[Mitgcm-support] xolas & cs

[mitgcm-support at dev.mitgcm.org]

Hello,

1) with aim.5l_cs on xolas, the f77 option -xarch=v9 produces
bad results. -dalign is OK and the model seems to run
faster with it. 
It could be interesting to use also -dalign with hs-cs
since -fast does not work with hs-cs (I confirm).

2) concerning the test 1+1=2 on xolas for hs-cs:
I get very small differences in cg2d_b (about 10e-15, 10e-16 
smaller that cg2d_b), so small that only few points are affected, 
and these points change when I change 1 parameter 
(e.g. No-Shapiro/with-Shapiro ...) or 1 compiler option ; 
but they are always close to the edges of the cube (i=32 or j=32).

Those differences are small enougth so that 
MONITOR & CG2D output are not affected in the 1+1=2 test,
but it's enougth to change the model state and I got
significant differences after 6 months.

I add an exchange of guNm1,gvNm1 just at the beginning of
solve_for_pressure and now the test 1+1=2 pass.

I agree that this is not the best solution but
it's not easy to figure out what happens in the overlap
of the cube, specially on xolas and when the differences
are small.

See you,

Jean-Michel