[mythtv-users] ffmpeg and multiple CPUs. Was: Should I be switching to 0.21?

[Phil Bridges]

On Mon, Apr 14, 2008 at 1:07 PM, Tom Dexter <digitalaudiorock at gmail.com> wrote:
> On Sun, Apr 13, 2008 at 1:33 PM, Tom Dexter <digitalaudiorock at gmail.com> wrote:
>  >
>  >  I just wanted to report back on my experience regarding CPU usage.  I
>  >  upgraded today from 0.20.2 to 0.21 SVN 16944 under Gentoo.  First of
>  >  all, the upgrade from everything I've tested appears to have gone
>  >  flawlessly.
>  >
>  >  My frontend (dedicated frontend only) is a Dell Dimension 4700C 3.0
>  >  Ghz with hyper threading enabled.  Under 0.20.2 I was using libmpeg2
>  >  and bob de-interlacing.  It should be noted that I'm using Viktor's
>  >  patch to allow bob de-interlacing on my 1080i interlaced display:
>  >
>  >  http://svn.mythtv.org/trac/ticket/2903
>  >
>  >  ...though other than allowing me to use bob, that shouldn't be a factor.
>  >
>  >  First thing after upgrading I setup my own custom playback profile
>  >  group to mimic my old settings with only one entry as follows:
>  >
>  >  Resolution: >= 0 0
>  >  Decoder: libmpeg2
>  >  Video Renderer: xv-blit
>  >  OSD Renderer: softblend
>  >  OSD fade: ON
>  >  Deinterlacer: bobdeint
>  >  Secondary Deinterlacer: none
>  >  Filters: none
>  >
>  >  For max cpus under the Decoder I assumed 2 made sense for my hyper-threaded CPU.
>  >
>  >  Because of all the talk about increased CPU usage in 0.21, I did
>  >  several checks before upgrading, noting the CPU usage on the frontend
>  >  for 720p (that doesn't require de-interlacing) as well as 1080i (which
>  >  of course does) for several different OTA channels, both watching
>  >  recordings and livetv (though those really didn't differ).
>  >
>  >  I can say beyond question that the CPU usage under 0.21 with the above
>  >  profile is identical to that under 0.20.2.  As a matter of fact, on
>  >  occasion when watching 720p it's slightly less.
>  >
>
>  I was experimenting with some changes to my above profile and I've
>  made a rather interesting discovery.
>
>  On my frontend I found that using the Standard (ffmpeg) decoder in
>  place of libmpeg2 in the above profile used a fairly significant
>  amount of extra CPU.  In 0.20.2 I also found it used more, but this
>  difference was quite a bit more significant.
>
>  Playing 720p recordings with my above profile (with libmpeg2) used
>  18-20% of both hyper-threaded CPUs, while the exact same profile with
>  ffmpeg used around 38%.  Playing 1080i recordings with my above
>  profile (with libmpeg2) used slightly less than 30% of both CPUs,
>  while that profile with ffmpeg used 50% or more, with mythfrontend
>  often using 100% of one CPU.  I believe that the difference in 0.20.2
>  was more in the 10% range as apposed to this near 20% change.
>
>  However...and this is the interesting part...if I use ffmpeg but
>  switch the number of CPUs to 1 instead of 2 the performance of the two
>  is virtually identical.  It seems that, at least with my
>  hyper-threaded 3 Ghz P4, ffmpeg works much better with a single core.
>
>  Very cool...now I can used captions if I want.  Speaking of ATSC
>  captions though...is there any way to control the font color of ATSC
>  captions?  I see all sorts of choices for fonts but nothing for color.
>   The grayish color I get right now over most video content can be
>  unreadable.  I also wish I could control the position...is that
>  controlled by the broadcaster?  I often have the captions chopped off
>  due to overscan.
>

That is interesting.  I wonder if the same applies for AMD chips?
IIRC, the on-screen documentation recommends not using libmpeg2 with
64-bit processors, but I may be incorrect.