Re: [Gimp-developer] c2g improvements

[Øyvind Kolås <pippin gimp org>]

On Sun, Jun 3, 2012 at 10:46 PM, Jacek Poplawski <jacekpoplawski gmail com> wrote:

there are few things... reading from 3 different arrays is 3 times slower than reading from 1 array, at least under Linux... but the overall algorithm does too much things that's why I proposed to create "c2g lite" which will not be so mathematically correct but give similar effect... for instance is relative brightness really so important in envelopes.h? also the luts are generated everytime from start, while I found out that generating one big array and use subset of data give similar results

Caching the LUTs sounds like an optimization that should go in directly, though I doubt that is what is taking most of the time. Beware that stress/c2g should also work on HDR images and not only on images with data in the 0.0-1.0 range.. For some of the rationale behing stress/c2g see

http://pippin.gimp.org/papers/Kolaas_11_jist.pdf which is a publication based on the approach explored by these GEGL ops.

/Ø

-- 
«The future is already here. It's just not very evenly distributed»
                                                 -- William Gibson
http://pippin.gimp.org/                            http://ffii.org/