[MITgcm-devel] TAF version 2.8

[Jean-Michel Campin]

Hi,

I have just added the storage of the OB fields for U & V (as it was already
done for T & S) and the recomputations go away with latest version of TAF (2.8.3).
in experiment obcs_ctrl.
Also the output_adm.txt is back to what it was with version 2.7.8

Cheers,
Jean-Michel

On Thu, Aug 21, 2014 at 07:44:45PM -0400, Jean-Michel Campin wrote:
> Hi,
> 
> exeriment obcs_ctrl looks less easy. The additional recomputations come
> from "the_main_loop" with this message:
> >  The responsible variables are : obeu,obev,obnu,obnv,obsu,obsv,obwu,obwv
> Now, one thing I noticed is that in autodiff_store/_restore.F
> and in obcs_ad_check_lev1_dir.h, we store the OB fields for T & S 
> but do not store the OB fields for U & V.
> May be this is the reason ?
> 
> Cheers,
> Jean-Michel
> 
> On Thu, Aug 21, 2014 at 06:44:51PM -0400, Jean-Michel Campin wrote:
> > Hi,
> > 
> > Yesterday TAF was been updated to version 2.8.
> > It fixes one issue we had with TLM (that was responsible for
> > global_ocean.90x45x15 fail with gfortran & -devel).
> > Also we are getting more recomputaion warnings for 
> > 2 experiments: global_ocean.cs32x15 (50 -> 52) and obcs_ctrl (0 -> 16).
> > 
> > With global_ocean.cs32x15, it seems that the previous code (without 
> > recomp warning for thsice_main.F) might be problematic regarding the case
> > thSIceBalanceAtmFW > 0 with calls to thsice_balance_frw & thsice_balance_frw_ad.
> > So, version 2.8 might just be picking something that previous version
> > (2.7.8) was ignoring.
> > It's not difficult to remove these recomputations (just adding 1 store dir)
> > and will check-in this modif.
> > 
> > I did not look yet at obcs_ctrl.
> > 
> > Cheers,
> > Jean-Mcihel
> > 
> > _______________________________________________
> > MITgcm-devel mailing list
> > MITgcm-devel at mitgcm.org
> > http://mitgcm.org/mailman/listinfo/mitgcm-devel
> 
> _______________________________________________
> MITgcm-devel mailing list
> MITgcm-devel at mitgcm.org
> http://mitgcm.org/mailman/listinfo/mitgcm-devel