Thanks for the report. I was able to reproduce the symptom. When commsander is the same thing as mpi_comm_world, quick appears to run fine, but when doing an alchemical "bookend" simulation (mm -> qm/mm) using a groupfile, the communicator that quick should be using is a subset of the available processes. It appears that quick is using mpi_comm_world, however.
Here are my observations:
1. "mpirun -n 2 sander.MPI -ng 2 -groupfile groupfile" hangs when using icfe=1 for both the mm and qm/mm
2. "mpirun -n 2 sander.MPI -ng 2 -groupfile groupfile" will successfully run the mm simulation the mm simulation when icfe=0; however, the qm/mm simulation hangs.
3. "mpirun -n 2 sander.MPI -O -p parm7 -c inp.rst7 -i qm.mdin ...." will successfully run the qm/mm simulation.
The quick code needs to be changed to accept a mpi communicator as an argument and exclusively use that communicator. ...at least that is my current working hypothesis.
-Tim
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Wed Sep 24 2025 - 08:00:02 PDT
