Issue metadata
Sign in to add a comment
|
1.2%-50% regression in webrtc_perf_tests at 13142:13143 |
||||||||||||||||||||||
Issue descriptionSee graphs below.
,
Jun 16 2016
Hi Alex, your CL (https://codereview.webrtc.org/2058753003) seems to have a generally positive effect on many perf stats related to encode time and cpu usage on Android. One thing that looks a bit worrying is that dropped_frames/screenshare_slides deteriorated by 50% on Nexus 9. And also many of the PSNR/SSIM related perf stats on various android devices deteriorated by 1-5%. Could you have a look at this, please?
,
Jun 24 2016
Adding Alex, see comment #2
,
Jul 1 2016
Alex, ping on this.
,
Jul 1 2016
Adjusting number of threads may reduce PSNR a little and this is expected, but should not have impact on the amount of dropped frames. Please assign to marpan@or jackychen@ for triage. Also it is possible to disable this optimization for Chromium builds if there is no easy fix, but keep it for standalone WebRTC Android builds.
,
Aug 15 2016
marpan@: Could you have a look at this, please?
,
Aug 17 2016
For screenshare in vp8, we only drop frames if we detect the current frame will be too big (i.e., on slide change) from a partial encoding (in which case we drop it and encode next frame at higher QP). Since number of threads does change the encoding a little, it may have changed the detection somewhat and dropped more. We'll need to verify by running the screenshare on nexus 9 device.
,
Oct 6 2016
,
Mar 31 2017
Is this still a valid issue?
,
Mar 31 2017
Re-open if this is still an issue. |
|||||||||||||||||||||||
►
Sign in to add a comment |
|||||||||||||||||||||||
Comment 1 by ivoc@chromium.org
, Jun 16 2016