Crate j2k_transcode_cuda 

CUDA acceleration for coefficient-domain JPEG to HTJ2K transcode stages.

Mirrors j2k-transcode-metal: it implements DctToWaveletStageAccelerator for direct DCT-grid to one-level 5/3 and 9/7 wavelet projections (and the fused 9/7 HTJ2K code-block path), so JPEG can be transcoded to HTJ2K without an IDCT->pixels->DWT spatial round-trip. The CPU scalar code in j2k-transcode remains the oracle and fallback; this crate never reimplements it.

The actual GPU kernels live in j2k-cuda-runtime CUDA Oxide projects and are loaded through the runtime’s CUDA Driver API host layer. The GPU path is gated behind the cuda-runtime feature; without it this accelerator behaves like Metal’s non-macOS path (Explicit -> typed Err, Auto -> Ok(None) scalar fallback).

Structs

CudaDctToWaveletStageAccelerator

Optional CUDA accelerator for j2k-transcode transform stages.

CudaRuntimeFailure

Diagnostic retained when the CUDA runtime rejects a transcode operation.

Enums

CudaTranscodeError

Error returned by the CUDA transcode accelerator.

Constants

CUDA_UNAVAILABLE

Stable message returned when the CUDA runtime is unavailable (feature not compiled, no device, or the transcode kernels were not built).