[Wien] MKL error: Cannot load symbol MKLMPI_Get_wrappers

[Peter Blaha]

Don't know if this is the problem here, but very often the problem is 
the blacks-library. There are several versions of blacks in the mkl and 
they must be chosen according to your mpi. Of course there is a version 
for intel-mpi, but it must be the correct library for each mpi.


Am 27.04.2016 um 16:42 schrieb Walid Hetaba:
> Dear colleagues,
>
> I'm trying to get the parallel version of WIEN2k 14.2 running on the
> institute's cluster, but I'm stuck now.
> I searched the mailing list for the error message I receive, but at
> first glance I did not find relevant postings.
>
> This is the problem I face:
> The serial version of WIEN2k is running fine (also when using qsub and
> the submission system).
> However, when trying to use the parallel version I get the following
> error when lapw1 is invoked in parallel mode:
> "Intel MKL FATAL ERROR: Cannot load symbol MKLMPI_Get_wrappers."
>
> Here the facts on hard- and software:
> I'm using WIEN2k 14.2 to run the example TiC case from the UG as a test
> to see whether all works out fine.
> I compiled the sofware using the Intel Cluster Studio with the Intel
> Fortran Compiler 16.0, Intel C++ Compiler 16.0, Intel MKL 11.3 and Intel
> MPI.
> I compiled it using FFTW3 and the MKL lapack, scalapack and blacs.
>
> The cluster consists of 37 nodes with 20 CPUs each. It is running a
> CentOS Linux distribution and it is using SGE as submission system. In
> the submission script orte is used as the parallel environment.
>
> This is what I've tried so far:
> I don't know why it can't find things from the MKL as I export all my
> environmental variables using the submission script, which I also
> checked by printing them.
> As additional information I include the submission script
> (WIEN2k-test-p.sh, which is still a bit messy as I adopted it from
> another program for testing reasons), the :parallel file with the error
> messages (which I renamed to .parallel so that Thunderbird on Windows
> can handle it), the output-file (TiC.o14540) and the .machines file.
>
> I get the same error messages when using OpenMPI instead of the
> IntelMPI, but the output is a bit different (see TiC.o14541).
>
> I did try different ways to export my environmental variables (to get
> rid of unnecessary stuff), also tried mpi as the parallel environment in
> the submission script instead of orte. I also tried to request a
> different number of slots,...
> But nothing off this got me rid of this error message.
>
> Has anybody experienced a similar problem or are there any hints what
> else I can check?
>
> Thanks for your help.
> All the best,
> Walid
>
> --
> Dr.techn. Walid Hetaba
>
> Fritz-Haber-Institut der Max-Planck-Gesellschaft
> Department of Inorganic Chemistry
> Faradayweg 4-6, 14195 Berlin, Germany
> T: +49 30 8413-4412
> hetaba at fhi-berlin.mpg.de
> http://www.fhi-berlin.mpg.de
>
>
>
> _______________________________________________
> Wien mailing list
> Wien at zeus.theochem.tuwien.ac.at
> http://zeus.theochem.tuwien.ac.at/mailman/listinfo/wien
> SEARCH the MAILING-LIST at:  http://www.mail-archive.com/wien@zeus.theochem.tuwien.ac.at/index.html
>

-- 
--------------------------------------------------------------------------
Peter BLAHA, Inst.f. Materials Chemistry, TU Vienna, A-1060 Vienna
Phone: +43-1-58801-165300             FAX: +43-1-58801-165982
Email: blaha at theochem.tuwien.ac.at    WIEN2k: http://www.wien2k.at
WWW:   http://www.imc.tuwien.ac.at/staff/tc_group_e.php
--------------------------------------------------------------------------