Hi Jed, > Just out of curiosity: Does anybody have any estimates on the number > of MPI 1.x implementations in use out there? > > > Satish probably knows more than I, but MVAPICH1 had releases not that > long ago, and it's still quite common on clusters. Then there are > oddities like the scali MPI that we encountered recently. (They seem to > have copied an mpi.h from an MPI-2 implementation, then only implemented > part of it, which may be a strict subset of MPI-1.) I wish people would > upgrade so we wouldn't have to check for failure to implement a 15-year > old API, but I expect PETSc will be supporting MPI-1 for some years yet.
Okay, I see. Better not to mention the additional delay introduced by package repositories... Best regards, Karli
