[Beowulf] A start in Parallel Programming?

Greg Lindahl greg.lindahl at qlogic.com
Tue Mar 13 16:32:06 PDT 2007


On Tue, Mar 13, 2007 at 12:59:49PM -0700, Jim Lux wrote:

> Which syntax compiler writers hate, because you can't use a parser 
> that always moves forward.  (But compiler writers love FORTH)

Er, compiler writers (like the ones around here) know that parsing is
the easiest part of compiling, so they don't stress about it.

Really, all this discussion about language misses the boat. You should
use whatever language is used in your environment. If you want to do
weather forecasting, that's Fortran95. As a student, it's trivial to
learn Fortran if you've picked up any other procedural language, and
if you're a scientist, it can't hurt to know Fortran. If you're going
to write some example programs, just be sure you're doing something
like a numerical program which is within Fortran's area of excellence.

I value students who know multiple languages over a student who's only
been exposed to one: there's no proof that the latter student is
mentally flexible.

The bit about not having libraries for Fortran isn't really true. It's
not hard to write glue routines to libraries in C, and in fact many
new important "Fortran" libraries are actually written in C (atlas,
fftw, MPI, hdf/netcdf, yadda yadda.) And I don't do 3D graphics in
Fortran: I have my Fortran code write out data, and I write a reader
for the graphics gizmo in whatever the preferred form is. Some gizmos
read HDF, some need some coversion.

The National Labs have a lot of codes which include big dollops of
F77, F95, C, and C++... in the same million-line code. They pay good
money, too.

-- greg




More information about the Beowulf mailing list