[mythtv] Patch for Via XvMC VLD QMatrix Decoding

Ivor Hewitt ivor at ivor.org
Mon Sep 20 10:22:22 EDT 2004

Hi Isaac,

>>As I mentioned on unichrome enhancing the via osd is my project for this
>>week. :-)
>>Yes, it would be possible to patch up the BlendtoI44 to create a colour
>>list... but when I did this before the performance was, um, less than
>>impressive. Or to hack up a pseudo colour version of the palette...
>>(which produced okish results but didn't get rid of the OSD stutter).
>How are you going to get a 16 color palette to look right, when all the OSD 
>themes assume a full color palette?  That's the entire reason it's in b&w.
I was planning on producing a 16 color OSD theme to go with it.
I understand why it's blended down to b&w.... however the cpu overhead 
of creating the full yuv osd surface and then mixing it down to I44 is a 
killer on the EPIA.... and I thought I'd try a different approach.

>>However, I was planning on creating a new OSDI44Surface type, to hold
>>and handle the OSD in native format instead. I'll let you know if I get
>That'll be even worse performance - the current blending code is MMX 
>optimized, and if you change the internal surface format, you'll lose that.
But there would be no blending required to display the final osd, it 
would be held in I44 format throughout.
Of course, it might be a complete non-starter and I've probably 
overlooked something fundamental... I just thought I'd give it a bash.

I tried a couple of colorized via OSD's before... the first generated a 
pallete by picking most frequently used colours, the second just has a 
non-greyscale palette, with hardcoded detection to pick out certain 
colours in the blend code.
The first simply slowed things down even more... The second produced a 
reasonable OSD but the performance was no better and to be honest I 
found I preferred the grey scale!


More information about the mythtv-dev mailing list