Automatically generate the correct platform-dependent parallel spawn command for examples

[MTCam]

In the current run_examples.sh . It should pass on most workstations which have mpiexec. However, this line will fail on LC machines due to it not using an appropriate parallel spawn command.

For quartz, we can easily just replace mpiexec with srun: srun -n <nproc> -ppdebug python <thing>-mpi.py. There is similar command for Lassen.

We could easily add hostname-parsing to the run_examples.sh script and be done with it (until the next unique platform) - but I wanted to ask the group. Is there a better, cleaner, or more robust way?

[MTCam]

Related issue #69

[matthiasdiener]

Instead of parsing the hostname, I would suggest looking for the presence of srun, jsrun, mpirun etc., and using the one that is found as the spawn command.

[MTCam]

Instead of parsing the hostname, I would suggest looking for the presence of srun, jsrun, mpirun etc., and using the one that is found as the spawn command.

Uh oh.

[mtcampbe@lassen708:~]$ which srun
/usr/tcetmp/bin/srun
[mtcampbe@lassen708:~]$ which mpirun
/usr/tce/packages/spectrum-mpi/spectrum-mpi-rolling-release-xl-2020.06.25/bin/mpirun
[mtcampbe@lassen708:~]$ 

[matthiasdiener]

Instead of parsing the hostname, I would suggest looking for the presence of srun, jsrun, mpirun etc., and using the one that is found as the spawn command.

Uh oh.

[mtcampbe@lassen708:~]$ which srun
/usr/tcetmp/bin/srun
[mtcampbe@lassen708:~]$ which mpirun
/usr/tce/packages/spectrum-mpi/spectrum-mpi-rolling-release-xl-2020.06.25/bin/mpirun
[mtcampbe@lassen708:~]$ 

Is that an issue?

if [[ command -v lrun ]]; then
# Run with lrun
elif [[ command -v srun ]]; then
# Run with srun
elif [[ command -v mpirun ]]; then
# Run with mpirun
fi
#etc.