mpeg4 decoder memory leaks

[atvan]

We have built inIntel mpeg4 and jpeg decoders in our product. Some of our clients are complainning about memory leaks of mpeg4 after thousands switches between different cameras. I tried to narrow down the leak and foundthe problemis with the mpeg4 decoder.The leak appears when deleting the decoder after the decoding is finished. ifI putthe decoderinto a umc_thread, deleting thedecoder willcause theleaks. But it works fine if I put it in the main loop. There is no problem with Jpeg decoder. I hope you can help me to solve this problem in some way.

The ipp sample version I'm using is 5.3.095.

With the attached code you can find
1. mpeg4_memory_leak_report.jpg, performance chart of private bytes and handle leaks
2. raw data under .\frames\ for running the test
3. three scenarios,twowhere the mpeg4 decoder works fine, and one where it leaks
4. jpeg decoder works fine in all the three scenarios

Thanks.

[Vladimir_Dudnik]

Could you please check this with IPP 6.0 samples. It might be fixed already in the latest version.

Vladimir

[atvan]

Quoting - Vladimir Dudnik (Intel)

Could you please check this with IPP 6.0 samples. It might be fixed already in the latest version.

Vladimir

Vladimir,

I repeat the test on 6.0, and I can see the same problem. The leak seems more serious on my desktop. When running on my laptop, the leak is less but after around 100 switches, I can still clearly see the leak.

I ransimple_playerin 5.3 with an mpeg4.avi media, after hundreds of switches, I can also see a leak. But I'm notclear if it's caused by mpeg4, because if I disable the video channel andplay only audio channel, the leak still exists.

I'm running the same test over 6.0 with simple_player.Though the testcode is still running, I don't think6.0 makes any different.The private bytes isstill growing slowly.

Regards

[atvan]

When I said switch I mean let the simple_player keep running PlaySingleFile. Now I can seethe 6.0simple_player is leaking after a couple hundred of runs.

For some reason I can not editmy previous reply. And it takes very long to open my post. Is it weird?


Thank you for your replay.

[Vladimir_Dudnik]

If you may provide test case or more details how do you reproduce the issue with IPP 6.0 sample it will help us in investigation.

Thanks,
Vladimir

[atvan]

It works fine with 6.0 so far. I guess I didn't do it right when updated to 6.0.