single precision compilation of cp2k

Axel akoh... at gmail.com
Sun Dec 2 22:13:58 UTC 2007


hi juerg,

On Nov 28, 5:26 am, Juerg Hutter <hut... at pci.uzh.ch> wrote:
> Hi
> It's almost 2 years since we tried this the last time.
> In priciple you only need the preprocessor flag __SGL.

this sounds simple enough. ;-)

> The problem you will face are the libraries;
> BLAS, LAPACK, SCALapack, MPI and the FFT libraries.

actually, there are a number of other small issues in
the code itself that have crept in. for example there are
several constants using 1.0D1 style syntax explicitly and
i even found an "implicit real*8". brrrrr.

even if nothing else comes out of it, trying a single prec.
compile could help to clean up the code (again). ;-)

> We had it working for non-parallel and the FFTSG library.

that would be our first target. the FFT would be replaced
with a machine specific version anyways (as well as BLAS).
attempting a parallel version would only make sense, if there
is a significant gain in the serial port.

> My guess would be
> BLAS, LAPACK  -> replace dxxxx by sxxx

if i'm reading the code right, f77_blas_generic should
take care of the blas part.

i'll keep you posted of the progress.

cheers,
    axel.

> ScaLapack  ???
> MPI should be ok
> FFT libraries; FFTSG should be ok, FFTW link single precision version
>
> Juerg
>
> ----------------------------------------------------------
> Juerg Hutter                   Phone : ++41 44 635 4491
> Physical Chemistry Institute   FAX   : ++41 44 635 6838
> University of Zurich           E-mail: hut... at pci.uzh.ch
> Winterthurerstrasse 190
> CH-8057 Zurich, Switzerland
> ----------------------------------------------------------
>
> On Tue, 27 Nov 2007, Axel wrote:
>
> > hi!
>
> > can anybody here give me a short summary of what would
> > be needed to compile cp2k in single precision.
>
> > thanks,
> >    axel.


More information about the CP2K-user mailing list