[mpich-discuss] abusing MPI_2DOUBLE_PRECISION to fake MPI_DOUBLE_LONG

Jeff Hammond jeff.science at gmail.com
Tue Aug 13 11:07:10 CDT 2013


Does MPICH preserve the ability of doubles to hold integers up to 2^53
properly when doing MPI_{MAX,MIN}LOC over Fortran
MPI_2DOUBLE_PRECISION?

I see no evidence in
http://www.mpi-forum.org/docs/mpi-11-html/node79.html that this isn't
possible, but I guess implementations might fall back to the C
implementation for MPI_DOUBLE_INT under the assumption that the
Fortran programmer only needs half of the bits in the second double of
MPI_2DOUBLE_PRECISION.

The motivation is that, until
https://svn.mpi-forum.org/trac/mpi-forum-web/ticket/319 is part of the
standard, there's no way to have loc be more than a 32b int, at least
from C.  This hideous hack via Fortran seems like the only viable
option for the time being.

Thanks,

Jeff

-- 
Jeff Hammond
jeff.science at gmail.com



More information about the discuss mailing list