs: &[J2kHtCodeBlockEncodeJob<'_>], ) -> Result<Option<Vec<EncodedHtJ2kCodeBlock>>, &'static str>

Optionally encode multiple HTJ2K code-blocks in one backend dispatch.

fn encode_htj2k_tile( &mut self, job: J2kHtj2kTileEncodeJob<'_>, ) -> Result<Option<Vec<u8>>, &'static str>

Optionally encode the complete HTJ2K tile packet body.

fn encode_ht_subband( &mut self, job: J2kHtSubbandEncodeJob<'_>, ) -> Result<Option<Vec<EncodedHtJ2kCodeBlock>>, &'static str>

Optionally quantize and encode one HTJ2K cleanup-only sub-band.

fn encode_packetization( &mut self, job: J2kPacketizationEncodeJob<'_>, ) -> Result<Option<Vec<u8>>, &'static str>

Optionally packetize prepared packet contributions.

fn encode_tier1_code_blocks( &mut self, _jobs: &[J2kTier1CodeBlockEncodeJob<'_>], ) -> Result<Option<Vec<EncodedJ2kCodeBlock>>, &'static str>

Optionally encode multiple classic Tier-1 code-blocks in one backend dispatch.

fn prefer_parallel_cpu_code_block_fallback(&self) -> bool

Return whether native CPU code-block fallback should use internal rayon parallelism.

fn prefer_parallel_cpu_tile_encode(&self) -> bool

Return whether whole-tile CPU-only batch encode may be parallelized by callers.