Compiler choice

[Axel]

ramon_garcia schrieb:

hi ramon,

> Hello, I am interested in running cp2k with hybrid density functional
> (B3LYP, PBE0). It seems that CP2k supports them, but it is sensitive

this is an experimental feature of the code,
so don't expect it to be simple to get right.

> to compiler choice. I have read:
>
> From http://groups.google.com/group/cp2k/browse_thread/thread/6c3db85a1990bab8/84dccce5b47d5016?lnk=gst&q=b3lyp#84dccce5b47d5016
>
> > As with all calculations including Hartree-Fock exchange you need an
> > up-to-date g95 compiler together with the libint-library in order to
> > run cp2k.
>
> Does this mean that I should refuse to use Intel or Portland
> compilers? Is the GFortran compiler a valid choice?

to the best of my knowledge, gfortran is a bit lagging behind
in implementing the required standards in a fully compliant way,
and i'm not sure whether even the latest development versions
are able to compile the non-hybrid code. the last time i tried with
a distribution supported version it failed.

intel compilers do in general a decent job, although one has to
pay attention to the patch levels and should update. from reading
the source code it seems that the very latest version might work.
originally, only recent g95 compilers were able to compile the hybrid
code. for the PGI compilers there are some hacks in the code, but
in general i would recommend against using them unless you have
no alternative. as with intel, it is imperative to have a very recent
version with all updates and/or try several patchlevels in case of
problems.

in summary, it is probably best to start with g95 and compile a
version
without hybrid functionals. get used to the input format and how to
run cp2k with quickstep and only once you are confident (may take a
while the learning curve is a bit steep at times) with that, i'd move
on to compile a version with libint. i remember that there were also
some problems with certain versions of libint...

as written in the sources of the good old xterm:
beware, here be dragons. ;-)

cheers,
   axel.



>
> Ramon Garcia Fernandez
> Ph D student
> http://cacharro.quim.ucm.es
> Universidad Complutense de Madrid