cp2k on POWER4 running Linux

Axel akoh... at gmail.com
Fri Jun 27 22:45:14 CEST 2008



On Jun 27, 4:18 pm, Eric Shamay <eric.... at gmail.com> wrote:
> Matt,
>
> I've run the BLACS and SCALAPACK testers and they all compiled,
> completed, and passed. That tells me that at least the MPI setup works

are you sure you ran those in parallel?

i have yet to see a machine where _all_ BLACS and ScaLAPACK
testers work perfectly. in most cases the deviations could be
traced back to optimizations in the underlying BLAS/LAPACK libs
and supplying those without optimizations and in full floating
point conformance reduced the number of failed tests to a minimal
number and with very, very minor deviations.


> for the pre-done libraries. After that I tried to compile and run the
> following code:
>
> ***** hello++.cc *******
> #include "mpi.h"
> #include <iostream.h>
>
> int main(int *argc, char ***argv) {
>
>   MPI_Init(argc, argv);
>
>   int rank;
>   int size;
>
>   MPI_Comm_rank (MPI_COMM_WORLD, &rank);
>   MPI_Comm_size (MPI_COMM_WORLD, &size);
>
>   printf ("Hello World! I am %d of %d\n", rank, size);
>
>   MPI_Finalize();
>
>   return 0;}
>
> **********************************
>
> I compile using the following command:
>            mpicxx -I$(MPI)/include -o hello hello++.cc
>
> Compilation works fine without error or warning. Then I run the
> following command on the executable:
>            mpirun -np 4 ./hello
>
> And the resulting error is:
>
> rank 2 in job 9  p690.cs.uoregon.edu_40420   caused collective abort
> of all ranks
>   exit status of rank 2: killed by signal 11
> rank 0 in job 9  p690.cs.uoregon.edu_40420   caused collective abort
> of all ranks
>   exit status of rank 0: killed by signal 11
>
> I've determined that the program doesn't make it through the MPI_Init
> stage and something in there is breaking.


that could happen, if mpirun and mpicxx belong to
different MPI packages. are you certain that there is
not mixup of that kind.

cheers,
   axel.

>
> Thanks,
> ~Eric
>
> On Jun 26, 5:06 pm, Matt W <MattWa... at gmail.com> wrote:
>
> > > Any idea what that could mean? An issue with my mpi setup or with the
> > > way I've compiled cp2k? I'm totally lost at this point.
>
> > Hi Eric,
>
> > I guess you should check whether you can run simple mpi progs and then
> > the BLACS/SCALAPACK tests correctly on you system.
>
> > Matt


More information about the CP2K-user mailing list