OS/2 Professional/OS2PRO194.ISO/os2/prgramer/adaptor/install.os2

Text File | 1994-01-04 | 5.8 KB | 112 lines

   This file describes the installation instructions for port of ADAPTOR,
software tool for Automatic Translation of Data Parallel Programs to the
OS2 platform.

   Please refer to readme.os2 for the description of the software and
for the list of requirements for sensible use of this software.

   Installation of this software is fairly painles. In adaptor\bin
directory you get 3 tools used for translation of Fortran programs
to Fortran programs that are parallel with the use of communication
libraries based on message passing paradigm:

 - fadapt.exe is the actual Fortran to Fortran translator
 - fstrip.exe is the tool to transform Fortran file generated by
              fadapt.exe to the usual form (it limits the length of the
              individual lines to be less or equal then 72 - or any
              other number between 50 and 150, but 72 is Fortran
              standard).
 - fsplit.exe Tool to split Fortran source file into files containing
              one routine each - functions, subroutines, block data
              and main program (that means each function is in separate
              file etc.). It is not directly used by the default setup
              of fadapt, it is written in FORTRAN (sic) and I am
              not even including source code, because to read command
              line options it uses some routines from the CERN library.
              I wrote it because I got frustrated not to have such a tool
              in OS2 (every compiler vendor of commercial Fortran
              compilers on UNIX platforms is shipping it with their
              compiler, but I am using f2c so I was forced to write
              it). But hey, it works.
Copy these 3 files to your usual tools directory or to the place in
your path.
fadapt.exe after invocation with the optional parameters (read original
documantation) generates around 10 files describing the process of
translating your code to parallel code and also generates makefile.
In order to be able to use makefile directly without ever looking at
it, you are asked to define two environmental variables:
 PVM_ROOT - it is the base directory of the directory structure for
            the PVM that you are supposed to use with this software
            (see readme.os2). For instance:  set PVM_ROOT=d:/pvm3
 DALIB    - this a place, where make (I am using dmake !) is supposed
            to find two libraries (dalib.a and unilib.a) + four
            object files (mhost.o, mnode.o, mnode1.o, mcube.o).
            They are shipped in directory adaptor\lib so you can
            set DALIB to point to this directory (or you can move them
            to \emx\lib and let DALIB to point there).
            Example:  set DALIB=d:/adaptor/lib


 Now you can go to the examples directory, go one level deeper into
simple and say:
         fadapt -H simple.f
If what happens is that you get various messages about different
stages of translation and it all ends up with the message that host.f
and node.f where successfully generated it seems to work. Now have
a look (al least this time) at the content of the makefile and
after that try to say: dmake in the same directory. If everything
goes OK, you are left with host.exe in that directory and node.exe
has been moved after creation to the directory where PVM expects
"worker processes" (refer to PVM documentation) which is typically
drive:/pvm3/bin/os2. Moving node.exe (so that it can be started by
PVM deamon) was the reason for defining PVM_ROOT. If you don't
define those two environmental variables, some reasonable defaults
are chosen. If you have PVM deamon running, try to run host.exe.
If everything works you are invited to read the original documentation
for getting real understanding how ADAPTOR works.
If nothing works you are invited to read the original documentation
for getting real understanding how ADAPTOR works plus to do some
thinking about what went wrong.

  You can recompile fadapt, fstrip plus libraries using makeadap.cmd
located in adaptor directory. If you need to modify the way makefile
is produced (because of the peculiarity of your setup) you have to
read (and at least partly understand) the file makef.c in adaptor\src
directory.


  Technical remarks:

  The port itself consisted in changing the names of the routines
to conform to 8.3 conventions and to change file makef.c in
adaptor\src directory. That file generates makefile and is the
most likely candidate for the one beeing changed if you want to
get different looking make file.

  As you might have seen, I assume that you have popular file
manipulation utilities from GNU collection: mv and rm.

  Fadapt can be compiled also with X-Windows support. I have avoided
to make it even more difficult to use this software, so I am avoiding
this option completely.

  One small warning: There is one small inconvenience present
in the current port. If you have DATA statements in your routines,
fadapt currently appends some additional definitions of its own to
the end of the declarative part of the routine. It means some INTEGER
and teh like definitions are appended after DATA statements. This
violates Fortran standard (though most commercial compilers don't
complain) and f2c complains about that. In this case you have to
move those data statement by hand. I have communicated the problem
to the authors and have some hope that it might be fixed in the next
release of ADAPTOR.
   Because of the character of the software, you are probably going
to have to work with generated code at least a little bit anyway.
But the quality of ADAPTOR as a tool is fairly high and it sure
can save a lot of time to people like me that would have to paralellize
the code by hand.

                                            Jan Ftacnik
                                     [e-mail]