Re: [AMBER-Developers] nofftw3 and MKL

From: Jason Swails <jason.swails.gmail.com>
Date: Tue, 14 Oct 2014 09:07:49 -0400

On Tue, Oct 14, 2014 at 7:55 AM, David A Case <case.biomaps.rutgers.edu>
wrote:

> On Tue, Oct 14, 2014, Ross Walker wrote:
>
> > I don't think anyone has taken the time to
> > update the interface to FFTW v3
>
> This doesn't appear(?) to be the case: fft1d.F90 has an "include
> <fftw3.f>"
> line.
>
>
> > Perri put in MKL's FFT for the CPU based hybrid precision model
> > but I don't think that was added for double precision in the trunk -
> Perri?
> >
> > If someone would like to volunteer to update PMEMD to support FFTW 3 and
> > test it (including performance) it would be much appreciated.
>
> pmemd (double precision) with FFTW3 seems to work fine for me (compiles,
> passes the test suite). I'll try to see if I can get some performance
> numbers.
>

​FWIW, this works for me as well. In fact, this was the necessary
workaround on cray machines to circumvent compiler optimization bugs with
PUBFFT.

The MKL-optimized FFTW implementation was added using this code path, to my
knowledge.

As for the "why don't we use FFTW for pmemd"... I think the answer is that
we can't, right? FFTW is GPL and pmemd is not, correct? Users are free to
link it themselves, but we can't do it automatically for them, I don't
think.

All the best,
Jason​

-- 
Jason M. Swails
BioMaPS,
Rutgers University
Postdoctoral Researcher
_______________________________________________
AMBER-Developers mailing list
AMBER-Developers.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber-developers
Received on Tue Oct 14 2014 - 06:30:02 PDT
Custom Search