HD playback performance

Devnn · n00b Joined: 03 Oct 2006 Posts: 25

I'm preparing my box for the growing HD video content available.
I've, however, just come across two samples, same resolution, one playing fine with ~50% CPU utilisation and the other is choppy with 100% usage.
My CPU is a pentium 4 3GHz. I'm considering an upgrade because of this.

I have no CPU flags enabled really, but isn't their effect on performance minimal? With choppy video, an improvement of 10% or even 20% would not be enough to be safe on this system. That's why I'm considering a dual core instead of the plain P4.

Does anyone have any tips to improve the playback by software? Is it possible to re-encode samples to the 50% CPU utilisation type losslessly?

col · Posted: Tue Aug 07, 2007 8:21 am Post subject:

you should have :

-msse3

make a big diff with video....although a P4-3.0G should have silky smooth HD playback consider I do with a slower CPU.

Gusar · Posted: Tue Aug 07, 2007 9:17 am Post subject: Re: HD playback performance

Devnn · n00b Joined: 03 Oct 2006 Posts: 25

Doh. You're right, I mixed up the outputs. The ffodivx is playing smoothly and the ffh264 is not.
I'll recompile mplayer and ffmpeg with

Gusar · Posted: Tue Aug 07, 2007 10:38 am Post subject:

Devnn · n00b Joined: 03 Oct 2006 Posts: 25

I recompiled mplayer with mmx, mmxext, sse, sse2 as use flags. And the 50% load video is now 10-20% and the 100% is now 50% and not choppy anymore. YAY! Just got to figure out how to give the same performace to xine now as it doesn't has CFLAGS options as use flags. Hopefully the CFLAGS in make.conf have an effect.

Thanks to everyone.

Btw. ssse3 is only for core2 CPU's. Is that what you meant, col?

col · Posted: Tue Aug 07, 2007 11:50 pm Post subject:

Gusar · Posted: Wed Aug 08, 2007 7:40 am Post subject:

sse3 (streaming SIMD extension 3) and ssse3 (supplemental streaming SIMD extension 3) are two different things. There is no sse3 support in mplayer, but there is ssse3. Only Xeon 5100 Series and Core 2 processors have ssse3.

Devnn · n00b Joined: 03 Oct 2006 Posts: 25

Binaries optimizations and dedicated support in the player seem to be two different worlds.

I recompiled ffmpeg, xine-ui and xine-lib with march=precott. This doesn't make any significant difference. Mplayer, however, just flies with the extended CPU instructions enabled in the USE flags. When running xine, it doesn't output anything about mmx support etc. like mplayer does.

I know mplayer has a ebuild that overwrites CFLAGS options and seem to take them from use flags instead, but xine doesn't show anything about CPU features anywhere.

Does anyone know how to make xine take avantage of mmx etc the way mplayer does?

Furthermode I don't see how mplayer can improve video playback so much by itself when it is using ffmpeg for decoding the videos. Doesn't mplayer and xine both use ffmpeg for video decoding and thus should have the same performance? Both programs had exactly the same performance before I enabled mmx etc. support in mplayer.

Gusar · Posted: Wed Aug 08, 2007 1:31 pm Post subject:

Devnn · n00b Joined: 03 Oct 2006 Posts: 25

Both the mmx flag and the EXTRA_ECONF variable helped on the performance in xine. The sample now play decently. Mplayer however still has better performance and wasn't affected by the ffmpeg recompilations. It's like mplayer uses another en embebbed ffmpeg, which would explain how useflags on mplayer could maybe improve performance of decoding. Either way, I just use mplayer, and xine is obviously affected by how ffmpeg is build.

Xine has a lot of settings, some of them could be impacting performance. Mplayer just seems the way to go if you want guaranteed video playback performance.

Gusar · Posted: Wed Aug 08, 2007 11:55 pm Post subject:

Devnn · n00b Joined: 03 Oct 2006 Posts: 25

This clarifies it all. Weird how nobody worries about an inefficient ffmpeg in these times with HD video emerging. Maybe people just buy new hardware, like I almost did.