Floating invalid in limvar.f90 and limthd_dh.f90

[mcastril]

Hi,

This issue is much similar to #1954. The problem is that we get a floating invalid in limvar.f90, line 135:

forrtl: error (65): floating invalid
Image              PC                Routine            Line        Source             
nemo.exe           0000000001B558B2  Unknown               Unknown  Unknown
libpthread-2.22.s  00002B5D8ABCDB10  Unknown               Unknown  Unknown
nemo.exe           0000000000F4AA69  limvar_mp_lim_var         135  limvar.f90
nemo.exe           0000000000CFA705  sbcice_lim_mp_sbc         389  sbcice_lim.f90
nemo.exe           000000000045CA63  sbcmod_mp_sbc_ini         374  sbcmod.f90
nemo.exe           000000000043CE20  nemogcm_mp_nemo_i         330  nemogcm.f90
nemo.exe           0000000000438D93  nemogcm_mp_nemo_g         120  nemogcm.f90
nemo.exe           0000000000438D5D  MAIN__                     18  nemo.f90
nemo.exe           0000000000438D1E  Unknown               Unknown  Unknown
libc-2.22.so       00002B5D8B0F76E5  __libc_start_main     Unknown  Unknown
nemo.exe           0000000000438C29  Unknown               Unknown  Unknown
forrtl: error (65): floating invalid
Image              PC                Routine

In that line there is this operation:

smt_i(ji,jj) = smt_i(ji,jj) + r1_nlay_i * rswitch * s_i(ji,jj,jk,jl) * v_i(ji,jj,jl) &

Apparently, looking at sbcice routine, the use of the s_i variable is the following:

    ierr = ice_alloc () ! ice variables       <-- Variables are allocated
      (Some stuff)
      CALL lim_itd_init ! ice thickness distribution initialization          <--  Does not use nor modify s_i
      CALL lim_hdf_init ! set ice horizontal diffusion computation parameters         <--  Does not use nor modify s_i 
      CALL lim_thd_init ! set ice thermodynics parameters              <--  Does not use nor modify s_i
      CALL lim_thd_sal_init ! set ice salinity parameters            <--  Does not use nor modify s_i  (only reads namelist parameters)
      CALL lim_msh ! ice mesh initialization               <--  Does not use nor modify s_i
      CALL lim_itd_me_init ! ice thickness distribution initialization for mecanical deformation   <--  Does not use nor modify s_i  (only reads namelist parameters)
      ! ! Initial sea-ice state
      IF( .NOT. ln_rstart ) THEN ! start from rest: sea-ice deduced from sst              <- Only if not a restart
         numit = 0
         numit = nit000 - 1
         CALL lim_istate           <-  Here the s_i variables are modified (only happens in the first chunk)
      ELSE ! start from a restart file
         CALL lim_rst_read        <--   Does not use nor modify s_i ( except if nn_icesal == 1, that in our case is 2)
         numit = nit000 - 1
      ENDIF
      CALL lim_var_agg(2)       < ---- Here the model crash    

Hence in our experiments, when we start a second year of simulation starting from restarts from the previous year, the run crashes at the beginning, presumably because the s_i variable has not been initiated due to the use of restarts.

We could solve this bug by initializing all the LIM variables at the ice.f90 routine.

[mcastril]

Hi,

We had more floating point issues at limthd_dh.f90, line 701. Using a debugger we could verify some NaN values in ht_i_1d variable, because this one was never initialized. We solved it by initializing to zero all arrays allocated in thd_ice.F90.

I can attach both files (ice.F90 and thd_ice.F90).

[ufla]

Adding some information about ht_i_1d in limthd_dh.F90:

The problem is that it is allocated for all points (jpij) in thd_ice_alloc, but only gets assigned values for indices kideb:kiut (1:nbpb when called from lim_thd) in lim_thd_dh. However, the statement

! --- ensure that a_i = 0 where ht_i = 0 ---
WHERE( ht_i_1d == 0._wp ) a_i_1d = 0._wp

(line 701 of limthd_dh.F90) accesses the full array. This will cause floating point exceptions, and crashes on systems that do not ignore these.

Besides, comparing ht_i_1d (or any other floating point number) to zero is not safe.

[mcastril]

Hi,

the assignation statement of ht_i_1d is an operation involving previous values (uninitialized) of ht_i_1d, so we can consider it "uninitialized" over all the longitude of the array, due to that values (involving NaN or random ones) can propagate though the operation (MAX).

[vancop]

obsolete for nemo v4