Open MPI logo

Open MPI User's Mailing List Archives

  |   Home   |   Support   |   FAQ   |  

This web mail archive is frozen.

This page is part of a frozen web archive of this mailing list.

You can still navigate around this archive, but know that no new mails have been added to it since July of 2016.

Click here to be taken to the new web archives of this list; it includes all the mails that are in this frozen archive plus all new mails that have been sent to the list since it was migrated to the new archives.

Subject: [OMPI users] Intel MKL PARDISO seg. faults wrapping it around MPI_INIT/MPI_FINALIZE
From: Alberto F. Martín-Huertas (amartin_at_[hidden])
Date: 2013-03-13 06:58:58

 Dear MPI users,

 I have a small serial program that reads a sparse matrix from a file
 and factorizes it
 using the sparse direct solvers included in PARDISO MKL. When I execute
 the program with
 the matrix of interest, PARDISO factorizes it succesfully. No SIGSEV is
 produced. However,
 if I transform this small sequential program to a message-passing
 program (i.e., wrapping its codes
 around MPI_INIT/MPI_FINALIZE and executing it using mpirun -np 1
 PROGRAM) then a SIGSEV appears. Therefore, the problem seems
 to be related to the combination of PARDISO MKL and the message-passing
 environment (OpenMPI).

 Some of the characteritics of the environment are:

    - MPI: OpenMPI 1.5.4 (find attached the output from ompi_info --all)
    - fortran compiler: ifort version 13.0.1
    - Linux Kernel: Linux version 3.0.13-0.27-default (geeko_at_buildhost)
                    (gcc version 4.3.4 [gcc-4_3-branch revision 152973]
 (SUSE Linux) ) #1 SMP Wed Feb 15 13:33:49 UTC 2012 (d73692b)
    - MKL libraries: MKL/13.0.1
    - CPU: Intel(R) Xeon(R) CPU E5-2670 0 @ 2.60GHz

 I have also reported this issue on the Intel MKL forum,
 and provided a reproducer of the issue on:

 The "weird" observation is that I could reproduce
 the issue on three machines based on the Intel SandyBridge-based
 multicore-CPUs, while I could not on a pair of non SandyBridge-based
 Intel archictures, using exactly the same software
 stack. I am afraid that Intel SandyBridge-based CPUs have any
 particularity with respect to virtual memory handling that
 cause PARDISO MKL to SIGSEV once MPI_INIT has been called.

 Could u please help me to find the root-cause of this issue?

 Thanks in advance.
 Best regards,

 Alberto F. Martín-Huertas
 Centre Internacional de Mètodes Numèrics a l'Enginyeria (CIMNE)
 Parc Mediterrani de la Tecnologia, UPC
 Esteve Terradas 5, Building C3, Office 210,
 08860 Castelldefels (Barcelona, Spain)
 Tel.: (+34) 9341 34223
 e-mail: amartin_at_[hidden]
 All personal data contained on this mail will be processed 
 confidentially and registered in a file property of CIMNE in
 order to manage corporate communications. You may exercise the rights 
 of access, rectification, erasure and object by
 letter sent to Ed. C1 Campus Norte UPC. Gran Capitán s/n Barcelona.