Intel® Integrated Performance Primitives
Deliberate problems developing high-performance vision, signal, security, and storage applications.
6711 Discussions

IPP 6.0 simple_player with -t2 weird behaviour

matthieu_darbois
New Contributor III
822 Views

Hello,

I tried to play an MPEG4 file I have to see the increase in performance while using multiple threads for decoding.

When using only one thread (-t1), I get 350fps decoding speed with almost no cpu used (.296s kernel time, .250s user time).

When using 2 threads (-t2), I get 400fps decoding speed but cpu usage go through the roof (13.5s kernel time, 5.312 user time).

I use a C2D E4400 (so 2 cores). The samples (6.0) were compiled with Intel Compiler 10.1.021 and IPP 6.0.0.062.

What I see is that a libiomp5md.dll!_kmp_launch_worker thread uses around 48% of cpu time doing mostly system calls. The libiom5md.dll loaded with simple_player is version 5.0.2008.215

I didn't find anything unusual while going through the source code. I know there's a little overhead when using multiple threads but it seems very unusual to have such difference.

Does anyone experienced the same problem ? Have you find the cause of this problem or is it normal ?

Thanks

ps : command line : simple_player.exe "file.mp4" -anul -vdx -t1 or simple_player.exe "file.mp4" -anul -vdx -t2

0 Kudos
3 Replies
matthieu_darbois
New Contributor III
822 Views

Hi again,

I've just compiled simple_player with intel compiler 11.0.061 and nothing has changed.

There's still that weird behaviour with the -t2 option.

0 Kudos
Vladimir_Dudnik
Employee
822 Views
Could you please try the same stream but specify to render to file (File Writer render, -vfw) instead of rendering to screen?
It might bean issue in renderthread synchronisation.
Vladimir

0 Kudos
matthieu_darbois
New Contributor III
822 Views

Hi,

I tried with -vfw and -vnul. The result is the same as before.

FYI, stream type is MP4ATOM, VideoType MPEG4, Resolution 640x480 and framerate 30fps.

0 Kudos
Reply