[Libav-user] libswscale speed regression

[Massimo Battistel]

Hi all,
I've recently upgraded an application based on libav*/libsw* to lastest
zeranoe builds.
Upgrade was fine, but I've experienced serious speed regressions at runtime
(I mean 2x of cpu usage).

Original libs version (2015/02/03 build):

  libavutil      54. 18.100 / 54. 18.100
  libavcodec     56. 21.102 / 56. 21.102
  libavformat    56. 19.100 / 56. 19.100
  libavdevice    56.  4.100 / 56.  4.100
  libavfilter     5.  9.103 /  5.  9.103
  libswscale      3.  1.101 /  3.  1.101
  libswresample   1.  1.100 /  1.  1.100
  libpostproc    53.  3.100 / 53.  3.100

Current libs version (2016/06/07 build):

  libavutil      55. 24.100 / 55. 24.100
  libavcodec     57. 45.100 / 57. 45.100
  libavformat    57. 37.101 / 57. 37.101
  libavdevice    57.  0.101 / 57.  0.101
  libavfilter     6. 46.101 /  6. 46.101
  libswscale      4.  1.100 /  4.  1.100
  libswresample   2.  0.101 /  2.  0.101
  libpostproc    54.  0.100 / 54.  0.100

>From what I can see, speed regression is in libswscale only.
Tipical libswscale usage I do is:

   1. Scaling to various sizes with SWS_BICUBIC enabled;
   2. Converting colorspaces from yuv420p to yuv422 and viceversa.

I've seen there has been a huge refactor in libswscale recently.

   1. Is the speed regression a known issue?
   2. Are there any workarounds available?
   3. Can I expect some optimizations in the near future?


Thanks,
MB
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://ffmpeg.org/pipermail/libav-user/attachments/20160622/b19fe0b2/attachment.html>