Hello Gimp Users and Developers,

This is a follow up of Bug 557950 (which in fact isn't a bug, according to Sven Neumann ;-)
http://bugzilla.gnome.org/show_bug.cgi?id=557950

As described in the “Bug”, scaling in Gimp 2.6 series is far slower, than it was in 2.4. Sven Neumann commented:

“We have completely changed the scaling implementation. The new algorithm is slower for some cases, but that is not a bug.”

Since there is no explanation WHY the algorithm was rewritten, I guess 2 possible reasons:

1.)The old code did something wrong in some cases 2.)The new code was necessary due to GEGL integration

For the first point, I compared scaling results from 2.4 and 2.6, and they are (ignoring some harmless alignment issues) 100% identical (using difference blend mode). I also cannot remember, that in the past years, the scaling routine in Gimp produced noticeable wrong results. (Beside the lanczos interpolation, that didn't work right, when it was introduced)

So my question is, isn't it possible, to have both algorithms in Gimp, and let the user decide which one he wants to use? (Option in Scale Dialog)

If it was due to point 2, the GEGL integration, than can we expect a faster version of the new scaling routine? Or will it be automatically faster, when GEGL is integrated more/better?

The current situation draws some users (not myself) to not use Gimp 2.6, and stick with 2.4 instead, because the difference in speed is so dramatically.

Sincerely, Claus Berghammer

Hello...

@Eric P:
Upscaling image:
500px -> 5000px (bicubic):
Gimp 2.6.1: 35,21 sec
Gimp 2.4.7: 6,9 sec

Downscaling layer: Image is 5000x5000 px, 2 white layers Scaling top layer to 2500x2500px (bicubic): Gimp 2.6.1: 7,85 sec
Gimp 2.4.7: 4,78 sec

@Sven Neumann: Thanks for your hint on the related bugthread. I will read it carefully to better understand the whole thing.

Claus

Eric P wrote:

Claus Berghammer wrote:

Hello Gimp Users and Developers,

This is a follow up of Bug 557950 (which in fact isn't a bug, according to
Sven Neumann ;-)
http://bugzilla.gnome.org/show_bug.cgi?id=557950

As described in the “Bug”, scaling in Gimp 2.6 series is far slower, than it
was in 2.4. Sven Neumann commented:

“We have completely changed the scaling implementation. The new algorithm is
slower for some cases, but that is not a bug.”

Since there is no explanation WHY the algorithm was rewritten, I guess 2 possible reasons:

1.)The old code did something wrong in some cases 2.)The new code was necessary due to GEGL integration

For the first point, I compared scaling results from 2.4 and 2.6, and they
are (ignoring some harmless alignment issues) 100% identical (using difference blend mode). I also cannot remember, that in the past years, the
scaling routine in Gimp produced noticeable wrong results. (Beside the lanczos interpolation, that didn't work right, when it was introduced)

So my question is, isn't it possible, to have both algorithms in Gimp, and
let the user decide which one he wants to use? (Option in Scale Dialog)

If it was due to point 2, the GEGL integration, than can we expect a faster
version of the new scaling routine? Or will it be automatically faster, when
GEGL is integrated more/better?

The current situation draws some users (not myself) to not use Gimp 2.6, and
stick with 2.4 instead, because the difference in speed is so dramatically.

Sincerely, Claus Berghammer

I'd be curious to see some benchmarks comparing 2.4 and 2.6 in this regard so that we know just how dramatically different the speed is.

Eric P. _______________________________________________ Gimp-user mailing list
Gimp-user@lists.XCF.Berkeley.EDU
https://lists.XCF.Berkeley.EDU/mailman/listinfo/gimp-user