Patch for speeding up the painting of the autobrush

[Cyrille Berger]

On Tuesday 02 February 2010, LukasT.dev at gmail.com wrote:
> Here are the results:
> http://lukast.mediablog.sk/i/AutoBrush_70px_rotated-new-randomLines.png
> http://lukast.mediablog.sk/i/AutoBrush_70px_rotated-old-randomLines.png

What bother me is that when I draw at 1600%, I can't see a difference between 
valueAt and interpolatedValueAt (of course I do that with small mask, could 
the problem only appear with bigger mask ?). But I also noticed that there is 
a 0.5/0.5 offset with both valueAt / interpolatedValueAt, but I do not know if 
Adrian is done fixing that area, or if I am uptodate enough, or if there is 
something to fix in autobrush.

> Could you please test with and without patch and report issues?
> 
> What I want to implement futher:
> o There are settings where the brush is symmetrical so I plan to compute
>  1/4  of the brush mask
> o Lookup table for atan2 ( lookup table for valueAt??? :) seems also
>  possible
> o move the rotation to the end of the pipeline so we will be
>  computing the brush mask unrotated and rotate when compositing
What would be the benefit of that ?

> o if that's not enough mip mapping (like we do in gbr brush)

-- 
Cyrille Berger