New issue
Advanced search Search tips
Note: Color blocks (like █ or █) mean that a user may not be available. Tooltip shows the reason.

Issue 811572 link

Starred by 1 user
Status: Verified
Owner:
hiroh@chromium.org
Closed: Feb 2018
Cc:
dcasta...@chromium.org
posciak@chromium.org
█ wuchengli@chromium.org
Components:
EstimatedDays: ----
NextAction: ----
OS: ----
Pri: 1
Type: Bug



Sign in to add a comment

Software decode regression on nyan_big

Project Member Reported by hiroh@chromium.org, Feb 13 2018 
Many frames are dropped in software decoding on nyan_big, rather than 65 branch. For instance, the dropped frame percent for h264 1080p video increases from 0% to 30%.
The regression happens at some point on 66 branch. Looks like the regression is unrelated to video compression format.
https://chromeperf.appspot.com/report?sid=3f069954ad35bfe54614062e085202c85df673915fd28247f62d9b47ab56034d

It's worth noting this regression doesn't affect on nyan_blaze.
https://chromeperf.appspot.com/report?sid=53d236780e2a8c14eb90fba717cb341cb840f864324fb04a1c12b46ff94a72f4

Comment 1 by posciak@chromium.org, Feb 13 2018

Cc: dcasta...@chromium.org
Labels: videoshortlist

Comment 2 by hiroh@chromium.org, Feb 13 2018

Description: Show this description

Comment 3 by hiroh@chromium.org, Feb 13 2018 

I couldn't reproduce this regression on my local nyan_big PreMP device. The percentage of dropped frames is 0%.
On the one hand, I could reproduce with lab device, chromeos4-row5-rack10-host13.cros, which is MP.

The regression happens between 10359.0.0 and 10388.0.0.
Since there is no nyan_big build image in the range, we didn't bisect any more.
The regression is probably due to Chrome change.
Chrome version is 66.0.3329.0 in 10359.0.0, and 66.0.3343.0 in 10388.0.0.
The regression happens with Chrome 66.0.3343.0 on ChromeOS 10359.0.0.

Comment 4 by hiroh@chromium.org, Feb 16 2018 

I'm bisecting.
66.0.3340.0 is bad and 66.0.3339.0 is good.

Comment 5 by hiroh@chromium.org, Feb 16 2018 

I located https://chromium-review.googlesource.com/c/chromium/src/+/898542 was a bad CL.
I realized that the CL was discussed in crbug.com/809670.
This CL may affect software decoding?
According to crbug.com/809670, the bug was fixed on Chrome ToT.
I should test with Chrome ToT on latest canary ChromeOS image.

Comment 6 by hiroh@chromium.org, Feb 16 2018 

This regression doesn't happen with Chrome ToT on ChromeOS 66.10408.0.0.
Let us wait for Chrome uprev and see the test results after that.

Comment 7 by hiroh@chromium.org, Feb 22 2018

Status: Verified (was: Started)
The software decoding regression was resolved in the last few builds. It is thanks to Chrome uprev.
https://chromeperf.appspot.com/report?sid=3f069954ad35bfe54614062e085202c85df673915fd28247f62d9b47ab56034d
►

Sign in to add a comment