The ABINIT Users Mailing List ( CLOSED )

[Xavier Gonze <gonze@pcpm.ucl.ac.be>]

Dear James and Steven,

As mentioned in the list of development efforts for v4.3, there
are several person working on the parallelisation of the FFT
inside ABINIT (this started already several months ago).
In v4.2, you find already the new set of FFT library routines,
written by S. Goedecker, that have shown excellent speed up
with up to 128 processors. v4.2 has also been prepared for
parallelisation, with a whole set of low-level routines
concentrating tasks to be executed in parallel, like
scalar products and other things that will need communication
between processors, see the directory Src_2spacepar.
To have ABINIT parallelized with correct
spread of memory usage accross processors is MUCH MORE than
using FFTW. You have a file that describe the present
status of FFT parallelisation in
~ABINIT/Infos/Notes_for_coding/FFT_in_parallel

If you are interested in coding within ABINIT, please,
- read the mails sent to the developer mailing list
- read the release_notes
- read the list of tasks ...
These are tools to coordinate the developers work.

Xavier


James Raynolds wrote:
> Steven,
>
>         Sounds good.  I'll get back to you.
>
> Jim
>
> -----Original Message-----
> From: Steven Homolya [mailto:Steven.Homolya@spme.monash.edu.au]
> Sent: Monday, October 20, 2003 10:34 PM
> To: 'forum@abinit.org'
> Cc: Lenore Mullin
> Subject: RE: [abinit-forum] FFTs in parallel
>
>
> On Mon, 20 Oct 2003, James Raynolds wrote:
>
>
> > Steven,
> >
> >         My colleague Lenore Mullin and myself at the University at Albany,
> > State University of New York, are working on FFT's and have OPENMP and MPI
> > versions of very efficient 1-d FFTS which we can provide you.  We are
>
> nearly
>
> > finished with a 3-d version which we believe to be the fastest available
> > anywhere.
>
>
> Thanks for the offer. Let me know when you have the multi-D version 
> done. Till then, I'll go with fftw, which I'm familiar with.
>
> Steve