[mythtv] EPIA M10k - XvMC: Very high CPU (even when paused!!)

Justin Hornsby justin.hornsby at gmail.com
Thu Dec 28 19:35:31 UTC 2006


On 28/12/06, Colin Guthrie <mythtv at colin.guthr.ie> wrote:
> Justin Hornsby wrote:
> > I see pretty much the same effect on my epia M10k-based frontend.
> > Xine, playing a recording (DVB MPEG2) takes less than 20% CPU with
> > xvmc, whereas mythfrontend is using around 50%.
>
> Yup sounds similar. Out of curiosity, what %age do you get if you play a
> recording and pause it?
>
> > Running mythtv 0.19 wasn't so resource hungry as I recall, so I think
> > it's definitely a change since then which is doing it.
>
> Yup I think I can concurr. I'd been running quite an old pre 0.20 SVN
> snapshot for a long time without many problems. But relatively recently
> I've started to notice these propblems.
>
> Are you running 0.20 or SVN? I'm running SVN.
>
> > What we need to do is build mythtv with debugging support enabled &
> > track it down properly - to get to the bottom of it, that is.  I can
> > live with it, but it'd be nice to have it nailed so to speak.
>
> I have a debug build here but it runs on an NFS root machine. I should
> be able to install various debugging programs so I'll see what I can do.
> If anyone has any advice as to how to track it down, it would be
> appreciated.
>
> As I write this email, however I've been playing around playing
> different recordings.
>
> It now seems that it's intermitant. I've definately had probelms with
> playback today hense why I wrote my original email.
>
> That said, when I now look I can see my frontend's CPU is nice and low
> (~15%) during playback.
>
> I switched a few recordings and they all seem OK now. It's really weird.
> It's like it sometimes decides to use non-XvMC playback. Typically, I
> cannot get it to break now (using network control as FE is in the other
> room). Every recording I find is playing nicely, even the one that was
> previously giving me grief!
>
>
> Perhaps someone who is more familiar with the video codec selection code
> could have a quick look and see if there is anything obvious that could
> screw up which decode method is used.... Even if they point at some
> useful debugging to insert that would be useful.
>
>
> For example, in my log I've had these, once was from a recording that
> didn't want to use XVMC and caused high CPU, the other is one that "just
> worked" (tm). When I tried to play the original again, it bloody worked
> properly..... man that's annoying!
>
>
> Good:
>
> Input #0, mpegts, from '/mnt/media/scratch/livetv/1629_20061228013500.mpg':
>   Duration: 01:01:28.2, start: 89352.551011, bitrate: 3549 kb/s
>   Stream #0.0[0xb02], 1/90000: Video: mpegvideo_xvmc_vld, yuv420p,
> 704x576, 1/25
> , 10000 kb/s, 25.00 fps(r)
>   Stream #0.1[0xb03](eng), 1/90000: Audio: mp2, 48000 Hz, stereo, 192 kb/s
>   Stream #0.2[0xb05](eng), 1/90000: Subtitle: dvbsub
>   Stream #0.3[0xb04](eng), 1/90000: Data: 0x0000
>
>
> Bad:
>
> Input #0, mpegts, from '/mnt/media/scratch/livetv/1629_20061227210000.mpg':
>   Duration: 02:06:26.6, start: 72854.927011, bitrate: 3833 kb/s
>   Stream #0.0[0xb02], 1/90000: Video: mpeg2video, yuv420p, 704x576,
> 1/25, 10000
> kb/s, 25.00 fps(r)
>   Stream #0.1[0xb03](eng), 1/90000: Audio: mp2, 48000 Hz, stereo, 192 kb/s
>   Stream #0.2[0xb05](eng), 1/90000: Subtitle: dvbsub
>   Stream #0.3[0xb04](eng), 1/90000: Data: 0x0000
>
>
> Ugly:
>
> Col.

Up until mythtv 0.20 I was using minimyth for my epia based frontend,
but it had serious trouble playing back dvb recordings which have
audio description tracks.  It only had problems if I was using xvmc
(not using xvmc isnt an option due to the very high cpu usage) and
deinterlacing.  Playing the same recordings with timestretch enabled
worked (and looked) great.

The problem showed itself as momentary pauses in playback every 20
secs or so, and mythfrontend -v playback showed me lots of NVP
prebuffering pauses & dropping frames to maintain audio sync.

Theories about the cause of the problem ranged from an alsa driver
issue to a problem with the maths the code uses to determine the a/v
sync.

Certainly food for thought.
-- 
Justin Hornsby
Moderator of www.mythtvtalk.com
email: justin(dot)hornsby(at)gmail.com
web: http://www.juski.co.uk


More information about the mythtv-dev mailing list