[Digikam-devel] [Bug 305382] editing NEF gives artefacts

[Gilles Caulier]

https://bugs.kde.org/show_bug.cgi?id=305382

Gilles Caulier <caulier.gilles at gmail.com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
   Version Fixed In|                            |lexa at lexa.ru
     Ever confirmed|0                           |1

--- Comment #7 from Gilles Caulier <caulier.gilles at gmail.com> ---
Marcel,

I can reproduce the problem with all RAW files from my test collection using
Bilinear demosaicing (and some other quality settings but not all).

I talk with Alex about this problem. He ask me if i can reproduce it with
official libraw tarball... I can confirm that it's not reproducible...

I written some test program to check if problem is relevant of :

1/ multithreading from digiKam core ==> no.
2/ DImg image loader ==> no.
3/ Extra compilation flags set by KDE ==> no.
4/ Check OPenMP => yes.

To reproduce the problem, you need a multicore processor, and libkdcraw/libraw
must be compiled against OPenMP. If you disabled OPenMP support, it's work.
There is no artefacts, but demosaicing is not parallelized.

I suspected in first a bug in Libraw. But the problem is that i cannot
reproduce the problem when it's compiled/linked against OpenMP.

THE question is why ???

- I checked again and again all CMake rules from libraw. I well separated
libraw compilation rules and libkdcraw into cmake scripts, just to be sure. I
take a care about all definitions passed to compile libraw in official tarball,
checked if i compiled libraw from libkdcraw with same dependencies than offcial
libraw (lcms, rawspeed, gpl2 and gpl3 extra pack, libjpeg, libjpasper) : It's
always reproducible... 

I cannot found the problem, and i don't have any idea... Help is welcome...

Gilles Caulier

-- 
You are receiving this mail because:
You are the assignee for the bug.