494116 - Video/Audio goes out of sync under high CPU load on Windows

Status: RESOLVED FIXED

(Core :: Audio/Video, defect)

Description

[Biju]

Bug #466699 fixed many video/audio out of sync issues.

But I see few cases where is still happening
1. when there is other processor load.
2. after skipping video


Steps Case 1.
1. in firefox goto http://tinyvid.tv/show/d3h9ycgenax0
2. play video 
3. run an application which consume processor power too much
   eg:- Compiling C++ code, run VMWare etc.
4. come back to video in firefox

Result:
Video/Audio go out of sync



Steps Case 2.
1. in firefox goto http://tinyvid.tv/show/d3h9ycgenax0 
2. play video 
3. skip video backward and forward


Result:
Video/Audio go out of sync

Comment 1

[Biju]

I am sorry, I accidentally added everybody to cc

Comment 2

[cajbir (:cajbir)]

Case 2 is bug 493958. I suggest changing this to "Video/Audio goes out of sync under high CPU load on Windows" or something like that.

Comment 3

[Biju]

Retested after bug 493958 got fixed, still this issue is there.
Even when there is no load, this will happen when we watch a long video,
and its very annoying while seeing 
http://www.double.co.nz/video_test/interview.ogg
Here some small slippages will cumulative to show a big difference by 30 minutes.
So we may want to check the sync every minute and re-sync by skipping frames
or pausing audio when there is no sound.

Comment 4

[cajbir (:cajbir)]

Unfortunately the audio backend we are using does not support getting the audio playback time which is why the sync Is drifting. We'll need to investigate a bit to see how to support this on windows.

Comment 5

[Matthew Gregan [:kinetik]]

Attachment: patch v0

Trivial patch.  Use the system audio timers on Windows by requesting SA_POSITION_WRITE_HARDWARE.

This fix seems to fix at least some cases of bug 494120, too.  Once the clock got too far ahead of the video, we might never catch up, so we stop showing frames at all (I ran in to this case while testing in a VM with heavy load on the host and in the VM).

Comment 6

[Matthew Gregan [:kinetik]]

Forgot to mention, but should go without saying: mochitests and reftests pass on Vista in a VM.

Comment 7

[Robert O'Callahan (:roc) (email my personal email if necessary)]

Comment on attachment 380992 [details] [diff] [review]
patch v0

This is a very simple change that improves A/V sync on Windows.

Comment 8

[Robert O'Callahan (:roc) (email my personal email if necessary)]

I think we still need approval before this can land on trunk

Comment 9

[Serge Gautherie (:sgautherie)]

Comment on attachment 380992 [details] [diff] [review]
patch v0

>+  sa_position_t positionType = SA_POSITION_WRITE_SOFTWARE;
>+#if defined(XP_WIN)
>+  positionType = SA_POSITION_WRITE_HARDWARE;
>+#endif

Nit: why not use an #else instead of double initializing?

Comment 10

[Matthew Gregan [:kinetik]]

Personal preference.

Comment 11

[Mike Shaver (:shaver -- probably not reading bugmail closely)]

Comment on attachment 380992 [details] [diff] [review]
patch v0

a191=shaver

Comment 12

[Matthew Gregan [:kinetik]]

Pushed http://hg.mozilla.org/mozilla-central/rev/2f8493c326dd

Comment 13

[Matthew Gregan [:kinetik]]

Pushed http://hg.mozilla.org/releases/mozilla-1.9.1/rev/77bcd5216032