• DocumentCode
    3703724
  • Title

    Parallelization of Kvazaar HEVC intra encoder for multi-core processors

  • Author

    Ari Koivula;Marko Viitanen;Jarno Vanne;Timo D. H?m?l?inen;Laurent Fasnacht

  • Author_Institution
    Department of Pervasive Computing, Tampere University of Technology, Tampere, Finland
  • fYear
    2015
  • Firstpage
    1
  • Lastpage
    6
  • Abstract
    This paper introduces key parallelization strategies of our Kvazaar HEVC intra encoder for multicore processors. The schemes implemented in Kvazaar are 1) tiles; 2) Wavefront Parallel Processing (WPP); and 3) picture-level parallel processing. Kvazaar is the only practical open-source HEVC encoder that supports all these schemes. In addition, its rate-distortion-complexity characteristics are superior to other public implementations in all-intra (AI) coding. Our experiments with high-quality encoder presets show that a C implementation of Kvazaar is 19% faster than the corresponding implementation of x265 for the same coding efficiency with 8 threads and 38% faster with 16 threads. With the high-speed presets, Kvazaar improves coding efficiency by 4.5% while being twice as fast as x265. The high-speed preset of Kvazaar obtains almost the same coding efficiency as the high-quality preset of f265 while being 24 times faster when 16 threads are used.
  • Keywords
    "Encoding","Parallel processing","Instruction sets","Open source software","Transforms","Video coding","Artificial intelligence"
  • Publisher
    ieee
  • Conference_Titel
    Signal Processing Systems (SiPS), 2015 IEEE Workshop on
  • Type

    conf

  • DOI
    10.1109/SiPS.2015.7345015
  • Filename
    7345015