[MITgcm-devel] do we need exch_z ?

[Jean-Michel Campin]

Hi,

Here is what I found, trying to run cs-32x32 on 6 cpu:

a) presently, there is no exch_Z in pkg/exch2 but
ini_curvilinear_grid.F call 3 times exch_Z (for xG, yG, and rAz)
The consequence is that the exch1 version of exch_Z (eesupp/src) 
is used instead, which is fine on 1 processor but make the model
blowing up after 2 iterations when trying on 6 cpu.

b) if I comment out the 3 calls to exch_Z, aim.5l_cs
produces identical results on 1 and 6 cpu.

But the problem is that rAz, xG, yG (and fCoriG) and no longer
valid outside (1:sNx+1,1:sNy+1). This is likely to cause
problems with high-order momentum advection (VI form) and
high-order viscosity. (Alistair, can you confirm ?).

c) I had a quick look at 
MITgcm_contrib/high_res_cube/code-mods/ini_curvilinear_grid.F
and did not see any call to exch_Z.
Is it possible that rAz is set to zero outside of (1:sNx+1,1:sNy+1) ?
and has some effect on the blowing up at the edgde of the cube ?

Jean-Michel