CESM1_2_1 restart test fails for SE dycore

[andreasb]

Hi all,

I have performed a "restart test", i.e. checking if the results are the same for a 2 day simulation versus a 1+1 day simulation (write + read restart). The results are binary identical for a finite volume experiment
-res 4x5_4x5 -case test4messy_restart -compset F_2000_CN

but they diverge for SE:
-res ne16_ne16 -case test4messy_restart_se -compset F_2000_CN

model version is CESM1_2_1.

I would appreciate any feedback on this.

kind regards,
Andreas

 

[jedwards]

On what machine did you run this and did you make any changes from the defaults?

 

[andreasb]

This was on my own machine (g95 with openmpi on ubuntu) with g95 changes (see below). I just now repeated the same test on yellowstone, where it passes the test, so this seems to be g95-related, which is not officially supportet anyway. I will also perform the test on a IBM AIX next week, so unless this will fail I suggest to ignore this...Thanks, Andreasse/share/dimensions_mod.F90#ifndef __G95__
  public :: qsize,qsize_d,ntrac_d,ntrac
#else
  public :: qsize,qsize_d,ntrac ! ntrac_d already public
#endifand se/share/ll_mod.F90:  recursive subroutine copy_node(node2,node1)

    type (node_t), intent(out) :: node2
    type (node_t), intent(in)  :: node1


    node2%Src    = node1%Src
    node2%Dest    = node1%Dest
    node2%id      = node1%id
    node2%valid    = node1%valid
#ifndef __G95__
    node2%prev       = node1%prev
    node2%next       = node1%next
#endif

  end subroutine copy_node./Machines/config_compilers.xml:
  /soft/netcdf/v4.1.1_g95_64/
  g95
  mpif90
  gcc
  mpicc
  -DLINUX -DFORTRANUNDERSCORE
  -g -fno-second-underscore -DNO_CRAY_POINTERS -ffree-line-length-huge -fmultiple-save -fbounds-check -ftrace=full -O0