[Users] compilation on marenostrum 4

[Erik Schnetter]

Helvi

The preferred way would be to create a Simfactory entry. The "unset MPI"
command would also go there.

If you are unfamiliar with Simfactory, then you would open an issue ticket
on the ET web site and attach the option list.

-erik


On Thu, Oct 19, 2017 at 8:26 AM, hwitek <hwitek at ffn.ub.es> wrote:

> Hi everyone,
>
> we could compile the toolkit on MareNostrum4 without any further
> problems ("unset MPI" did the trick), and I want to run some further
> tests before providing the configuration script.
>
> What is the standard way to include new configuration files into the
> toolkit? Should I send it to someone in particular?
>
> cheers,
> Helvi
>
> On 2017-10-18 19:16, hwitek wrote:
> > Hi Miguel, everyone,
> >
> > regarding your first point: I had a similar problem on its
> > predecessor,
> > MareNostrum 3.
> > The trick was to unset the global MPI variable, so in my bashrc I had
> > something like
> >
> > module load intel
> > ...
> > unset MPI
> >
> > I hope this helps! I'll also have a go at the compilation myself
> > tonight.
> >
> > cheers,
> > Helvi
> >
> >
> > On 2017-10-18 18:52, Miguel Zilhão wrote:
> >> hi all,
> >>
> >> i've been compiling the latest ET on marenostrum 4, and there were a
> >> couple of issues i'd like to ask:
> >>
> >> 1. on marenostrum, they bundle a lot of mpi-related tools on modules.
> >> in particular, one of the
> >> default modules that is loaded is called 'impi'. this module sets the
> >> global variable $MPI to 'impi', ie
> >>
> >>   $ echo $MPI
> >>     impi
> >>
> >> this seems to severely interfere with the Cactus building process. at
> >> the configuration stage i get
> >> the following:
> >>
> >> Configuring with flesh MPI
> >>    Warning: use of flesh MPI via MPI option is deprecated and should
> >> be replaced with the thorn
> >> ExternalLibraries/MPI and its MPI_DIR option
> >>    MPI selected, but no known MPI method - what is "impi" ?
> >>
> >> to workaround this problem, i've unloaded the module (module unload
> >> impi), which undefines the $MPI
> >> variable. after this, the configuration stage works fine.
> >> this seems to me like a bug, though. after all, a cluster-wide global
> >> variable called $MPI seems
> >> like a natural thing to exist in a lot of these machines... should
> >> the building of Cactus rely on
> >> the non-existence of such a variable?
> >>
> >> the other inconvenient thing is that, by unloading the impi module,
> >> one also removes from $PATH the
> >> mpirun command. so one has to unload the module for compiling the
> >> code, and then load it back to be
> >> able to run it.
> >>
> >> 2. i was not able to compile with any of the provided intel compilers
> >> (with gcc it worked fine). i
> >> know there are known issues with some versions of the intel
> >> compilers; is there some sort of list of
> >> intel compilers that are known to work? i could maybe try to ask the
> >> technical support if they could
> >> make those specific versions available...
> >>
> >> thanks,
> >> Miguel
> >> _______________________________________________
> >> Users mailing list
> >> Users at einsteintoolkit.org
> >> http://lists.einsteintoolkit.org/mailman/listinfo/users
>
> --
> ===========================
> Dr. Helvi Witek
> Marie-Curie Research Fellow
> Dep. Fisica Quantica i Astrofisica
> Universitat de Barcelona
> ===========================
> _______________________________________________
> Users mailing list
> Users at einsteintoolkit.org
> http://lists.einsteintoolkit.org/mailman/listinfo/users
>



-- 
Erik Schnetter <schnetter at cct.lsu.edu>
http://www.perimeterinstitute.ca/personal/eschnetter/
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.einsteintoolkit.org/pipermail/users/attachments/20171019/d9ea29a6/attachment-0001.html