lam-request_at_[hidden] wrote:
>> Can you run octave without the mpirun part? This usually means that
>> the command you are trying to run either can't properly execute on
>> the current architecture or is not found in your path.
>
> Hi, Mike,
> this is a log of how Brian's test works in my system:
>> [javier_at_xxx Hello]$ octave -q --eval Hello
>> [...]
>> Help on MPI: help mpi
>> Help(this demo): help Hello
>> Hello, MPI_COMM_world! I'm rank 0/1 (xxx)
> I can start octave without mpirun, but when I try:
> [marin][~/octave/mpitb/Hello]> octave -q --eval Hello
> [...]
> Help(this demo): help Hello
> -----------------------------------------------------------------------------
> It seems that at least one rank invoked some MPI function before
> invoking MPI_INIT. The only information that I can give is that it
> was PID 27255 on host marin.
> -----------------------------------------------------------------------------
>
> So, Hello just fails
Well, proving it's Hello's fault would be rather involved, since this is
Hello's code:
> function Hello
> % The classical greeting: "Hello (MPI_COMM_) world! I'm rank m/n"
> [...]
> info = MPI_Init;
> [info rank] = MPI_Comm_rank (MPI_COMM_WORLD);
> [info size] = MPI_Comm_size (MPI_COMM_WORLD);
> [info name] = MPI_Get_processor_name;
>
> fprintf("Hello, MPI_COMM_world! I'm rank %d/%d (%s)\n", rank, size,
> name);
>
> info = MPI_Finalize;
>
> [...]
So there is no single line of code before MPI_Init. Should every comment
be deleted, the resulting text would be
> function Hello
> info = MPI_Init;
> n0 localhost:3:origin,this_node
That worked fine. You mentioned 3 cpus in your single localhost node,
and there they are.
> The output I get when I run: mpirun -np 2 octave -q --eval Hello
> is:
> MPI: marin: 0x132c0000464b9eb5: octave: Command not found.
> MPI: could not run executable (case #4)
> Killed
Hmpf, I tried to find the text "case #4" in mpirun sources to no avail.
> I had to put -np instead of -c. with -c I just get:
> MPI: bad process count
Ouch! That's probably not LAM's mpirun, but other people's mpirun.
> It was mentioned that maybe the architecture is not good. The system
> I am running it on is an Itanium2 SGI Altix 4700.
> Any ideas?
Double-check you have indeed LAM/MPI installed. Try the "which" and
"laminfo" commands suggested in MPITB web.
http://atc.ugr.es/~javier/mpitb.html#Installing
My guess is you are not using LAM's mpirun.
Brian can probably provide not only a better guess on the "case #4" and
"bad process count" error messages, but also better instructions based
on LAM's FAQ pages. If he suggest something different, do follow his
advice instead of mine :-)
-javier
|
