Maps in MPI for quick structure generation

Alloy Theoretic Automated Toolkit (ATAT)
1

Hello

I compile maps in mpi by "make mpi" in the atat folder.
Couple of questions/remarks:

  1. Do I need to do "make clean" before this, or does it overwrite a previous make?
  2. I work on a supercomputer and the mpiCC is set and working (I use gcc with openmpi).

The compilation works.

I then take the mpimmaps exec and move to the folder I am using (or equivalently move to the bin folder in my home dir and add bin to the PATH, but whatever).

But then I run:

mpirun -n 2 mpimmaps -c=3

and the output from maps is printed twice.

To me, this signals that the MPI is not working: maps runs in serial, one instance on each core.
But I need to run in true parallel of course (this is to speed up the structure generation). I wonder why the compilation shows no issues and the exec is created.

I assume the expected behavior of maps is to print only once its statements to stderr/out (e.g., I get printed "finding best structure" twice, but I assume the expected behavior is printing this once, just find the new structure in parallel).

How can I debug this/has others come across similar issues?

Thanks

2

Doing a make clean first is safer.

The tracing info sent to stdout and stderr is duplicated but the structure generation part is truly parallel. Sorry for the confusing output.

If in doubt you could compile genstr with MPI (which only generates structures and nothing else) and check that you get a clear speedup when running in parallel.