Adding babl_chromaticities_make_space, babl_trc and
babl_trc_gamma to public api and ability to setup arbitrary
BablSpaces without loading from ICC profiles.
When viewing PNGs with attached ICC profiles the gegl binary produces black
if the two-table extension is active. Something is wrong in babl even accepting
these paths.
Øyvind Kolås [Wed, 20 Jun 2018 18:43:53 +0000 (20:43 +0200)]
babl: add ~ variants to gray color models
We do not add premultiplied sRGB trc variants even if this exists for RGB
since it is the type of premultiplied alpha cairo expects. The less code
that does non-linear compositing the better, excluding some possibilities
also reduces the combinatorial overhead of finding conversion paths in babl.
Øyvind Kolås [Wed, 20 Jun 2018 10:33:28 +0000 (12:33 +0200)]
babl: add R~G~B~ color models and pixel formats
From a given BablSpace which defines RGB chromaticities and
per component TRCs, the following are the defined formats:
R'G'B' with the TRCs applied
RGB linear data
and now a new one
R~G~B~ with the sRGB TRC, this permits any BablSpace to be the
native format of a buffer in an application while having both
a 'perceptual' in R~G~B~ and linear in RGB that is possible
to derive from it. Importantly it is also possible to recover
the babl space used to create the buffer.
Øyvind Kolås [Tue, 19 Jun 2018 17:29:48 +0000 (19:29 +0200)]
babl: rename functions that used to refer to specific gamma
The definitions of components and formats for linear and nonlinear formats in
model-rgb and model-gray are now using the TRC of a BablSpace, thus in the
format syntax ' now means nonlinear, not sRGB / 2.2 gamma - even though it ends
up meaning that for formats create without specifying the space.
Ell [Thu, 14 Jun 2018 05:54:52 +0000 (01:54 -0400)]
babl-fish: fix concurrent babl_fish()/process() calls with palette formats
In babl_fish(), don't call _babl_fish_rig_dispatch() on existing
path fishes during each lookup. We already call this function on
all path fishes upon creation (in particular, after commit b34a9672c8c9a42923a8f0b2650af30b1efe635c), so this isn't necessary.
This fixes an issue where the _babl_fish_rig_dispatch() call in
babl_fish() would temporarily modify the value of babl->fish.data,
causing a concurrent babl_process[_rows]() call in another thread
to pass the wrong 'data' value down to the conversion function. In
particular, this could cause palette conversions to segfault.
This API is internal only for now so we can change it, changing to use a
hard-coded pre-quantized chromaticities set for the internal sRGB space.
This preserve the gray axis matrix equalization performed for other spaces used
with babl, ideally we can improve the gray-axis equalization to match and
return to using it also for sRGB.
Ell [Wed, 30 May 2018 09:45:17 +0000 (05:45 -0400)]
babl-memory: allow babl_malloc(0) to succeed
Allow a babl_malloc() call with size == 0 to succeed, returning a
dummy allocation, instead of asserting on this condition. The new
babl-palette code tries to allocate an empty buffer for the radii
array when creating a single-color palette, which would previously
lead to a crash.
Ell [Fri, 18 May 2018 12:04:52 +0000 (08:04 -0400)]
babl-palette: speed up closest-color search
When constructing a palette format, calculate the distance between
each pair of colors in the palette, and for each color, construct a
list of all other colors and their distances from it, sorted by
distance. When searching for the closest palette color to a given
input pixel, based on the assumption that nearby pixels have
similar color, use the color list of the previous pixel's best
match, and use the triangle inequality to stop the search early.
Øyvind Kolås [Fri, 18 May 2018 09:02:04 +0000 (11:02 +0200)]
babl: use hashtable approach in double reference
Strictly speaking we should be doing the reference using
conversions to R'G'B'A double and using the distance metric
there; possibly getting a slight difference in result at
comparison ties.
In use GEGL ends up using both the really fast 8bit code paths
and the quite fast floating point code paths that use 8bit
hashtables for the lookup of conversion.
For working at acceptable performance with large indexed images, it is
recommended to run with BABL_TOLERANCE=0.0385 or highger set in the environment
when launching GIMP, this permits using a hashtable based optimization that
still isn't accurate enough according to babl-expectations.
Øyvind Kolås [Thu, 17 May 2018 10:53:42 +0000 (12:53 +0200)]
meson.build: synchronise with autotools version
WARNING: distributions should still not use meson to build packages - since
the meson build end up containing all instructions available on build host.
Debarshi Ray [Sat, 28 Apr 2018 23:15:46 +0000 (01:15 +0200)]
CIE: Add an SSE2 version of "RGBA float" to "CIE Lab alpha float"
On an Intel i7 Haswell, it now takes 0.13s to convert a 15 megapixel
buffer from "RGBA float" to "CIE Lab alpha float" instead of the
earlier 0.27s.
SSEx doesn't have integer multiplication or division operations, and
using bit shifts to implement integer divisions by powers of 2 seems to
introduce errors. Therefore, it was problematic to use the cube root
approximation from Hacker's Delight, which uses quite a few integer
divisions to make the initial guess. Instead, Halley's method of
approximating the cube root seems more SSEx friendly because the
initial guess requires only one integer division, which we can manage
by jumping through a relatively small number of hoops.
The scalar version of Halley's method seems to have originated from
http://metamerist.com/cbrt/cbrt.htm but that's not accessible anymore.
At present there's a copy in CubeRoot.cpp in the Skia sources that's
licensed under a BSD-style license. There's some discussion on the
implementation at http://www.voidcn.com/article/p-gpwztojr-wt.html.
Note that Darktable also has an SSE2 version of the same algorithm,
but uses only a single iteration of Halley's method, which is too
coarse.
Here's some more discussion on the cube root approximation algorithms:
https://bugzilla.gnome.org/show_bug.cgi?id=791837
This will permit slightly faster than the generic fast paths for some possibly
desired intermediate conversions. This also adds alpha stripping - which fills
in the last babl fast path currently reported missing on every launch of GIMP.
GIMP makes use of "Y' u8" to "R'G'B' u8" and we had no paths covering it,
making it go through floating point. This commit adds code paths for expanding
single grayscale u8 to R'G'B and R'G'B'A as well as R'aG'aB'aA u8.
extensions: add rgba8_rgb8 conversion that copies 24/32bit chunks
Slightly speeding up RGB u8 (without alpha) painting in GIMP, the rgba to rgb
and reverse conversions are used together with the very fast SIMD conversions
that operate on chunks of 4 components.
projects / babl.git / log