On Thu, 12 Jun 2008, Edward Chapin wrote:
>> > FFTW has the option to store the real and imaginary parts in
>> > separate real-valued arrays. However, the default (which I'm
>> > currently using) is to define arrays of type
>> >
>> > typedef double fftw_complex[2];
>> >
>> > It looks like much of starlink knows what to do with complex numbers
>> > because they are a native fortran type. Does the C interface support
>> > some implementation of complex numbers? (a cursory glance at the C
>> > interface documentation for NDF leads me to suspect that it
>> > doesn't).
>>
>> Is it not just a case of calling ndfMapz rather than ndfMap?
>
> Yes, indeed it is. I guess this mostly answers my question: it looks
> like complex numbers (not the hermetian representation noted above) are
> generally implemented in an NDF as two distinct continuous arrays, for
> real and imaginary parts respectively. It seems like this is the
> preferable method of storage given the 1-d transforms that I'm doing.
>
> I can easily switch to asking FFTW to use two distinct arrays for real and
> imaginary parts (rather than having the real/imaginary parts interleaved as I
> do currently), and use ndfMapz.
> It looks like this will produce an output that will be easy to visualize
> using Kappa routines.
David,
since, as far as I know, KAPPA doesn't support COMPLEX data types, that
last statement isn't true, is it?
Peter.
|