[MPI3 Fortran] Deprecate mpif.h?
nmm1 at cam.ac.uk
Fri Mar 5 04:14:26 CST 2010
On Mar 5 2010, Supalov, Alexander wrote:
> Thanks. I think there's some difference between "deprecating" and
> "freezing". The former implies people will have to move to the mpi3
> module once. The latter means they can "just" sit tight and forget about
> MPI-3. Hence the higher presumed impact of the deprecation.
Yes, and that's why "deprecate" is right. Regrettably, for unavoidable
reasons, the current MPI interfaces break all of the Fortran standards
from 1966 onwards. The design minimised the places where they do so,
but some just could not be bypassed.
Worse, the problems that those breaches cause are getting worse, and
that's going to continue. The two main ways that I know of where that
is happening are inter-procedural optimisation (which conflicts with
the way choice arguments break the type rules) and asynchronicity
(which is increasingly caused by shared memory parallelism, of all
sorts, and also conflicts with inter-procedural optimisation).
I.e., in the long term, all programmers and all programs should convert
to MPI3. But, given that the TR which it will depend on is still under
development, "in the long term" means decades. But there really is no
alternative, because the language infrastructure on which the existing
MPI interfaces depend is getting increasingly unreliable.
More information about the mpiwg-fortran