[mythtv] MacOS X MMX yuv420_2vuy Patch

[Todd Ignasiak]

On 9/13/06, Andrew Kimpton <awk at awkward.org> wrote:
> Quoting Todd Ignasiak <ignasiak at gmail.com>:
>
> >
> > The 0.20 final seems to be working very well for me.  I haven't used
> > it a lot yet, but thus far it's been stable and performance is
> > improved over previous versions.
> >
> > I noticed two minor issues, can anyone else confirm them on their Mac?
> >
> > 1) When I turn off the vector acceleration option, I no longer get a
> > crash.  I now get a distorted picture, where the colors are primarily
> > purple and green, and the distorted image is tiled on the screen
> > several times.
> >
> What you're seeing is the original YUV420 planar data. The Y/Luma
> plane will look a lot like a black/white version of the video. The
> UV/Chroma planes will tend to be between purple and green.
>
> Since the data is laid out in memory as seperate planes (Y then U then
> V) you'll see them as tiles on the output window if they're not being
> convered prior to display.
>
> If you turn off the 'accelerate' option then the code doesn't use the
> MMX conversion routines, and on Intel machines also doesn't attempt to
> setup QuickTime to do the conversion (since that crashes Intel
> machines). Hence the od images.

Thanks for the explanation.. that makes sense in terms of the tiling.
  I was just surprised that it was not crashing because all the
previous versions crashed when defaulting to QuickTime for the
colorspace conversion.

After looking at the code a bit closer, I see that the
'get_yuv2vuy_conv' function now sets it to a normal C conversion if
not MMX/Altivec (the old code set it to NULL, and QT crashed).    So,
in theory, video output should still work without the vector code,
just not as fast.    But, at least for me, it is not working as
expected.

> > 2) AC3 to SPDIF passthrough results in messed up sound.   After
> > glancing through the code, I don't think SPDIF passthrough is in the
> > Mac mythtv code.  Has anyone else tried AC3 audio?
>
> There's some commentary about this on the Wiki - they sense there is
> that some required byteswapping is missing. I don't have a AC3/SPDIF
> decoder attached to my Mac so I can't give any further insight.

Yeah, that was my comment in the Wiki, but I'm not confident that
that's really the problem.  I initially thought it was that because
the effect sounds very much like the standard audio output on early
x86 builds -- which turned out to be an endian issue.    But, I didn't
see any AC3/SPDIF code in the Mac audio output, so the similarity in
sound may not be relevant.      I was just trying to get confirmation
from someone else, PowerPC or x86, to verify that it's not just me.