[MITgcm-devel] gfortran AD testing

[Jean-Michel Campin]

Hi Patrick,

I had a quick look at the AD test on dickens with gfortran
+ check-bounds & floating operation check :

After fixing the lockey_2 expression in mom_calc_visc.F
(using: lockey_2 = i+Olx + (sNx+2*Olx)*(j+Oly-1) 
     &                   + (sNx+2*Olx)*(sNy+2*Oly)*(lockey_1-1)
instead of:
        lockey_2 = i + sNx*(j-1) + sNx*sNy*(lockey_1-1)
)

I am left with this list of problems (see below).
I don't know how serious the dm95 tapering problem is
(floating exception) but the other looks "fixable".

Cheers,
Jean-Michel

PS: should I check-in the lockey_2 fix in mom_calc_visc.F ?
---------------------------------------------------------------

global_ocean.90x40x15   <- lockey_2 ;
 floating exception in adgmredi_slope_psi, 
 GM_taper_scheme.EQ.'dm95', Pb with COSH function

global_ocean.cs32x15
 floating exception in adgmredi_slope_limit, 
 GM_taper_scheme.EQ.'dm95', Pb with COSH function

lab_sea & lab_sea.noseaicedyn
 Pb with argument list order in SUBROUTINE ADCTRL_BOUND_3D 
 which does not match "call adctrl_bound_3d" in ad_taf_output.f

lab_sea.noseaice
 floating exception in adgmredi_slope_limit,
 GM_taper_scheme.EQ.'dm95', Pb with COSH function

offline_exf_seaice
 Array reference out of bounds for array 'ndaymonth', 
 lower bound of dimension 1 exceeded (in file 'cal_timepassed.f', at line 657)

OpenAD
 floating exception in adgmredi_slope_limit,
 GM_taper_scheme.EQ.'dm95', Pb with COSH function

tutorial_global_oce_optim
 floating exception in adgmredi_slope_limit,
 GM_taper_scheme.EQ.'dm95', Pb with in "smod)/gm_sd)**2)/gm_sd" 

tutorial_tracer_adjsens
 floating exception in adgmredi_slope_psi,
 GM_taper_scheme.EQ.'dm95', Pb with in "smod)/gm_sd)**2)/gm_sd"