Video playback using the Wandboard VPU

This subforum is for discussing blogposts/articles and for other comments about the wandboard site.

Re: Video playback using the Wandboard VPU

Postby nicolauz » Tue May 27, 2014 4:52 pm

I re-checked the two commits and did only the "minimal" thing: re-adding the "if (mem->cpu_addr != 0)" check.
I've attached a patch ... for simplicity reasons I didn't do any intending! So don't commit this and push it somewhere, because it's very wrong (code style vice). :)

Now I got rid of the errors!

And: the slight memory leaks I mentioned seem to be unrelated to video-playback .. I think :)
Attachments
readd-cpu_addr-neq-zer_check.patch
(829 Bytes) Downloaded 240 times
nicolauz
 
Posts: 7
Joined: Tue May 06, 2014 8:29 am

Re: Video playback using the Wandboard VPU

Postby CruX » Thu May 29, 2014 11:09 am

Those patches made the overall situation better but I still had numerous issues with 3.0.35.
I'd strongly recommend to use the newer 3.10.17 (if this is possible in your scenario), it works much better.

Cheers
CruX
 
Posts: 68
Joined: Sun Oct 27, 2013 1:29 pm

Re: Video playback using the Wandboard VPU

Postby nicolauz » Mon Jun 02, 2014 3:19 pm

not that easy @ 3.10.17

First I don't have a 3.10.17 tree for the customer board I'm dealing with .. second I'm seeing significant opengl performance problems with 3.10.17 on nitrogen6x.


Is any of the issues a memory leak?... I'm asking because:

After tuning down everything (e.g. stopping most services, disabling journald logging,...) I discovered that I still have a very small memory-leak.
It's irregular, but it's a couple of 10kb ever couple of render-loops (not everytime!). So after 5000-6000 play cycles memory is used-up and the process freezes. It's definitely user-land memory because it's free again after killing the process (=> restarting the process results in it playing again).

/* edit
Actually testing whether it's qtmultimedia or gstreamer was simple ... just the main.c I previosly posted .. ;)
And yes I see the same memory leak ... I'll let it run for a while now to see it's behavior .. next item: testing on nitrogen6x
/*

... I already know that it's imx6 (maybe even customer-board) specific.
nicolauz
 
Posts: 7
Joined: Tue May 06, 2014 8:29 am

Re: Video playback using the Wandboard VPU

Postby Tapani » Tue Jun 03, 2014 2:57 am

First, glad it is not a kernel space memory leak. That would have been more severe.

Second, have you tried hunting it down using tools like Valgrind?
Tapani
Site Admin
 
Posts: 685
Joined: Tue Aug 27, 2013 8:32 am

Re: Video playback using the Wandboard VPU

Postby nicolauz » Tue Jun 03, 2014 1:48 pm

I tried ... but Qt5[Multimedia], galcore and gstreamer are currently not fit to be used with valgrind.
The amount of false-positives (or negative, how ever you want to see it) is so massive that it's impossible to see any real leaks.

It would require massive investigations to find all the rules so that valgrind will ignore a lot of those false-positives for Qt5, galcore and gstreamer .. and then it's a matter of not making any mistake during that work to not hide the right-positives ...

I'm not seeing myself being able to do that (not enough time and not enough experience writing proper exceptions for valgrind).

Has anyone else had some luke with valgrind? .. any good versions? an additional set of exceptions?



Also:
I further tested, it seems the memory leak is also there when using gstreamer only, and it's also happening on nitrogen6x .. so I would strongly assume it's also happening on wandboard. (use my main.c from previous posts to test, you'll start to see a leak after a couple of hundred play-cycles).


Greets
Thomas
nicolauz
 
Posts: 7
Joined: Tue May 06, 2014 8:29 am

Previous

Return to Article talkback and site feedback

Who is online

Users browsing this forum: No registered users and 2 guests