CP2K build with Intel Compiler and MKL

[Axel Kohlmeyer]

On Sep 25, 11:21 am, Matthias <matthia... at psi.ch> wrote:
> Hi Ondrej,
>
> I have just experienced the same behavior using the Intel compiler on
> a ia64 system (Altix 4700, Itanium2). I tried several ifort compiler
> releases, but none worked properly for all tests (many NaN, seg.
> faults etc.). Finally, I gave up with the Intel compiler, even if it
> would be obviously the first-choice compiler especially for an Itanium
> system and switched to gfortran 4.4.1 which works even fine with the
> pre-installed MKL. Thus the performance degradation is not that
> serious, but still the price to pay for working CP2K executable.

one comment on this issue. for the example mentioned in the
discussion,
it looks to me, by comparing the output from different compilations,
as
if the regtest is done in a way that at some point there is a
numerical
instability in the code, and that for some families of compilers, the
result flips one way (where the run finishes), and for others the
other
way (where it crashes). so i would argue that it is not always
justified
to blame the compiler outright, even if there are some compilers that
get it "right". apart from the fact that the input may cause these
instabilities
due to the nature of the underlying algorithms, one also has to
consider
to consider that these kinds of instabilities can be reduced by
changing
the code to be more careful, and that for package software as large
and
as complex as cp2k, it is almost bound to happen in some areas of the
code, due to the different ways how different people write code.

that being said, i have also made the experience that particularly on
itanium processors, the intel compilers are much more unstable than
elsewhere. the good news is, that it looks as if the itanium is going
the way of the dodo and that it soon will not be worth worrying about
it.

cheers,
   axel.

> However, if somebody knows about a working release/flag combination
> for the Intel compiler, I would also appreciate this information.
>
> Matthias
>
> On Sep 25, 3:38 pm, Ondrej Marsalek <ondrej.... at gmail.com> wrote:
>
> > And for added fun, even with almost the simple arch file that I can think of:
>
> >http://paste.pocoo.org/show/141444/
>
> > I get a failure on the same test. It looks a little bit different, but
> > still has NaNs for densities. If there is anyone out there who uses
> > this Intel compiler and MKL and his build can pass all the tests, I
> > would really like to head from you.
>
> > I don't think this is relevant, but to provide some extra information,
> > just in case. This is on Ubuntu 9.04 and a Core2 E8400 CPU.
>
> > Ondrej
>
> > On Fri, Sep 25, 2009 at 15:24, Ondrej Marsalek
>
> > <ondrej.... at gmail.com> wrote:
> > > I have installed the Intel compiler version 11.0.083 (and the included
> > > MKL), which is supposed to work based on what Axel said here on the
> > > list as well as this page:
>
> > >http://software.intel.com/en-us/articles/build-cp2k-using-intel-fortr...
>
> > > However, with the arch file pasted here (optimization switched off
> > > intentionally):
>
> > >http://paste.pocoo.org/show/141441/
>
> > > I still get a wrong build, that, for example, produces this output on
> > > the test si8_pmix_otdiag.inp:
>
> > >http://paste.pocoo.org/show/141440/
>
> > > I would appreciate any suggestions on how to make this work. Regards,
> > > Ondrej
>
> > > On Wed, Sep 23, 2009 at 19:23, Ondrej Marsalek
> > > <ondrej.... at gmail.com> wrote:
> > >> Dear all,
>
> > >> I have found that my build of CP2K with the Intel Fortran Compiler
> > >> 10.1.018 and MKL 10.0.4.023 has trouble passing the tests (NaNs appear
> > >> and so on). Apparently, it is miscompiled. The same setup has worked
> > >> (passed all the tests) about a year. I have only updated CP2K from CVS
> > >> and used a recent OpenMPI (1.3.3). I have intentionally kept the
> > >> versions the same as before, to have something that is known to work.
> > >> I have two questions related to that.
>
> > >> Has anyone else experienced a similar problem?
>
> > >> Does anyone have a tested combination of recent Intel compiler and MKL
> > >> versions, perhaps together with an arch file that I could use as
> > >> inspiration?
>
> > >> Thanks for any replies,
> > >> Ondrej