[Pw_forum] Compilation of espresso-4.0.4
Axel Kohlmeyer
akohlmey at cmm.chem.upenn.edu
Fri Mar 27 18:01:20 CET 2009
On Fri, 2009-03-27 at 09:21 -0700, Amos Leffler wrote:
> Hi Forum,
> I have been trying to compile espresso-4.0.4 using
> LAM/MPI-7.1.4. The latter was correctly installed as noted by the
> program itself. I was able to configure the espresso program in
amos,
that doesn't mean anything. can you compile a simple MPI program
written in fortran with it? for example the 'pi' example referred
to at one of the LAM/MPI tutorial.
> parallel mode as noted by the output of the configure file. However,
configure only tests whether a parallel compiler wrapper is
installed and whether it can compile, link and run a simple,
non-parallel executable. it assumes that it is installed
correctly for parallel codes if it works on a minimal serial
test. otherwise, you have much bigger problems than getting
espresso compiled.
> when I tried to use the "make all" command it started to compile but
> stopped in the file "fft_paralled.f90" on line 234 with the message
> "mpi_double_complex has no implicit type". I have run into the
> same error with OpenMpi-1.2.9 and Axel K. thought that it was
> due to problems with the OpenMpi. It should be noted that espresso
as i was already noting in my previous response, this is
not a problem of OpenMPI by itself, but a problem of an
incorrect installation. so either you must have a messed
up the installation of openmpi, or you have an incompatible
or bogus mpif.h file somewhere where it is picked up by the
mpi compiler wrapper. Both, OpenMPI and LAM/MPI work
fine with espresso, and since you get the same error message
with both packages, the latter seems more likely.
> does compile properly in single mode.
that only confirms that the problem is the MPI installation.
> Hopefully there is a simple work-around.
the simplest workaround is always to hire somebody, who knows
about these things. the symptoms you describe are very obvious
to check and sort out, so there should be enough people somewhere
close to where you are located to have a look at your machine(s),
detect what is messed up and sort it out.
as discussed many times with people having compilation problems,
these are very difficult to sort out remotely, particularly if
there are indications of an inconsistent installation.
it is obviously not a defect of quantum espresso by itself, or
else you would not be the only person reporting it. there are
a lot of people using either openmpi or lam/mpi.
depending on how large your deployment is, you could also consider
wiping the machine and installing a clean distribution. e.g. fedora has
been working for me out of the box (it has suitable packages for
openmpi, gfortran, atlas as part of the distribution) since version 6,
i.e. since two years ago. i'd assume that others like debian 5, ubuntu
08XX or RHEL/CentOS 5.x work just as well.
it only gets nasty, if you have to include support for a specific
high-speed interconnect or include program launching through a
specific batch system, both of which would require a custom compile.
adding support for non-native compilers to openmpi can be done without
a recompile, all you need is a suitable symlink to opal_wrapper (or
opal_wrapper-64, or /etc/alternatives/mpicc in some distros) and
a matching
>
> Amos Leffler
please remember the policy of this forum
to include your affiliation. thanks.
axel.
> _______________________________________________
> Pw_forum mailing list
> Pw_forum at pwscf.org
> http://www.democritos.it/mailman/listinfo/pw_forum
--
=======================================================================
Axel Kohlmeyer akohlmey at cmm.chem.upenn.edu http://www.cmm.upenn.edu
Center for Molecular Modeling -- University of Pennsylvania
Department of Chemistry, 231 S.34th Street, Philadelphia, PA 19104-6323
tel: 1-215-898-1582, fax: 1-215-573-6233, office-tel: 1-215-898-5425
=======================================================================
If you make something idiot-proof, the universe creates a better idiot.
More information about the Pw_forum
mailing list