> I've tried to avoid FFTW for a few reasons:
>
> 1) Huge code bloat -- this thing is not small. It takes a long time to
> build, and the code overhead for a Photoshop plug-in is significant.
Agreed. For most scientific programs, adding another shared library
to the link line isn't even worth thinking about, but if you are
writing embedded code or plug-ins, you may fall in the few percent for
which FFTW isn't appropriate.
> 2) Funky calling API -- you have to use a much more complex API to
> talk to it, mostly due to the prediction engine for what will be the
> most efficient implementation on that hardware. Yes, this is how
> they accomplished their s peed, but the pain for the programmer is
> significant.
I don't really see this. I changed my code from Numerical Recipes and
the change was a few lines. In fact I can go back and forth with an
#ifdef and the total number of changed lines is tiny.
> 3) It writes files. They save their "preference" information and frankly, I
> want to be in control of what disk access takes place.
This is only if you want persistent wisdom. I'm pretty sure it doesn't
write files unless you instruct it to.
But, as always YMMV, so use whatever is most appropriate for your task.
Cheers,
-- Paul
--
Dr. Paul S. Ray WWW : http://xweb.nrl.navy.mil/personnel/paulr/
NRL, Code 7655 E-mail: email@hidden
Washington, DC 20375 Phone : (202) 404-1619
_______________________________________________
scitech mailing list | email@hidden
Help/Unsubscribe/Archives: http://www.lists.apple.com/mailman/listinfo/scitech
Do not post admin requests to the list. They will be ignored.
