[mythtv] Re: request: OSX mythfrontend and winmyth for CVS backend

[Jeremiah Morris]

On 15 Nov 2004, at 12:57 PM, Gavin Hurlbut wrote:

> Woohoo!  I compiled CVS as of last night, and sure enough, it came up 
> nicely
> on my 500MHz G3 iBook.

Congratulations!

> This gives me enough ammunition to start looking at tweaking for my 
> poor
> non-Altivec processor :)

Given that I can barely run video on my 800Mhz Altivec-enabled TiBook, 
that's an ambitious job, but I certainly wish you the best of luck!

On my Shark runs, the bulk of the time is evenly split between 
libavcodec's video decoding, and QuickTime's YUV conversion and 
blitting.  Nigel and I have discussed an alternate video driver using 
OpenGL rather than QuickTime, which might be faster, so that's one 
avenue to explore.  (QuickTime is supposed to use hardware acceleration 
when it can, but I don't know how to put it to the test.)  On your G3, 
the non-vectorized libavcodec routines may overwhelm any QuickTime 
speed penalty, though.

> Nice.  Just please make sure that the CVS way works too :)  hehe.

What we need now is more folks who can keep up with CVS and the dev 
list, and complain loudly when patches or checkins break the Mac build. 
  :->  The more eyes we can have on the bleeding edge, the easier it 
will be to keep the OS X version running.  Believe me, I'd rather be 
building binaries than worrying about tree breakage.

> I ran Shark on my iBook, and found out something interesting.  about 
> 20% of the
> CPU time is used to resample my audio.

Ouch.  The resampling should be taking place natively in the Core Audio 
driver, but the audio framework provides a way to use libsamplerate 
instead.  Can you send me your Shark log?  If Core Audio is being a CPU 
hog, I can send you patches to use libsamplerate for the conversion, 
which might work better.

- Jer