MPI2 problem on IBM Power PC

[Pietro Vidossich]

Dear all,

 we have problems running CP2K on the IBM Power PC cluster at the Barcelona
Supercomputing center. The error we receive is the following:

libdbcsr|  MPI error 1641487 in mpi_alltoall @ mp_alltoall_i44 : Other MPI
error, error stack:
MPI_Alltoall(711).................: MPI_Alltoall(sbuf=0x3c3cde10, scount=2,
MPI_INTEGER, rbuf=0x3c3cd3c0, rcount=2, MPI_INTEGER, MPI_COMM_WORLD) failed  
MPIR_Alltoall(175)................:    
MPI_Type_create_indexed_block(166): MPI_Type_create_indexed_block(count=32,
blocklength=2, array_of_displacements=0x3c3c6670, MPI_INTEGER,
newtype=0xffffff29628) failed

and may happen after any number of steps.
Unfortunately, the support of BSC has not been able to solve the problem (as
you can see from the message below). If you run the latest code on Marenostrum
or you have any suggestion on how to make the code working on Marenostrum, we
would really appreciate your help.

 kind regadrs, 
pietro vidossich




---------- Forwarded Message -----------
From: Xavi via RT <sup... at bsc.es>
To: vi... at klingon.uab.es
Sent: Mon, 21 Jun 2010 08:06:05 +0200
Subject: [rt.bsc.es #46701] RES - New Activity Report

Hi Pietro,

Unfortunately we have not any good news on this. We have tried with
latest versions of the code, multiple compilation techniques and even
different mpi implementations, but with no luck. It seems there is a
memory problem that makes the program crash.

I can try again with the very latest version to see if the problem is
already solved. I see in the changelog that a memory leak has been fixed
recently

I let you know if i can get a working installation.

Regards,

On Dom. Jun. 20 10:03:23 2010, vi... at klingon.uab.es wrote:
> New report for ActivityBCV-2010-1-0023: Unraveling the reaction and
>    drug activation mechanisms in catalase-peroxidases by means of
>    QM/MM Car-Parrinello simulations:
> UnixGroup: ub07
> Site: MareNostrum
> Date: 2010-06-20
> Sent by: Pietro Vidossich
> Report type: Petition
> Text:
> Hello, as you may remember from a previous report, the current version
>    of CP2K is not stable on Marenostrum because of some problem with
>    MPI2. so far we were using an older version of the code which was
>    fine, but now we discovered that this version has a bug which does
>    not allow us to perform certain calculations. The bug has now been
>    fixed by the cp2k developers, obviously in the latest code. Thus,
>    we would really need to run with the new code. Have you solved the
>    problems with MPI2? On May 3 you told me you were working on it,
>    but i didn't hear from you since then. Please let us know.
>    kind regards, pietro
>

-- 
Xavi Abell�n (xavi.a... at bsc.es)
BSC-CNS
C/ Jordi Girona 31              WWW: http://www.bsc.es
Barcelona (Spain)               Tel: +34-93-405 42 28 
e-mail: sup... at bsc.es          Fax: +34-93-413 77 21
------- End of Forwarded Message -------