[MITgcm-devel] OpenAD testing broken

[Jean-Michel Campin]

Hi,

With the addition of find_hyd_press_1d.f in pkg/openad/openad_ad_diff.list
the experiments that change are:
Y Y Y Y 14>13< 8 pass  global_ocean.90x40x15
Y Y Y Y 16>16< 9 pass  isomip
Y Y Y Y 15>16< 8 pass  OpenAD
Y Y Y Y 14>13< 5 pass  tutorial_global_oce_biogeo
Y Y Y Y 14>16< 0 pass  tutorial_global_oce_optim
compared to yesterday:
Y Y Y Y 16>16<16 pass  global_ocean.90x40x15
Y Y Y Y 16>16<16 pass  isomip
Y Y Y Y 16>16<16 pass  OpenAD
Y Y Y Y 16>16<16 pass  tutorial_global_oce_biogeo
Y Y Y Y 16>16< 0 pass  tutorial_global_oce_optim

I will check-in this addition, and since none of the
experiments are failing (must be related to different machine
truncation), will wait until tomorrow to update few of them.

Cheers,
Jean-Michel

On Fri, Mar 11, 2016 at 10:47:18AM -0500, Jean-Michel Campin wrote:
> Hi,
> 
> The changes I made yesterday are breaking the OpenAD built (except for
> halfpipe_streamice which does not use much of the ocean setting).
> 
> Nothing changed in forward test (since I kept the same default
> and make sure we get the same machine truncation).
> No issue with TAF since most of the changes/additions are in set_ref_state.F,
> called from initialise_fixed.F (which is not seen by TAF).
> 
> With OpenAD, it's different. I can fix the compilation step by adding:
>  find_hyd_press_1d.f
> at the end of:
>  pkg/openad/openad_ad_diff.list
> but for some reasons (not very clear to me) the results are affected
> (only checked with 1 exp. but going to process the other exp.):
> now:
>  Y Y Y Y 14>13< 8 pass  global_ocean.90x40x15
> yesterday:
>  Y Y Y Y 16>16<16 pass  global_ocean.90x40x15
> 
> Two questions:
> 1) Any idea why results are changing ?
> 2) Is there an other way to fix the problem that does not involve
>   differentiating this new routine "FIND_HYD_PRESS_1D" (and calling
>   routine "SET_REF_STATE") ?
> 
> Cheers,
> Jean-Michel
> 
> _______________________________________________
> MITgcm-devel mailing list
> MITgcm-devel at mitgcm.org
> http://mitgcm.org/mailman/listinfo/mitgcm-devel