MPI-related fatal error when importing Dolfin

Issue #107 invalid
Timothée Lecomte created an issue

Dear all,

I am using Fenics on MacOS X (Snow Leopard). I used to have it working fine until recently when I updated some packages from Homebrew (a package manager for MacOS, similar to MacPorts). I then rebuilt Fenics using Dorsal, and I get the following error :

$ python
Python 2.7.5 (default, Aug 28 2013, 10:09:22) 
[GCC 4.2.1 (Apple Inc. build 5666) (dot 3)] on darwin
Type "help", "copyright", "credits" or "license" for more information.
>>> from dolfin import *
Attempting to use an MPI routine before initializing MPICH
<python exits>

I have the following packages built with Homebrew: MPICH 3.0.4 Boost 1.54 (with MPI support) CGAL 4.2 Python 2.7.5 Swig 2.0.10 Parmetis 4.0.2 Vtk 5.10.1

In Dorsal, I use the following:

PACKAGES=(
petschypre
slepc
scotch
fiat
ufc
ufl
ffc
viper
instant
dolfin
)

When I remove PETSc from Dorsal (petschypre is just PETSc with the Hypre preconditioners enabled) (and slepc since it depends on PETSc), the above error does not appear and Dolfin works fine. Of course, I then miss the PETSc assemblers and solvers, and I am only left with uBLAS.

When building dolfin, the following message sums up the packages that are found on my machine:

-- The following optional packages were found:
-- -------------------------------------------
-- (OK) OPENMP
-- (OK) MPI
-- (OK) PETSC
-- (OK) SLEPC
-- (OK) UMFPACK
-- (OK) CHOLMOD
-- (OK) SCOTCH
-- (OK) PARMETIS
-- (OK) CGAL
-- (OK) ZLIB
-- (OK) PYTHON
-- (OK) SPHINX
-- (OK) VTK
-- (OK) QT
-- 
-- The following optional packages were not be found:
-- --------------------------------------------------
-- (**) TRILINOS
-- (**) PASTIX
-- (**) HDF5

I suspect there is something wrong with MPICH3 but I tried to revert to MPICH2 1.5 without success.

Can you please help solve this issue ?

P.S. : I am not using the binaries for MacOS offered on the website for three reasons: 1) I have found that the plotting is buggy compared to what I get when I build the stack myself, 2) I have Numpy, Scipy and friends built for the python interpreter from Homebrew, 3) I need the AMG preconditioners provided by Hypre (for fluid mechanics problems).

Comments (1)

  1. Log in to comment