[MITgcm-devel] tutorial_global_oce_biogeo Adjoint

[Jean-Michel Campin]

Hi,
I did some test on my laptop, and things look good:
With gfortran, the difference between 1 tile and 4 tiles (as now) are
much reduced: 15 digits for cost & grad ; minimum 13 digits for all
ptracers stats. I also get same results (with standard 4 tiles set-up)
on dickens with gfortran and also with g77 on faulks.
I also tried with ifort on my laptop and it's much better:
15 & 16 digits for cost & grad and minimum 16 digits for tracers
(was failing before with only 2 & 4 digits for cost & grad and a minimum
of 3 digits for tracers).
Will check-in a new output_adm.txt.
Cheers,
Jean-Michel

On Mon, Apr 12, 2010 at 10:56:30PM -0400, Jean-Michel Campin wrote:
> Hi Patrick,
> 
> I think I found the Pb with multi-tile tutorial_global_oce_biogeo AD test:
> 
> It comes from fe_chem.F (which is not tested in the forward experiment
> since we have "#undef  ALLOW_FE " in code/DIC_OPTIONS.h) but is used
> in the AD test (in code_ad/DIC_OPTIONS.h : #define  ALLOW_FE).
> 
> I will check-in the fix now. 
> Will see tomorrow if we need to update output_ad.txt (it take some 
> time to compile+run and would like to test more than 1 compiler).
> 
> Cheers,
> Jean-Michel
> 
> _______________________________________________
> MITgcm-devel mailing list
> MITgcm-devel at mitgcm.org
> http://mitgcm.org/mailman/listinfo/mitgcm-devel