Troy

The code/Makefile/datasets are available using anonymous FTP from
ftp.cs.man.ac.uk:/pub/cnc/michael/Dec02/code.tar.gz (nearly 200Mb I'm afraid); I
compile/link it using 'make MPI2occtest'

When compiled and run on our system, I'm interested in the output in fort.221,
part of which is below


  1846   DEBUG: h0(3,2,1)= 6.9204015353931241E-6
  1847   doing i= 2 of 3
  1848   targetPE= 28  slavePE= 27
  1849   done get from win_n_h0
  1850   done get from win_iswm1
  1851   done get from win_imt_m
  1852   done get from win_jswm1
  1853   date/time:Thu Dec 12 20:01:52 2002
  1854   done all win scalars post GETs
  1855   m= 134 of 1
  1856   myI,myJ= 2*2 targetI, targetJ= 2*181 on PE rank 28
  1857   h0 was 0.E+0
  1858   date/time:Thu Dec 12 20:01:57 2002
  1859   DEBUG: h0(3,2,1)= 6.9204015353931241E-6
  1860   DEBUG: myI,myJ,nr_h0: 2*2,  1
  1861   h0 now 6.86486895749535764E-6
  1862   DEBUG: h0(3,2,1)= 0.E+0


As you can see, concentrating on h0(3,2,1) it is set to a value (much earlier in
the listing) but for some reason it becomes zero (line 1862) even though that
mpi_get is writing to h0(2,2,1)

The relevant Fortran code is in MPI2src/MPI2comms.F -- it's pretty messy since
I've been cutting down and messing with this code to try and work out what's
going wrong. (Somebody at SGI informed my there's no need for MPI_Fence hence
why most/all of them are commented out.)

If you could enlighten me to my mistake/oversights it would be a very, very nice
Xmas present!

Cheers,
Michael