Performance of FVCOM on AMD nodes

[jimc101]

Tests indicate FVCOM performs poorly on AMD nodes when built with the intel compiler. This is highlighted by excessive run times on low core counts, and random segmentation violation issues on higher counts. The cause of this is still being investigated.

[jimc101]

Update - the issues appears to depend on the version of the Intel compiler used to build FVCOM. Poor performance has been observed when building FVCOM with Intel2020/2, and also later releases. In a test performed with the Intel2018/2 compiler, run times were significantly improved.