[MOAB-dev] failed to run parallel test case: moab-5.1.0 with mpich-3.3

[Vijay S. Mahadevan]

Dear Lorenzo,

> I indeed had a system wide mpi installation which I promptly removed.
> Can you tell me how did you catch that in config.log, I'm not able to
> figure it out.

Lines 131/5246 and 239/5259 in your previous config.log shows mpicc
and mpicxx from different locations. So I surmised potentially
multiple installations in the path, which got picked up automatically.
Usually, explicitly specifying the compilers would avoid such an
issue, even if you do have multiple wrappers in your path.

> Back to the issue, I removed the system wide mpi and also tried to
> specify the compilers as you suggested but got the same behavior.
> Please find in attachment the new config.log
> As I mentioned in the first email I encountered the exact same problem
> on a AMD epyc cluster, again with ubuntu as os.

The newer config.log shows that the wrappers are consistent now. So I
am still puzzled why `make check` did not actually run the tests.

Can you do the following two variations:

1) Re-verifying your current approach
  a) Go to test/parallel and do `make clean && make check | tee make_check.log`
  b) Send us make_check.log so that we can see if things compile but
do not run or nothing actually happens there.

2) Retry out-of-source
  a) Perform `make distclean` in your in-source build
  b) Do: `mkdir build && ../configure <configure arguments> && make
all && make check`

Let me know if (1) yields something useful and if not, (2) should
resolve the issue. If (2) does work, then we may have a problem with
in-source builds and I can check that on a workstation to replicate
the problem. Hope my instructions above aren't confusing.

Best,
Vijay

On Sun, Jun 2, 2019 at 10:47 AM Lorenzo Botti <bottilorenzo at gmail.com> wrote:
>
> Dear Vijay,
> I indeed had a system wide mpi installation which I promptly removed.
> Can you tell me how did you catch that in config.log, I'm not able to
> figure it out.
>
> Back to the issue, I removed the system wide mpi and also tried to
> specify the compilers as you suggested but got the same behavior.
> Please find in attachment the new config.log
> As I mentioned in the first email I encountered the exact same problem
> on a AMD epyc cluster, again with ubuntu as os.
>
> Bests
> Lorenzo
>
>
>
> Il giorno ven 31 mag 2019 alle ore 16:42 Vijay S. Mahadevan
> <vijay.m at gmail.com> ha scritto:
> >
> > Hi Lorenzo,
> >
> > Yes that certainly is strange. You do have these defined in src/MOABConfig.h.
> >
> > | #define HAVE_MPI 1
> > | #define HAVE_HDF5_H 1
> > | #define HAVE_HDF5 1
> > | #define HAVE_HDF5_PARALLEL 1
> >
> > So configuration seems to have picked up a stable combination.
> >
> > One thing that stands out from config.log is that mpicc is not
> > prefixed with the right MPI path. Do you have "mpicc" on your path
> > pointing to a different installation ? Can you retry configuration by
> > explicitly setting the compilers like below ? Let us know if that
> > resolves the issue.
> >
> > CC=/home/ihabia/src2/mpich-3.3/mpich-f/bin/mpicc
> > CXX=/home/ihabia/src2/mpich-3.3/mpich-f/bin/mpicxx
> > FC=/home/ihabia/src2/mpich-3.3/mpich-f/bin/mpif90
> > F77=/home/ihabia/src2/mpich-3.3/mpich-f/bin/mpif77 ./configure
> > --enable-optimize --disable-imesh
> > --with-mpi=/home/ihabia/src2/mpich-3.3/mpich-f
> > --with-hdf5=/home/ihabia/src2/hdf5-1.10.5/hdf5
> > --prefix=/home/ihabia/src2/moab-5.1.0/moab
> >
> > Best,
> > Vijay
> >
> > On Fri, May 31, 2019 at 10:33 AM Lorenzo Botti via moab-dev
> > <moab-dev at mcs.anl.gov> wrote:
> > >
> > > Dear MOAB developers,
> > > I cannot build the parallel version of moab with mpich-3.3.
> > > I tried on two different architectures with same result, parallel
> > > examples hangs without any error message, see below
> > > make  check-TESTS
> > > make[3]: Entering directory '/home/ihabia/src2/moab-5.1.0/test/parallel'
> > > make[4]: Entering directory '/home/ihabia/src2/moab-5.1.0/test/parallel'
> > >
> > > Please find in attachment the config.log
> > >
> > > One option is to build with openmpi but I cannot understand why mpich
> > > does not work.
> > > Best regards
> > > Lorenzo