C++ programming (was Newbie Alert: Beginning parallel program ming with Scyld)

Mark Hahn hahn at physics.mcmaster.ca
Tue Oct 15 15:20:56 PDT 2002


> But seriously folks.  Would I get in TOO much trouble if I were to point
> out that if one is writing serious parallel numerical code, Fortran has
> much to recommend it, such as relative ease of parallelization and the
> fact that you can't get bogged down in trying recode the API just because
> you have easy access to raw system calls?

language choice is orthogonal to any interesting parallel programming issue.

and I don't really see how these claims can be true: Fortran itself
says nothing about parallelization, so you must be talking about HPF
or Fortran+OpenMP.  but I'd also claim that C/C++ has for many years now
had "parallel-friendly" aliasing semantics, and I don't see any reason
it wouldn't offer the same "ease".  I don't know of any way that Fortran
is better suited for message-passing (and remember, this is a non-shared
memory clustering list...)

as for the bogging down claim, well, it's self-evident to me 
that lack of expressive power is never a good thing.  in the classroom,
sure, don't give beginners a lot of rope.  but I am a trained 
professional, and *do* write C++ of the mild variety: no exceptions, 
light use of templates, but classes and overloading, definitely,
and even some virtuals on stuff where it makes sense like guis.)




More information about the Beowulf mailing list