S3.3-h High parallelisation on new plateforms

Last edited Timestamp?

This page gives a non exhaustive description of the work (to be) done to optimise the code for massively parallele processing
There main strategies are follows :

• reduce the number of communication in the baroclinic part of the code
• generalise the time-splitting solver (educe the number of communication in the barotropic part of the code)
• re-activate the extended halo (switch from 1 to 2 overlap row/column)

---

1. Reduce the number of communication (baroclinic part)

1.1 Vertical mixing

Suppression of avmu and avmv arrays.

----

2. Reduce the number of communication (barotropic part)

---

3. Extended overlap halos