July 31, 1999
Gentle Reader,

This is a note sent to a Fujitsu VPP user who had written to call
attention to irregularities along the southern boundary of the
innermost nest of a multi-domain simulation.

-Rotang

--------------


I have a possible fix for you.  In the file MPP/RSL/Makefile.RSL,
please change the line that reads:

    @ RDP=`fgrep RSL_DEFAULT_PADAREA $(RSLLOC)/rsl.h | cut -f 3 -d ' '` ; export RDP ; \
      if [ $(MPP_TARGET) = vpp ] ; then \
       echo '      PARAMETER(MIX=MIX_G)' >>parame.incl ;\  <<<<<<<===== this one

to read, instead:

    @ RDP=`fgrep RSL_DEFAULT_PADAREA $(RSLLOC)/rsl.h | cut -f 3 -d ' '` ; export RDP ; \
      if [ $(MPP_TARGET) = vpp ] ; then \
       echo '      PARAMETER(MIX=MIX_G+2)' >>parame.incl ;\
                                      ^^
                                   add '+2'


(Note that the lines start with a tab character).

This may or may not be the problem, depending on whether your innermost
domain i-dimension and the setting of MIX in the configure.user file
are identical. If they are, there might have been problems such as what
you were seeing.  If MIX in your configure.user file was larger than
the i-dimension of the innnermost nest, then this probably wasn't the
problem but it's a good fix to have in the code anway.

The extra 2-cells in the I dimension are needed because of the way the parallel
code initializes the domains with interpolated data from the parent.  Adding 2
makes MIX a multiple of 3, which is the nesting ratio.

Once you've made the change to the code, you should do a complete reinstall and
recompile:  

   type:
         make uninstall
         make mpp