I’ve been trying out the partitioned communication that was added in openmpi 5. I am aware that “The current implementation is an early prototype and is not fully compliant with the MPI-4.0 specification” as it says in the docs. However the deadlock issue that I have encountered doesn’t seem to pertain to that (at least as far as I can tell).

I’m implementing an Allreduce algorithm inside the mca/coll framework.