[Digikam-devel] raw speed in gqview

[Arnd Baecker]

On Wed, 8 Nov 2006, Gilles Caulier wrote:

> On Wednesday 08 November 2006 09:35, Arnd Baecker wrote:
> > Hi,
> >
> > before using digikam I used gqview
> > (http://gqview.sourceforge.net/) a lot.
> > Recent versions have quite a few interesting features,
> > in particular it can display the jpeg embedded in several
> > camera raw image formats.
> > It turns out that this is fast, i.e. noticeably faster than F3 in digikam.
> > E.g. for EOS 350D .cr2 files it takes digikam
> > a bit above 2s whereas gqview takes around 1s on my machine.
> > Moreover, changing back and forth between two images is instantaneous
> > (so obviously some caching is used for this.)
>
> This is the same way : dcraw extraction. In fact the method used by both
> programs is the same.

Looking at gqview's source:
format_canon.c was a patch by D.M. German,
  and a link to http://turingmachine.org/~dmg/libdcraw/gqview/
is given.
For format_fuji.c, format_nikon.c and format_raw.c
it is stated that
"Original version 2005 Lars Ellenberg, base on dcraw by David coffin."

But since then it seems (at least according to the CVS history)
no direct transfer of from David Coffins dcraw has happened.
So this code-base has been developed separately from then on.
(I.e.: if newer versions of dcraw are much slower than
the one around 2005, this might explain also part of the speed difference)


> The diference can be explained like this :
>
> - digiKam use a kioslave, and this can be slowest than to use a direct loading
> on the main app thread.

So this could be the main reason for the approx. factor of
2 speed difference.
Hmm, if this is true, another reason to go for the
"libkdcraw" library you suggested?

> - digiKam use an embedded dcraw implemention witch is compiled with the same
> option than digiKam. If you is in debug, dcraw will do it too.

I just checked my installation log and all I find is "-O2",
(in contrast, the gqview I compiled had:  "-g -O2")

> - digiKam do not use a cache mecanism. This is the real problem than we will
> fix in the future.

That would indeed be great!

I think there was a B.K.O entry on this ...
Whom could one ask to get BKO working again (or is it just me having
problems??). I can successfully ping to bugs.kde.org, but no http
requests work.

Many thanks, Arnd