[MITgcm-devel] an other new TAF version (2.8.2)

[Jean-Michel Campin]

Hi,

Just an update on this: 
I have tested few changes (including the split: dic_atmos -> dic_ini_atmos + dic_atmos)
and this works fine with the new version of TAF (now 2.8.3), 
without adding any new storage.

I am going to check-in these changes after testing the OpenAD case.

Cheers,
Jean-Michel

On Tue, Aug 26, 2014 at 04:44:44PM -0400, Jean-Michel Campin wrote:
> Hi,
> 
> OK, only these 2 experiments: 
>   tutorial_dic_adjoffline  (e=0, w=19)
>   tutorial_global_oce_biogeo  (e=0, w=24)
> have Pbs, and it seems that the causes of additional recomputations
> are similar in the 2 cases:
> < TAF RECOMPUTATION LOOP WARNING DOLOOP_STMT ad_input_code.f:XXX in the_main_loop
> <  extensive recomputations are required.
> <  The responsible variables are : atpco2
> < TAF WARNING TAF RECOMPUTATION WARNING DOLOOP_STMT ad_input_code.f:XXX in the_main_loop
> 
> One thing that might help would be to take out of S/R dic_atmos.F
> the part done at initialisation (when called from dic_surfforcing_init.F
> with argument istate=0) and to put it into a new S/R dic_ini_atmos.F 
> This would simplify dic_atmos.F (no need for argument "istate")
> that would only b called from one place, in gchem_forcing_sep.F 
> 
> Cheers,
> Jean-Michel
> 
> On Tue, Aug 26, 2014 at 02:14:23PM -0400, Jean-Michel Campin wrote:
> > Hi,
> > 
> > TAF has been upgraded to new version (2.8.2) this morning (between 7 & 7:30)
> > and it seems that we are getting more recomputations for few experiments. 
> > At least 2 of them (but may be more) changed, new=2.8.2 / old=2.8.1 :
> > < Y Y Y Y  0>14< 8 pass  tutorial_dic_adjoffline  (e=0, w=19)
> > < Y Y Y Y  0>13< 6 pass  tutorial_global_oce_biogeo  (e=0, w=24)
> > ---
> > > Y Y Y Y 14>14< 8 pass  tutorial_dic_adjoffline  (e=0, w=1)
> > > Y Y Y Y 14>13< 6 pass  tutorial_global_oce_biogeo  (e=0, w=6)
> > 
> > I am currently re-running last-night testreport with the same code
> > on baudelaire in:
> >  /scratch/jmc/test_baudelaire/MITgcm_Aug26/verification/
> > with previous testreport run saved in:
> >  /scratch/jmc/test_baudelaire/MITgcm_Aug26/verif_v2_8_1/
> > 
> > The little annoying thing is that these 2 experiments are now taking
> > much more time to run (because of re-comp): 
> > new=2.8.2 / old=2.8.1 on a slow & old 32-bit machine:
> > User:  2712.581 sys: 17.730 clock:  2762.222 <- tutorial_dic_adjoffline
> > User:  1607.022 sys:  7.395 clock:  1640.744 <- tutorial_global_oce_biogeo
> > 
> > User:   574.999 sys:  5.822 clock:   596.977 <- tutorial_dic_adjoffline
> > User:   569.409 sys:  4.006 clock:   593.723 <- tutorial_global_oce_biogeo
> > 
> > Cheers,
> > Jean-Michel
> > 
> > _______________________________________________
> > 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