<div dir="ltr"><div>Hi all,</div><div>I've recently upgraded an application based on libav*/libsw* to lastest zeranoe builds.</div><div>Upgrade was fine, but I've experienced serious speed regressions at runtime (I mean 2x of cpu usage).</div><div><br></div><div>Original libs version (2015/02/03 build):</div><div><br></div><div><div> libavutil 54. 18.100 / 54. 18.100</div><div> libavcodec 56. 21.102 / 56. 21.102</div><div> libavformat 56. 19.100 / 56. 19.100</div><div> libavdevice 56. 4.100 / 56. 4.100</div><div> libavfilter 5. 9.103 / 5. 9.103</div><div> libswscale 3. 1.101 / 3. 1.101</div><div> libswresample 1. 1.100 / 1. 1.100</div><div> libpostproc 53. 3.100 / 53. 3.100</div></div><div><br></div><div>Current libs version (2016/06/07 build):</div><div><br></div><div><div> libavutil 55. 24.100 / 55. 24.100</div><div> libavcodec 57. 45.100 / 57. 45.100</div><div> libavformat 57. 37.101 / 57. 37.101</div><div> libavdevice 57. 0.101 / 57. 0.101</div><div> libavfilter 6. 46.101 / 6. 46.101</div><div> libswscale 4. 1.100 / 4. 1.100</div><div> libswresample 2. 0.101 / 2. 0.101</div><div> libpostproc 54. 0.100 / 54. 0.100</div></div><div><br></div><div>From what I can see, speed regression is in libswscale only.</div><div>Tipical libswscale usage I do is:</div><div><ol><li>Scaling to various sizes with SWS_BICUBIC enabled;<br></li><li>Converting colorspaces from yuv420p to yuv422 and viceversa.</li></ol><div>I've seen there has been a huge refactor in libswscale recently.</div></div><div><ol><li>Is the speed regression a known issue?<br></li><li>Are there any workarounds available?</li><li>Can I expect some optimizations in the near future?</li></ol><div><br></div></div>Thanks,
<div>MB</div><div><br></div></div>