[QLab] video crossfade

[Christopher Ashworth]

On Nov 2, 2007, at 11:04 AM, luckydave wrote:
>
> try making a 512x384 video, and telling QLab to make it a
> resizable window with a size of 1024x768. OpenGL stretches it,
> and it usually looks pretty good. It'll keep file size down, so  
> you'll see less hiccups.
> Also it gives QLab fewer pixels to digest when it's working out your
> crossfade.

Note that this does not help with the crossfade issue because the  
final number of pixels is the same, and that's what produces all the  
work.

See this post for a detailed discussion:

http://lists.figure53.com/htdig.cgi/qlab-figure53.com/2007-August/002544.html

In general, crossfades are going to be bottlenecked by the graphics  
card, rather than the CPU or hard disk.  I'd guess that messing with  
the media encoding, etc, will not help in this case.  It's most likely  
going to be a question of: "how many pixels is the graphics card  
calculating on the final display device?"  The only way to modify the  
result for that question is to:

1) reduce the number of pixels in the final display
2) get a better graphics card

Without knowing anything about how other software does crossfades, I'd  
take a guess that they're putting some of that work on the CPU, and  
thus getting more power to do smoother fades when the GPU can't cut it.

The point about crossfades not being ideal right now is valid, and I  
take the point.  I'll try to find some time to explore how I can  
optimize this further, possibly by pulling in the CPU power to augment  
the GPU, etc.  Or looking to see if there are any further  
optimizations I can perform with the current architecture.

We're in a bit of a transition period with graphics on the Mac right  
now, because Apple is pushing developers to use all the built-in Core  
Animation, Core Video, OpenGL technology for all our graphics needs,  
but the graphics hardware they're shipping  with the machines is not  
really the top of the line, so when people really push it (like full- 
screen crossfades on high-res devices on the GPU), we start to hit the  
bumps in the road of the current state of the art.

Anyway, I digress.  I hear your point and will try to allocate more  
time to look if there is anything even further I can do to improve the  
situation, as this is a common issue that folks bring up.

Thanks,
Christopher