[MITgcm-devel] Beaufort experiment on mac os x

[Jean-Michel Campin]

Hi Martin,

Thanks for doing these tests. If you could sent the testreport
output as testreport -a jmc at mitgcm.org does, I think it could
be useful.
Regarding bottom_ctrl_5x5, it's good that you manage to avoid the NaNs,
and you have now the same numbers as on baudelaire, so it's good.

Regarding your fix to ini_masks_etc.F & update_masks_etc.F, 
since it does not change the results, could you take care of 
checking-in these changes ?

And finally, regarding this option to check for un-initialised variables,
there is indeed a problem in admom_fluxform (with non-lin free-surf or
bottom control) but did not have time yet to figure out the cause.

Cheers,
Jean-Michel

On Fri, Mar 16, 2012 at 03:36:09PM +0100, Martin Losch wrote:
> with 
>     NOOPTFILES='ini_masks_etc.F update_masks_etc.F gmredi_calc_psi_bvp.F'
> (and not changes to ini_masks_etc.F) I have no problems (do you need my testreport output? I forgot the -a) on my PowerBookPro with gfortran 4.6.0
> 
> Now I also managed to make bottom_ctrl_5x5 work on my linux box, and it fails with 
> Y Y Y Y 16> 3<16 FAIL  bottom_ctrl_5x5
> for both versions of ini_masks_etc/update_masks_etc, but I guess that's normal (judging from recent baudelaire results).
> 
> So I'd say, we can safely use the modified code and avoid -O3 problem with apple gfortran.
> 
> Martin
> 
> PS. Side effect: bottom_ctrl_5x5 adjoints produces NaNs when I tell gfortran to initialise by default with inf (-finit-real=inf), but I saw this is also happening e.g. on acesgrid with linux_amd64_open64, so there seems to be an initialization problem in the adjoint (or forward?) code.
> 
> 
> 
> _______________________________________________
> MITgcm-devel mailing list
> MITgcm-devel at mitgcm.org
> http://mitgcm.org/mailman/listinfo/mitgcm-devel