v0.2.0-beta — GPU decoding
Pre-release
Pre-release
GPU decoding pipelines plus parallel-throughput fixes.
pipeline="cuda"(NVIDIA, Linux/Windows) andpipeline="metal"(Apple GPU, macOS) decode on the GPU via the BRAW SDK- Throughput fixes (fmParallel + deferred plane copy): ~2.3× (Metal) / ~1.6× (CUDA) vs CPU at 4.6K — see
doc/gpu-benchmark.md - Removed the no-op
alphaparameter (BRAW has no alpha channel)