
CHANGES:
  In gfort: changed the fortran names gdsumset_, etc to gdsum and
now use the macros i.e. GDSUM rather then calling the gdsumset directly.

  Changed gscatter and gtoken to gscatterset and gtokenset to match the
other routines and so there would be no conflict with the fortran one.

  added psdestroy_ and pspartition_ to gfort.c

BUG:
  Fixed GCOLX man page and GCOL man page.

BUG:
 commdef.h line 46 gline = gsize was bug, should be *gline = gsize 
  
BUG:
  examples/example1.c had a bug in it. Note that pval must always 
be distinct for each different PSPartition and greater then 0. This
is because the way GMSMSG determines the message type you could get 
the same type for different procset. Note that this means that the
example in the talk slides is incorrect also.

BUG: not changed
  PIiFormP4PGFile you take the address of parameters; this could be
dangerous!!

BUG:
  Bug in initp4.c line 45 you set P4ARGC = argc but then in 85 if the 
-np # option was used you remove two arguments from argv but you 
do not change P4ARGC To fix it I simply set P4ARGC    = argc;
P4ARGV    = argv; below the call to PIiFormP4PGFile rather then before
it.

OTHER:
  Need a RECVTYPE ??

CHANGES:
  mfort.c provides  fortran interface for the message passing
  and procset

OTHER:
  we have to decide on a format for include files for fortran

CHANGES: In order to be able to use p4, etc. libs in makefiles
  I have included more things in bmake. $ARCH.$COMM which 
  defines a $CLIB  and a $COPT; I have tryed to update all 
  makefiles; I have not yet added all the new ones needed in bmake
  now we only need make BOPT=O COMM=pvm not make COMM=pvm OPT=-Dpvm

CHANGES: Got the pvm version working, major changes to commpvm.h
  and initpvm.h

BUG: in blkcm/compile/valid.c  around line 136 you have i loops inside 
i loops

BUG or CHNAGE:
made GCOL and GCOLX have a datatype associated with them; otherwise things
go bad for PVM.

BUG: NOT FIXED 
in blockcomm compile/findown.c you pass structures through GCOl
this should be changed
because of problems with hetrogenious systems. You use to use 
MSG_OTHER, in GCOL this WON'T work with PVM; the bytes count isn't an
accurate count of how many bytes were put in. I changed GCOL see above 
note above, now  compile/findown.c uses MSG_INT which works ok 
but for another system which may pad structures differently message
passing structures is always a dangerous game.

RULE: must use MSG_OTHER only to text strings; never for structures.

Need two clocks like SYsecond(); one for cpu time and one for real 
time.

Maybe change buildlib to match makefile interface
   bin/buildlib BOPT=O COMM=p4 ARCH=intelnx  ....
  simplier is usually better;  
