[vdr] Softdevice or Xine?
jouni.karvo at tkk.fi
Tue Sep 27 21:17:09 CEST 2005
Darren Salt writes:
> > The reason why a separate xine is nice is that currently the setup eats up
> > all the CPU cycles. [...]
> Hmm? I'm seeing around 30% CPU usage with gxine (2GHz Athlon, Radeon 9200,
> 266MHz bus, AGP 4x).
I used earlier a 2.6GHz P4 HT processor, and the processor load was
approx 20% or less. It was really a surprise that a 2.4GHz non HT P4
would be so much slower. So yes, I suspect too that there is
something weird going on.
It might be some setting in the kernel compilation parameters (no so
obvious as MTRR, though). Top reports quite low percentages for
different processes. 60% is shown to be user and 40% sy time, which
is weird since the highest %CPU is with XFree86 at approx 6%.
Perhaps something about copying data unoptimally or weird context
switching frenzy? (DMA?) I checked the kernel timer settings already
(now use HPET).
> > Btw. if someone has any hints on how to reduce used cpu cycles, they are
> > warmly welcome. And yes, I use the fglrx driver - without that the picture
> > is jerky.
> Something's evidently not right with your setup. Are you using Xv? Have you
> tried the native R200 support in a recent 2.6 kernel and X?
Yes, I use xv (xshm simply is too slow). I have not tried kernel
radeon driver - and the current kernel is 2.6.9 (I have not yet been
able to get the ATI driver to work on newer kernel versions)
> A patch which recently went into xine-lib CVS HEAD helps somewhat (it
> optimises the video buffer allocation somewhat, which affects things like
> software deinterlacing).
There is something else fishy, since changing software deinterlacing
options does not change cpu usage considerably.
Thanks for your comments.
More information about the vdr