Also, note that fftw (http://www.fftw.org/faq/) has F77 bindings which
are easily callable from F90. It does require a c compiler for the
initial build, but typically we find a perfomance boost of 30%-50%
over over fftpack for our large 2-d FFTs (on Solaris and Intel Linux).
It also supports MPI out of the box (although we haven't tested
this yet ourselves).
Regards, Phil
Jörg Stiller writes:
> Thank you for all contributions to my question concerning FFT in
> Fortran 90/95.
>
> Here is my summary:
>
> (1) Tao Pang provides a compact complex FFT for N=2^M along
> with his book "An Introduction to Computational Physics.
> More information including the program source is found at:
> http://deimos.physics.unlv.edu/~pang/cp.html
>
> (2) Tom Criss sent a complex FFT for N=2^M. However, because
> of using declarations like integer*4, this code is not fully
> standard compliant.
>
> (3) Aleksandar Donev provided a link to Zdzisaw Meglicki's
> online textbook on Advanced Scientific Computing. The URL
> is http://noir.ovpit.indiana.edu/B673/ . The book provides
> a nice introduction into discrete FT and FFT along with
> several algorithms and corresponding F90 source code.
>
> (4) Simon Clarke provides F90 versions of fftpack/vfftpack
> including multible complex/real/sine/cosine transforms
> for arbitrarily sized N. The code is found at:
> http://www.maths.monash.edu.au/~src/archive/fftpk
> BTW: Wouldn't it be good to put this code to netlib too?
> So far, only a F77 version is provided there.
>
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|