Issue 908314 link

Starred by 1 user

Issue metadata

Status:	Started
Owner:	hiroh@chromium.org
Cc:	posciak@chromium.org tfiga@chromium.org
Components:	OS>Kernel>Video
EstimatedDays:	----
NextAction:	----
OS:	Chrome
Pri:	1
Type:	Bug
videoshortlist

video_VideoDecodeAccelerator import case failed on kevin

Project Member Reported by hiroh@chromium.org, Nov 26

Issue description

import case failed because of mismatch of md5sum values of decoded frames on kevin.
https://stainless.corp.google.com/search?view=list&build=R72&test=video_VideoDecodeAccelerator.vp9.import&board=%5Ekevin%24&first_date=2018-10-29&last_date=2018-11-26&status=GOOD&status=WARN&status=FAIL&status=ERROR

Comment 1 by hiroh@chromium.org, Nov 26

Cc: posciak@chromium.org tfiga@chromium.org
Components: OS>Kernel>Video
Labels: videoshortlist

I am looking into this today.
Looks like the md5sum value on vp9 is deterministic. I saw the dumped yuv and the image looks correct. The difference from the correct image is really small, some bits in the middle of image. I wonder there may be something wrong in hardware or driver on rk3399 devices. But it may be hard to figure out the place...
I attached dump yuv images.
The difference between them can be seen by hexdump and diff tool (like meld).

correct_test-25fps.vp9.0100_320x240_I420.yuv
112 KB Download

bad_test-25fps.vp9.0100_320x240_I420.yuv
112 KB Download

Comment 2 by tfiga@chromium.org, Nov 30

Perhaps some cache synchronization issues? Is the difference always the same for every run? Could you convert both correct and bad to normal image files (e.g. png)? Also, there are some image diff tools, even online if you google for "image diff online". I guess it could be convenient to see the diff using such tool.

►

Issue 908314 link

Issue metadata

video_VideoDecodeAccelerator import case failed on kevin

Issue description

Comment 1 by hiroh@chromium.org, Nov 26

Comment 1 Deleted

Comment 2 by tfiga@chromium.org, Nov 30

Comment 2 Deleted

Sign in to add a comment