|
|
Blitz Support : |
From: Andrius Kurtinaitis (andrius.kurtinaitis_at_[hidden])
Date: 2004-01-03 04:13:00
Hello,
Julian C. Cummings wrote:
> The point of my previous e-mail was simply to say that you can't have it
> both ways. You have to decide whether your best storage option is to have
> all components for an element stored together or each component stored
> separately in contiguous memory.
Yes, I understand it. I've given up with the interlaced arrays. I would
like to use the multicomponent array not for performance, but for
notation clarity, because all my data points and all the functions are
working with a three complex numbers.
I think first i'll try to use FFTW routines on the extracted single
components of the multicomponent array using the guru interface of the
FFTW. If it is not possible or too difficult, i'll copy the data.
I don't think I can write an universal FFTW wrapper for all kinds of
blitz array atorage schemes right now. Maybe someday in the future... ;-)
Thanks a lot.
Andrius
> P.S. There was an e-mail exchange recently about creating a Blitz wrapper
> for the FFTW package. This is a good idea in principle, but we would need a
> nice design to handle the various scenarios of multiple array dimensions,
> real vs. complex FFTs, in-place transforms vs. data copying, etc. If we had
> a set of FFT functions to invoke directly on a Blitz Array, this would
> address your problem with multicomponent Arrays, but the performance outcome
> might be worse than if you just copied each component into contiguous
> memory.
>
> Dr. Julian C. Cummings
> Staff Scientist, CACR/Caltech
> (626) 395-2543
> cummings_at_[hidden]