Open MPI logo

Open MPI User's Mailing List Archives

  |   Home   |   Support   |   FAQ   |   all Open MPI User's mailing list

Subject: [OMPI users] Open MPI 1.3 segfault on amd64 with Rmpi
From: Dirk Eddelbuettel (edd_at_[hidden])
Date: 2009-01-26 15:33:48


I am chasing a segfault when I use Open MPI (1.3) with Rmpi (0.5.6), the MPI
add-on package for R that is written and maintained btyby Prof Hao Yu (CC'ed)

I should prefix that the code runs just fine on 32bit Debian system at home.
However, on amd64 running Ubuntu 8.10, I am seeing segfaults upon
initialisation. I use the same R and Open MPI packages on both systems,
suitably recompiled. One of the bigger toolkit difference is the 1.5.26
version of libtool on Debian versus 2.2.4 on Ubuntu.

Gdb doesn't want to step into the Open MPI code; I used debugging symbols for
both R and Open MPI that are available via -dbg packages with the debugging
info. So descending one function at a time, I see the following calling
sequence

  MPI_Init
  ompi_mpi_init
  orte_init
  opal_init
  opal_paffinity_base_open
  mca_base_components_open
  open_components

where things end in the loop over oapl_list() elements. I still see a
fprintf() statment just before

   if (MCA_SUCCESS == component->mca_register_component_params()) {
  
in the middle of the open_components function in the file
mca_base_components_open.c

Does this make any sense? I was at first worried that the dynamic loading
faild -- yet this does not seem to be the case as the mpi, open-rte and
open-pal libraries are loaded and I also see code from some of the modules
being executed. I somehow fear that something is colliding with GNU R, but
despite some familiarity with R I have to admit that I do not know here this
could come from.

Any pointers would be appreciated.

Regards, Dirk

-- 
Three out of two people have difficulties with fractions.