On Thursday, August 21, 2003, at 11:57 AM, Ed Wesley Wells wrote:
> Now, whenever I try to start my mpi application (remember, it linked
> and
> compiled fine) I get an error telling me:
>
> ======================
> Thank you,It seems that [at least] one of the processes that was
> started with
> mpirun did not invoke MPI_INIT before quitting (it is possible that
> more than one process did not invoke MPI_INIT -- mpirun was only
> notified of the first one, which was on node n0).
>
> mpirun can *only* be used with MPI programs (i.e., programs that
> invoke MPI_INIT and MPI_FINALIZE). You can use the "lamexec" program
> to run non-MPI programs over the lambooted nodes.
> ======================
There are a number of reasons this could be happening. I would guess
that something is causing your application to abort before MPI_INIT
occurs. For various reasons, LAM can only give the "your application
died due to signal XX" we are all way too familiar with if your
application made it through MPI_INIT. Before MPI_INIT, you always see
the above error, regardless of *why* your application died.
stdout and stderr forwarding are initialized as soon as your
application is started, so printf() will always display back to the
terminal running mpirun. I would add some printf()s around MPI_INIT to
make sure that you aren't reaching MPI_INIT. Then run your application
as a singleton process (while the LAM environment is booted), meaning
just start it directly from the shell:
% ./foo <your app's args>
Only one process will exist in MPI_COMM_WORLD, but who cares - you
aren't getting to the point where communication can occur. With this,
the shell will report why the application died, or you can start the
app under gdb. What is going wrong should be more obvious with this
information, so it should help us track the problem down.
Brian
--
Brian Barrett
LAM/MPI developer and all around nice guy
Have a LAM/MPI day: http://www.lam-mpi.org/
|
