[MITgcm-devel] [MITgcm-cvs] MITgcm/pkg/dic CVS Commit

[Martin Losch]

Hi Jean-Michel and users/developers of the dic-package,

I (hopefully) fixed the adjoint of bio_export.

I changed car/phos_flux for better vectorization. The code looks quite different now, because I needed to take the k-loops outside the i/j-loops. Please have a look at it to see whether you can live with these changes. I changed the logic a little, but the recompuations (because of depth_u/l) have gone now.

Note: phos/car_flux are enormously expensive. In the verification experiment GCHEM_FORCING_SEP takes three times longer than SOLVE_FOR_PRESSURE, mainly because of phos_flux (but car_flux is also a contributor). On the vector computer, phos_flux uses about 10% of the entire computer time, because of this: (depth_l-zbase)**(-Kremin). The statement is executed with a double k-loop (k=1,Nr, and ko=k+1,Nr). Maybe, there is a better way of coding this parameterization?

Martin


On Aug 1, 2010, at 1:51 PM, Jean-Michel Campin wrote:

> Hi Martin,
> 
> Looks like the adjoint experiment that uses pkg/dic are broken
> (does not matter machine/compiler, they are all bad):
>> Y Y Y Y 12> 0<FAIL  tutorial_dic_adjoffline
>> Y Y Y Y 16> 5<FAIL  tutorial_global_oce_biogeo
> Can you fix this problem soon ?
> Thanks,
> Jean-Michel
> 
> On Fri, Jul 30, 2010 at 11:02:43AM -0400, Martin Losch wrote:
>> Update of /u/gcmpack/MITgcm/pkg/dic
>> In directory forge:/tmp/cvs-serv13258/pkg/dic
>> 
>> Modified Files:
>> 	bio_export.F 
>> Log Message:
>> make bio_export vectorize on SX8
>> 
>> 
>> _______________________________________________
>> MITgcm-cvs mailing list
>> MITgcm-cvs at mitgcm.org
>> http://mitgcm.org/mailman/listinfo/mitgcm-cvs
> 
> _______________________________________________
> MITgcm-devel mailing list
> MITgcm-devel at mitgcm.org
> http://mitgcm.org/mailman/listinfo/mitgcm-devel