Jorge,
I am not that familiar with UCX, but I hope that will help:
The changes I mentioned were introduced by
https://github.com/open-mpi/ompi/pull/8549
I suspect mpirun --mca pml_ucx_tls any --mca pml_ucx_devices --mca pml ucx
...
will do what you expect
Cheers,
Gilles
On Mon, Sep 13, 2021
Dear Gilles,
Despite my last answer (see below), I am noticing that
some tests with a coarray fortran code on a laptop show a
performance drop of the order of 20% using the 4.1.1 version
(with --mca pml ucx disabled), versus the 4.1.0 one
(with --mca pml ucx enabled).
I would like to experime
