Memory Bug requries compilation flag to zero all arrays to overcome in AMM15

[deazer]

Context

AMM15 cannot be run in NEMO4 Beta on a large number of nodes without the use of a flag to initialize all arrays to 0.
e.g. -ez on cray

Analysis

AMM15 will not restart if using a certain number of nodes. Below tis number it will run but wil not be reproducible across
different domain decompositions without the -ez flag.

Initial analysis isolated the arrays ahmt, ahmf in ldfdyn.F90

these can be set directly to zero, e.g.

ahmt(:,:,:) = 0._wp
ahmf(:,:,:) = 0._wp

and the model will at least run but remains not reproducible

Further analysis is required to isolate remaining uninitialized data that is causing the reproducibility problem in AMM15

Note the default configurations do not appear to be affected by this bug.

Fix

...

[smasson]

Very strange...
I don't see why initialise ahmt, ahmf to 0 in ldfdyn.F90 would change something...
Which value of nn_ahm_ijk_t do you use?