[mpich-discuss] MPI Implementation on 1 module

[Nicolas Rosner]

Hello Gulshan,

> We were expecting that MPI_FINALIZE command
> will stop all the nodes expect the master node

May I suggest that you re-read the documentation for MPI_FINALIZE? You
seem to be expecting more from the call than it actually does.


> The Call MPI_FINALIZE(err) does not stop the additional processors

Although we can't infer with any certainty what you precisely mean by
"stop them", it doesn't sound like it should at all. As the manpage
will confirm, you should keep in mind that

* the call is not intended to be called from one rank to stop all
others (rather, you should ensure that every rank calls it once it's
fully done)

* the call is not supposed to abort other pending calls, cancel any
nonblocking tasks or discard any potential in-flight messages for you
(rather, you should ensure that nothing is pending, at the app level,
before you call Finalize)


Best regards
Nicolás