CI: improve Time to Solution
- TTS on Alps using CPU only (single node) to be improved.
- The workload is best with square number for rank-count.
I am not sure, if it is a good idea to change the CPU_BIND from cores to sockets. I would also keep the explicit MPI_TYPE specification pmi2 instead of relying on the default one even if it might be the same currently.
I tested CPU_BIND=sockets since then. Though, it was suggested by CSCS Documentaiont for ALPS and CP2K in particular (https://docs.cscs.ch/software/sciapps/cp2k/#running-on-daint). I can change it back once the PR passes for the first time.
In fact, I would like to avoid binding/pinning procs to CPU cores in a consecutive fashion (mainly because of OpenMP). I think the underlying MPI implementation has a default scheme for placing the procs (once forced to a socket ;-).