Print

Print


Steve Lionel and Erik Schnetter
Have both raised threads as a method
Of achieving parallel code.

I'll gladly add this if people can provide me with details
Of compilers that do this. 

I've also been contacted by a company that offers
Tools to parallelise existing serial 77, 90 and 95 code.

If anyone has any details in this area I'll add that too. 

Latest draft is below.

Parallel Programming with Fortran
=================================

The following Fortran standards have not addressed
the issue of parallel programming

  Fortran 66
  Fortran 77
  Fortran 90
  Fortran 95
  Fortran 2003

This has meant that there has been no
standard way of doing parallel programming in Fortran.

This has meant that a small number of 
ways of parallel programming developed 
independently of the Fortran Standards
Committees. These have included

  HPF
  MPI
  OPENMP

and each of these is covered in more depth below.

Thing's have changed with the proposed Fortran 2008
standard where co-arrays are an integral 
part of the language.

Two tutorials on parallel programming are given below.

http://www.mhpcc.edu/training/workshop/parallel_intro/MAIN.html

http://users.actcom.co.il/~choo/lupg/tutorials/parallel-programming-theory/p
arallel-programming-theory.html

1. Co-Array Fortran
-------------------

Co-array Fortran is a small extension to Fortran 95.
It is a simple, explicit notation for data decomposition,
such as that often used in message-passing models,
expressed in a natural Fortran-like syntax.
The syntax is architecture-independent and may be implemented
not only on distributed memory machines but also on
shared memory machines and even on clustered machines. 

This is a proposal for Fortran 2008.

See

http://j3-fortran.org/doc/meeting/173/05-208.txt

Older references can be found at

http://www.co-array.org/

and

http://lacsi.rice.edu/software/caf/

2. HPF
------

The High Performance Fortran Forum (HPFF),
a coalition of industry, academic and laboratory representatives,
works to define a set of extensions to Fortran 90 known
collectively as High Performance Fortran (HPF).
HPF extensions provide access to high-performance
architecture features while maintaining portability across platforms. 

http://dacnet.rice.edu/Depts/CRPC/HPFF/index.cfm

HPF Compilers

http://dacnet.rice.edu/Depts/CRPC/HPFF/compilers/index.cfm

Requires source code modification.

3. MPI
------

MPI is a library specification for message-passing,
proposed as a standard by a broadly based committee of vendors,
implementors, and users.

http://www-unix.mcs.anl.gov/mpi/

http://www-unix.mcs.anl.gov/mpi/mpich/

Requires the installation of the MPI library (some compiler
companies offer a bundle of compiler and MPI library)
and source code modification.

4. OPENMP
---------

The OpenMP Application Program Interface (API) 
supports multi-platform shared-memory parallel 
programming in C/C++ and Fortran on all architectures,
including Unix platforms and Windows NT platforms. 
Jointly defined by a group of major computer hardware and software vendors,
OpenMP is a portable, scalable model that gives
shared-memory parallel programmers a simple and flexible
interface for developing parallel applications for platforms
ranging from the desktop to the supercomputer.

http://www.openmp.org/drupal/

Resources (including compilers)

http://www.openmp.org/drupal/node/view/9?PHPSESSID=e69b7c73d8caca5ce49e5c229
882fe78

Requires source code modification.

5. Automatic
------------

By this is meant automatic parallelisation of the code 
without source code modification.

Table of compilers and supported parallel options
-------------------------------------------------------

               Co Array  HPF   MPI   OPENMP   Automatic

Absoft
                               Y     Y        

Compaq
  Tru64 Unix             Y           Y

Cray
  PVP                          Y     Y         Y
  T3E           Y              Y
  X1            Y              Y     Y

Fujitsu
  Sparc                              Y

HP
  HP-UX                              Y

IBM
  XL                                 Y         Y

Intel
                                     Y         Y

Lahey/Fujitsu
  Pro Linux                          Y         Y

NAG
  NAGWare                      Y

NEC
  SX                                           Y

Pathscale
                                     Y         Y

Portland Group
                         Y           Y         Y

SGI
  IRIX                               Y         Y

SUN
                               Y     Y

               Co Array  HPF   MPI   OPENMP    Automatic

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

Parallel Software Products
offer a tool to help parallelise Fortran 77, Fortran 90
or Fortran 95 code.

Their web address is

http://www.parallelsp.com/index.htm

The tool can generate either MPI calls or Openmp calls.


Cheers

Ian Chivers 
-----Original Message-----
From: Fortran 90 List [mailto:[log in to unmask]] On Behalf Of
Lionel, Steve
Sent: 22 August 2005 17:00
To: [log in to unmask]
Subject: Re: Parallel programming update

> 
> 5. SMP
> ------
snip 
> Several hardware manufacturers make SMP systems, and several compiler 
> companies now offer parallel programming on SMP systems.

It makes no sense to me to list SMP as a separate category for compilers.
SMP is a hardware architecture, primarily addressed in Fortran compilers
through OpenMP or automatic parallelization.  Cluster or distributed
parallel processing is primarily targeted by MPI, though you can use MPI on
a single SMP system and there exists "Cluster OpenMP"
as well, so the distinction is blurred. 

Furthermore, MPI is typically implemented as a library, not needed explicit
compiler support.  Most any compiler can support MPI if you have a suitable
implementation.

My recommendation is that you focus on the programming models available
(OpenMP, MPI, autoparallel) and not get bogged down in hardware
implementations. I'll also note that there are additional parallel
programming models, such as POSIX threads, or OS-dependent thread libraries,
that can be used from Fortran as well.  All a Fortran compiler would need to
take advantage of these is support for "recursive" routines, VOLATILE,
perhaps the ability to get the address of a variable (LOC()), and a
thread-safe run-time library.  If you expanded your definition of parallel
to that, there are many other compilers that could be mentioned (including
the HP OpenVMS compiler, and even DEC/Compaq/HP VAX Fortran which supports
autoparallel).

Steve