Custom Query (100 matches)

With XIOS the model crashes on Ada with XIOS (but not without), with the following error message:

terminate called after throwing an instance of 'xios::CNetCdfException'
  what():  Error when calling function nc_enddef(ncId)
NetCDF: HDF error
Unable to end define mode of this file, given its id: 65536
forrtl: error (76): Abort trap signal
Image              PC                Routine            Line        Source             
libifcoremt.so.5   00002ADA7C66B3EF  for__signal_handl     Unknown  Unknown
libpthread-2.12.s  00002ADA7A7F27E0  Unknown               Unknown  Unknown
libc-2.12.so       00002ADA7EEED495  gsignal               Unknown  Unknown
libc-2.12.so       00002ADA7EEEEC75  abort                 Unknown  Unknown
libstdc++.so.6.0.  00002ADA7AA90BCD  _ZN9__gnu_cxx27__     Unknown  Unknown
libstdc++.so.6.0.  00002ADA7AA8EB86  Unknown               Unknown  Unknown
libstdc++.so.6.0.  00002ADA7AA8EBD1  Unknown               Unknown  Unknown
libstdc++.so.6.0.  00002ADA7AA8EDE8  Unknown               Unknown  Unknown
...
gcm_48x36x19_phyl  0000000000D33951  iophy_mp_histwrit         974  iophy.f90
gcm_48x36x19_phyl  0000000002EE3365  phys_output_write         458  phys_output_write_mod.f90
gcm_48x36x19_phyl  00000000018A0037  physiq_mod_mp_phy        4581  physiq_mod.f90
gcm_48x36x19_phyl  00000000013D9905  callphysiq_mod_mp          70  callphysiq_mod.f90
gcm_48x36x19_phyl  0000000001397F34  calfis_loc_               718  calfis_loc.f
gcm_48x36x19_phyl  00000000009B2B81  call_calfis_mod_m         205  call_calfis_mod.f90
gcm_48x36x19_phyl  0000000000596751  leapfrog_loc_             796  leapfrog_loc.f
gcm_48x36x19_phyl  000000000042AC4B  MAIN__                    444  gcm.f90

Problem found on Ada using LMDZ r3543 (but also occurs for other recent revisions) and XIO revision 1659.

Sur Jean Zay, avec les options de debug le bench plante (en séquentiel) pour la r3572 dans pbl_surface lors du calcul de tke_x(i,k,is_ave):

       IF (iflag_split .eq.0) THEN
        wake_dltke(:,:,nsrf) = 0.
        DO k = 1, klev+1
           DO j = 1, knon
              i = ni(j)
!jyg<
!!              tke(i,k,nsrf)    = ytke(j,k)
!!              tke(i,k,is_ave) = tke(i,k,is_ave) + ytke(j,k)*ypct(j)
              tke_x(i,k,nsrf)    = ytke(j,k)
              tke_x(i,k,is_ave) = tke_x(i,k,is_ave) + ytke(j,k)*ypct(j)
!>jyg
           ENDDO
        ENDDO

pour i=14, k=1, car ytke(j,k)=nan (probablement pas/mal initialisé)

To better match with other models (ORCHIDEE, NEMO, ...), it would make sense to have the arch.path follow the same rules, i.e. definitions using 3 sets of keywords :

SOMETHING_LIBDIR ="-L....."
SOMETHING_LIB="-l..."
SOMETHING_INCDIR="-I...."

Of course thins implies adapting the makelmdz[_fcm] scripts accordingly.

An important benefit of having matching arch.path is that the file can then be shared by all models when compiling a combination (e.g. LMDZ+DYNAMICO+XIOS+IOIPSL)