[mythtv-users] What does the mythtv-vid branch and OpenGL renderer have to offer

[Mark Kendall]

On 2/6/07, Viktor Avramov <drvik at bigpond.net.au> wrote:
> What is the purpose of the mytht-vid branch??

Just in case you hadn't spotted it, Daniel laid out his aims a while back:-

http://cvs.mythtv.org/trac/ticket/2287

> What do the new opengl-renderer patches offer???
>
> From what I can understand, the 3D engine of the video card will be used to
> assist with video playback (which should be a good thing right??)  but now
> all of a sudden one needs a Geforce 6200 as a minimum just to playback SD
> material, whereas a GF4 series was fine before...

Firstly, and maybe most importantly from your perspective, using
opengl isn't compulsory. It's just another display option. If xv, an
FX5200 and bobdeint give you all you need, then just stick with that.

On the other hand, using opengl can (and it is my no means complete)
give you further options to improve the picture quality while
theoretically not increasing the cpu load-

 - the opengl osd is displayed at the screen resolution rather than
video resolution and makes full use of all of the screen (when
displaying 4:3 on a 16:9 dispay etc)
 - the double frame rate deinterlacers do work well (though still some
fixes to go in to svn) - think bob without the flickering.

While there are many other options to pursue with opengl, my personal
end game is to have a quality edge detection/motion adaptive
deinterlacer and lanczos or similar upscaling filter to allow me to
display my standard def sources at 1080p in realtime. Secondary to
that, it would be nice to see that happening significantly faster than
realtime for hardware assisted transcoding. (these are just my
thoughts/aims - Daniel and others may have different views!)

In terms of performance, while GPUs are certainly fast, working with
video is not a particularily efficient use of their capabilities and
hence the current GPU requirements. That said, Daniel has come up with
some ideas to speed up the fragment programs used for yuv to rgb
conversion and I've been looking at using agp/pci-express memory to
speed up the transfers to the video card (which appears to be a
significant bottleneck).

... and there is, of course, no reason why all of this cannot be
implemented in software - it's just nobody has written the code.

regards

Mark

(P.S. an FX5200 was bleeding edge hardware about 4 years ago...)