Custom Query (964 matches)

The makefile change introduced in ​src:ticket:854 needs to be updated to include libDRIVER.a twice. We need to include it again for cycling dependencies in old linkers. We need to have:

#--------------------------------------------------------------------------
#  Notice that the token "libraries" is initialized with the ROMS/Utility
#  library to account for calls to objects in other ROMS libraries or
#  cycling dependencies. These types of dependencies are problematic in
#  some compilers during linking. Such libraries appear twice at linking
#  step (beginning and almost the end of the ROMS library list).
#--------------------------------------------------------------------------

   libraries := $(SCRATCH_DIR)/libNLM.a $(SCRATCH_DIR)/libDRIVER.a \
                $(SCRATCH_DIR)/libUTIL.a

• The update includes changes to several files to allow mixed single and double precision in some applications. In single precision, we need to ensure that time and scale factors are in double precision. For example, in wrt_ini.F we have:

       real(r8) :: Fmin, Fmax
       real(dp) :: my_time, scale
  

  Recall that in single-precision the floating-point kind parameters are set in mod_kinds.F as:

          integer, parameter :: dp = SELECTED_REAL_KIND(12,300)  ! 64-bit
          integer, parameter :: r4 = SELECTED_REAL_KIND(6,30)    ! 32-bit
  # ifdef SINGLE_PRECISION
          integer, parameter :: r8 = SELECTED_REAL_KIND(6,30)    ! 32-bit
  # else
          integer, parameter :: r8 = SELECTED_REAL_KIND(12,300)  ! 64-bit
  # endif
  

• Modified routine wclock_off to report also the total profile times in the MPI communications for all nested grids (if NESTING is activated), and its total for all numerical kernels in 4D-Var applications.

• Corrected a violation to the Fortran standard that doesn't allow to have a subroutine and variable to be named the same. Some Fortran compilers relax that rule. However, gfortran complains. I missed this one. The variable increment is renamed to NLincrement.

Many thanks to Mike Shen for ​reporting this issue.

• Corrected segmentation fault in close_io.F when closing the BRY, CLM, and FRC files in the structure. The array index to the files component could go out-of-bounds.

• Updated timers.F to report the minimum and maximum processes wall CPU time during profiling in distributed-memory applications.