Re: [Gimp-user] GIMP 2.10-RC1 Previews Loading Slowly

[Øyvind Kolås <pippin gimp org>]

On Sat, Apr 14, 2018 at 5:44 AM, Robert Bieber <gimp biebersprojects com> wrote:
> I've had some similar issues working with large images, here's some things
> I've found.  First of all, if your image is in 16-bit precision, your
> operations are going to be slow.  GEGL is apparently optimized for 8-bit int
> and 32-bit float, so paradoxically you'll actually get a much smoother
> experience if you change the image precision to 32 bit floating point.
> Trying to,   That's gotten the new GIMP to be pretty usable for me.  It's
> still not as zippy as the 2.8 series was, but it's also crunching a lot more
> data over 8-bit images.

A perhaps more accurate description is that most operations are now performed
on linear 32 bit floating point data, regardless of which precision
pixels are stored with.
Fast path conversions from storage formats of 8bpc and 32bit float
have been around
in babl for many years, but for most permutation of gray/rgb/alpha in
16bit precision,
both float and integer no such short cuts existed and babl would
roundtrip the data to
64bit floats. As of babl-0.1.46 released earlier this week most gaps
for these pixel-formats
have been filled with fast paths. There is still room for making the
16bit code paths faster with
SIMD but GIMP-2.10rc2 should already be much better than 2.10rc1, for
further details
see also https://www.patreon.com/posts/babl-fast-path-18052156

/pippin -