Expand description
CUDA reconstruction and resource-lifecycle boundary for JPEG XR.
Parsing and entropy decoding stay in jxr-native; this crate consumes only
the validated macroblock-major reconstruction contract from jxr-core.
Structs§
- Cuda
Batch Destination - Exclusively writable homogeneous batch in one caller-owned CUDA allocation.
- Cuda
Batch Destination Completion - Completion metadata and retained storage for a caller-owned dense batch.
- Cuda
Batch Destination Submission - Pending dense decode retaining every resource and exclusive destination.
- Cuda
Batch Submission - Pending ordered collection of CUDA reconstruction submissions.
- Cuda
Buffer Pool Diagnostics - Exact-size CUDA scratch reuse counters.
- Cuda
Decode Plan - Validated CUDA reconstruction plan retaining CPU-produced coefficients.
- Cuda
Decoder Session - Reusable CUDA runtime session with cached kernels, uploads, streams, and scratch.
- Cuda
Destination - Exclusively writable JPEG XR surface in a caller-owned CUDA allocation.
- Cuda
Destination Completion - Completion metadata for a caller-owned CUDA destination.
- Cuda
Destination Submission - Pending decode retaining exclusive destination access through completion.
- Cuda
Kernel Manifest - Static identity for one CUDA phase implementation.
- Cuda
Resident Batch - Completed homogeneous batch in one CUDA allocation.
- Cuda
Resident Batch Submission - Pending homogeneous reconstruction into one internally owned allocation.
- Cuda
Submission - Pending CUDA reconstruction retaining all device resources through completion.
- Cuda
Upload Cache Diagnostics - Immutable coefficient upload-cache counters.
- Dense
Cuda Batch Layout - One tightly packed image repeated contiguously in a CUDA allocation.
- Resident
Cuda Image - Immutable completed CUDA-resident JPEG XR output.
Enums§
- Cuda
Error - Failure at the JPEG XR CUDA backend boundary.
- Cuda
Route Decision - Concrete route selected by the CUDA planner.
- Kernel
Stage - Ordered JPEG XR CUDA reconstruction phase.
Constants§
- CUDA_
AUTO_ THRESHOLD - Provisional automatic-routing threshold in reconstructed coefficients.
- RECONSTRUCTION_
KERNELS - Ordered complete reconstruction manifest.
Functions§
- plan_
cuda_ route - Resolve a CUDA route without probing or mutating runtime state.