[mythtv-users] New Mac OS X Build Process--MacPorts

Jean-Yves Avenard jean-yves at avenard.org
Wed Aug 22 10:55:26 UTC 2012


On 22 August 2012 10:37, Yeechang Lee <ylee at pobox.com> wrote:
> I concentrated so much on look and feel in my last comparison of the
> prebuilt and MacPorts self-compiled mythfrontends that I completely
> neglected the completely unimportant topic of playback! Turns out here
> is a substantial difference. The MacPorts version uses 75% on Activity
> Monitor to play a MPEG-2 1080i recording with Kernel 2X deinterlacing;
> the prebuilt one uses about 115%. (Makes me now glad that for years
> I've only used self-compiled MythTV RPMs on Linux.) Being able to use
> VDA erases the gap; I see 45% with both frontends when playing a 720p
> h.264 file.

The difference you are seeing is entirely dependent on which version
of Qt you have used and how it was compiled and it depends on which
CPU you are using...

In my experience:
64 bits compile on Core 2 Duo is about 15% slower than a 32 bits build.
The Qt 4.8 (which I use in my build) is also slower than the Qt 4.6.
Unfortunately, using Qt 4.6 on Lion cause some artifacts and hang not
occurring with Qt 4.6.
The worse is that Qt 4.8 fixes some bugs, but introduces others, so
it's hard to say weither one is better than another.

What I am sure about, is that you can, most likely, achieve exactly
the same results as what macport is doing...

I don't believe a Kernel 2X deinterlacer would ever take 115% of CPU,
there must be something else at play.
My 2.4GHz Core 2 Duo mac mini plays any mpeg2 1080i with less than 65% CPU.

It also depends if you are using OpenGL or Qt painter.

Doing it over macports is the best way going forward for sure.... But
that would require people to build the code themselves.
If you have to package the application in a bundle, you will end up
rewriting most of the perl packager.


More information about the mythtv-users mailing list