[Dune] [ISTL]: Is (HAVE_MPI && !HAVE_PARMETIS) a viable configuration?
Bård Skaflestad
Bard.Skaflestad at sintef.no
Sat Nov 20 11:44:50 CET 2010
Dear Dune,
Revision 1365 of dune-istl introduced a new function, commGraphRepartition(), into dune/ist/repartition.hh . As far as I can tell, this function is viable only when one has ParMETIS support (as indicated by the HAVE_PARMETIS preprocessor symbol set during the configuration process). In particular, commGraphRepartition() uses the ParMETIS 'idxtype' and directly invokes the ParMETIS function 'ParMETIS_V3_PartKway()'. Moreover, unless I'm misreading the code, 'commGraphRepartition' is protected by HAVE_MPI but makes no direct mention of HAVE_PARMETIS.
At the moment I don't have ParMETIS installed on my system, but I do have MPI. Effectively, ISTL revision 1365 has broken the build (of at least dune-pdelab-howto) for me. On the other hand, ISTL (or possibly PAAMG) has for a long time "strongly recommended" using ISTL with ParMETIS in parallel configurations.
I guess my question really boils down to whether or not (HAVE_MPI && !HAVE_PARMETIS) is expected to work (or at least build) in recent editions of the ISTL. If not, I'll update my local installation.
Sincerely,
--
Bård Skaflestad <Bard.Skaflestad at sintef.no>
SINTEF ICT, Applied mathematics
More information about the Dune
mailing list