Release of rav1e 0.5, AV1 encoder

rav1e 0.5.0, the encoder for the AV1 video encoding format, has been released. The product is developed by the Mozilla and Xiph communities and differs from the reference implementation of libaom written in C / C ++ languages ​​by increasing the speed of coding and increased attention to security (compression efficiency still lags behind). The product is written in the Rust programming language with assembler optimizations (72.2% - assembler, 27.5% - Rust), the code is distributed under the BSD license. Ready builds are prepared for Windows and macOS (builds for Linux are temporarily omitted due to problems with the continuous integration system).

rav1e supports all major AV1 features, including support for intra- and inter-frames (intra- and inter-frames), 64x64 superblocks, 4:2:0, 4:2:2, and 4:4:4 chroma subsampling , 8-, 10- and 12-bit color depth encoding, RDO (Rate-distortion optimization) distortion optimization, various modes of inter-frame change prediction and transformation detection, bitrate control and scene truncation detection.

The AV1 format is noticeably ahead of H.264 and VP9 in terms of compression capabilities, but due to the complexity of the algorithms that implement them, it requires significantly more time for encoding (in terms of encoding speed, libaom lags hundreds of times behind libvpx-vp9, and thousands of times behind x264). The rav1e encoder provides 11 levels of performance, the highest of which allows you to achieve speeds close to real-time encoding. The encoder is available both as a command line utility and as a library.

The new version contains the following changes:

  • Significant acceleration of the codec;
    Release of rav1e 0.5, AV1 encoder
  • Fixing a bug that caused the encoder to crash at certain video sizes;
  • Use of AVX2 instructions for a significant (up to 13 times) acceleration of Wiener estimation for 16 bits per channel. Similarly, the use of SIMD instructions has been added, which made it possible to speed up calculations up to 7 times in similar conditions;
  • Tons of minor fixes and optimizations for x86, arm32 and arm64 platforms.

Source: opennet.ru

Add a comment