[MITgcm-devel] level 4 recomps

[Matthew Mazloff]

If anyone else wants to try to help with this it is easily  
reproducible with the verification experiments.

For example, I just added:
#define AUTODIFF_4_LEVEL_CHECKPOINT
to
lab_sea/code_ad_seaice_ocean/ECCO_CPPOPTIONS.h
and
       integer    nchklev_4
       parameter( nchklev_4      =    1 )
to
lab_sea/code_ad_seaice_ocean/tamc.h

and this caused taf to generate extensive recomputations

I am not sure when this happened as I haven't used level 4  
checkpointing in awhile....

Anyway, any other verification experiment you suggest I break?  Any  
insight is appreciated!

Thanks
Matt





On Apr 21, 2011, at 2:26 PM, Matthew Mazloff wrote:

> Hi Martin,
>
> Unfortunately no, it is still broken with even the latest code.  I am
> really having trouble tracking this down too....
>
> -Matt
>
>
>
> On Apr 21, 2011, at 12:21 AM, Martin Losch wrote:
>
>> Does that have anything to do with the new freeze_internal.F (that
>> will cause recomputations). Make sure that you have already the
>> latest where Dimitris has excluded this call of ALLOW_AUTODIFF_TAMC
>> defined.
>>
>> M.
>> On Apr 21, 2011, at 6:05 AM, Matthew Mazloff wrote:
>>
>>> Hello,
>>>
>>> I'm getting recomputations when I
>>> #define AUTODIFF_4_LEVEL_CHECKPOINT
>>> for the latest version of the code.  I can't figure out what is
>>> causing them, and TAF only states:
>>>
>>>
>>> TAF RECOMPUTATION WARNING DOLOOP_STMT ad_input_code.f:521248 in
>>> the_main_loop
>>> triggered by      :
>>> apressure0
>>> ,apressure1
>>> ,aqh0
>>> ,aqh1
>>> ,area
>>> ,atemp0
>>> ,atemp1,detahdt,dwatn,empmr0,empmr1,eta,etah,etan,evap,gs,gsnm1,g
>>> t
>>> ,gtnm1
>>> ,gunm1
>>> ,gvnm1
>>> ,heff
>>> ,hflux0
>>> ,hflux1
>>> ,hsalt
>>> ,hsnow
>>> ,lwdown0
>>> ,lwdown1
>>> ,lwflux0,lwflux1,obns,obns0,obns1,obnt,obnt0,obnt1,obnu0,obnu1,
>>> obnv0
>>> ,obnv1
>>> ,pload0
>>> ,pload1
>>> ,pmepr
>>> ,precip0
>>> ,precip1
>>> ,qnet0
>>> ,qnet1
>>> ,qsw0
>>> ,qsw1,runoff,runoff0,runoff1,salt,saltflux0,saltflux1,seaice_sigm
>>> a1
>>> ,seaice_sigma12
>>> ,seaice_sigma2
>>> ,seaicemasku
>>> ,seaicemaskv
>>> ,sflux0
>>> ,sflux1,siceload,snowprecip0,snowprecip1,sss0,sss1,sst0,sst1,stress
>>> divergencex
>>> ,stressdivergencey
>>> ,surfaceforcingtice
>>> ,swdown0
>>> ,swdown1
>>> ,swflux0,swflux1,taux0,taux1,tauy0,tauy1,theta,tice,totphihyd,uic
>>> e
>>> ,ustress0
>>> ,ustress1
>>> ,uvel
>>> ,uwind0
>>> ,uwind1
>>> ,vice
>>> ,vstress0
>>> ,vstress1
>>> ,vvel,vwind0,vwind1,wspeed0,wspeed1,wvel,xx_aqh0,xx_aqh1,xx_atemp0,x
>>> x_atemp1
>>> ,xx_obcsn0
>>> ,xx_obcsn1
>>> ,xx_precip0
>>> ,xx_precip1
>>> ,xx_swdown0,xx_swdown1,xx_uwind0,xx_uwind1,xx_vwind0,xx_vwind1,zeta
>>> still to be recomputed : ycg2d_i_l1,ycg2d_i_r7,ycg2d_i_r8
>>> TAF WARNING ad_input_code.f:521248 : divided adjoint has not been
>>> generated
>>>
>>>
>>>
>>> Any ideas?
>>> Thanks
>>> Matt
>>>
>>>
>>>
>>> _______________________________________________
>>> 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
>
>
> _______________________________________________
> MITgcm-devel mailing list
> MITgcm-devel at mitgcm.org
> http://mitgcm.org/mailman/listinfo/mitgcm-devel