Thanks Freddie, that was helpful. The error was in my environment settings. I do in fact only have openmpi installed, but the environment is not set up exactly right, apparently.
when I execute PyFr with /opt/openmpi-1.8.3/bin/mpirun -np 1 pyfr run -b cuda it runs. I now have another question, but will open a new thread. V. On Wednesday, May 4, 2016 at 11:14:30 PM UTC+8, Freddie Witherden wrote: > > Hi Victor, > > On 04/05/2016 00:20, Victor Major wrote: > > I am trying to run the Euler vortex example, but I am stymied with the > > following error: > > > > ..//python3.4/site-packages/mpi4py/MPI.cpython-34m.so: undefined > symbol: > > ompi_mpi_op_no_op > > > > I am guessing that this has to do with OpenMPI version against which > > mpi4py configured itself. > > > > I am running 1.8.3, but can change this to anything actually as long as > > I know which version I need. Also... I am trying to use the CUDA backend > > and am not attempting to run a parallel job so why is OpenMPI being > invoked? > > PyFR should work with any recent version of MPI. However, it is crucial > that when you change your MPI library that you also recompile mpi4py. > My guess for the above is that mpi4py has been compiled against > something like OpenMPI while the current MPI is say mpich2. > > When PyFR starts up it has no idea if it is running as a parallel job or > not. The only way we can find this out is by initialising MPI and > seeing how many ranks there are. Hence, it is always necessary for PyFR > to initalise MPI. > > Regards, Freddie. > > -- You received this message because you are subscribed to the Google Groups "PyFR Mailing List" group. To unsubscribe from this group and stop receiving emails from it, send an email to [email protected]. To post to this group, send an email to [email protected]. Visit this group at https://groups.google.com/group/pyfrmailinglist. For more options, visit https://groups.google.com/d/optout.
