Various tweaks to the unit tests that are needed to get them to complete successfully on the COSMA7 Skylake machines and AMD Rome processes, both using the Intel 2018 compiler (although these do also work for GCC/7.3.0 on COSMA7).
The volatile hints are hackish as the errors when calling cbrtf() (which caused
a segfault) are only seen when optimizing and all the local variable references are optimized
away, but seem to work.
The updates to testPeriodicBC are to fix an issue running on COSMA6, which looks
to be down to using aligned structs in non-aligned ways. Like not using posix_memalign
to allocate cells, and passing aligned structs by value, i.e. on the stack.