Compiling with high optimisation

  1. A new 'high' optmisation level exists which should preserve all our basic tests, like NRUN vs CRUN and PE decomposition tests.
  2. It should also be scientifically equivalent to safe optimisation
  3. For GA
    • 5% faster on 1 thread
    • 7% faster on 2 threads

Problem is that 1) and 2) are NOT currently true for UKESM model

  • I've down graded optimisation on three files so that 1) works for UKESM model
  • A 50 year coupled run shows that this is still not scientifically equivalent to safe optimisation