John J. Lee
jjl at pobox.com
Sun Sep 7 00:39:05 CEST 2003
Basile STARYNKEVITCH <basile-news at starynkevitch.net> writes:
> >>>>> "John" == John J Lee <jjl at pobox.com> writes:
> John> "Michael Peuser" <mpeuser at web.de> writes:
> Michael>> Same argument holds for supercomputing as well. I may be
> Michael>> wrong but I doubt that the ASCIs will ever see much
> Michael>> Python in their production lifetime.
> John> ASICs, you mean? Well, no, but so what? I don't think
> John> anybody has ever *claimed* that Python is suitable for that
> John> kind of application.
> No, ASCI are big US government (DoE or DoD) supercomputers (used
> notably for nuclear weapons computations).
OK. (actually, um, do ASICs have any code in them at all? can't
actually remember what they are...)
> However, contrarily to Michael's belief, it won't surprise me at all
> that some big numerical computations are driven by scripting languages
> (scripts which call big number crunching primitives coded in C or C++
> or Fortran). At least in Europe, several number crunching applications
> are driven by scripts. Of course, a huge fraction of the CPU time (ie
> >= 98%) is spend in numerical routines coded in Fortran or C. Only a
> tiny fraction of the work is spent in interpreting scripts.
Indeed, IIRC David Beazley wrote SWIG for just that purpose
(controlling molecular dynamics simulations of some kind from Python).
I think there's a paper about it, probably on the SWIG site.
Plenty of other examples, of course. Tcl is still quite popular for
this purpose, since that's what Ousterhout designed it for! And I
know Python has been ported to various Cray machines.
More information about the Python-list