This reduces the time needed for initialization and increases the time
spent running a compute workload on all cores.

Follow the reference implementation more closely

There is no Julia equivalent for this code, and we want to run on all
cores anyways, so just let OpenMP do its thing