Re: using MPI_Init() without mpirun

Al Geist (geist@msr.epm.ornl.gov)
Tue, 30 Apr 1996 19:55:33 -0400 (EDT)

>My main point was that MPI-2 should not specify behavior in the absence of
>mpirun since as far as MPI standards are concerned there is no mpirun.

I agree with your point Joel. I don't want to focus on mpirun (or SPAWN).
I don't care how an implementation decides inside MPI_INIT
how big to make COMM_WORLD. Maybe it checks an envar,
maybe the startup passes some "magic" args, maybe the MPP just
checks the partition size and says that is the how big it will be.

I want to define that if MPI_INIT is clueless to a choice to make for
COMM_WORLD it defaults to size=1 and continues rather than
aborting with an error.

My guess is the Paragon implementation of MPI_INIT is always clueful
and so you are unaffected by the proposal.

Al Geist