Changes
- Speed levels had been retuned
- RDO-Lookahead is now speed-level dependent by default
--help
now shows colors- Unify spatial and temporal adaptive quantization
- The encoding time is reduced by 52.61% at speed level 0.
- General 3~6% improvement in BD-Rate
- The encoders would use the global threadpool unless configured otherwise
- Allows to use rayon if wasm is built with atomics support
- Scene-change logic improvement, now it faster and better.
- Overall 1~2% faster
- Around 1% improvement in BD-Rate
- Import ASM from dav1d 0.8.1, 0.8.2, 0.9.0, 0.9.1 for AVX2, SSE2, SSE3, NEON.
- More than 50% improved encoding times for 10bit files.
Fixes
- Initialize residual when less than the transform width is visible.
- This fixes the squared artifacts and incorrectly encoded edge of a frame, this was particularly evident for still images.
- Correctly align the memory when using AVX512.
New features
- by-gop parallel encoding is available as part of the
channel-api
no-scene-detection
encoding option- Make sure the dequantize tests do not use random values outside the possible range.
- v_frame provides now a simple downscaler.