Ada for numerics (was: Let's help GNU/FSF with their dilemma

[ig25@fg70.rz.uni-karlsruhe.de (Thomas Koenig)]

[X-Posted to comp.lang.ada,comp.lang.fortran, F'up]

In gnu.misc.discuss, Craig Burley <burley@gnu.ai.mit.edu> wrote:

>That's for sure.  There are several languages out there that seem
>to be well-designed for concise expression of scientific and numeric
>processing.  C/C++ is not one of them, but Ada, even APL, might well
>be -- I haven't looked into these myself, but numeric types seem to
>find them to be pretty decent.

As far as numerical stuff is concerned, Ada 95 doesn't impede compiler
optimization as much as other languages (such as C) do; I'd say it's
about on par with FORTRAN 77 in that respect.  As far as I can see,
its 'in' and 'out' parameters should give a compier enough hints
to avoid the nasty aliasing problem, which is the bane of C.

If you want to do serious numercrunching on vector and parallel
machines, you're probably better off with High Performance Fortran or
Fortran 95, which are MUCH better at describing data (in)dependence with
FORALL and similar things.

IMHO, the people who design Fortran 2000 could do much worse than take a
close look at Ada 95 when they want to add some kind of object -
orientation to Fortran.  OTOH, the people who design the next Ada
standard might also take a look at High Performance Fortran.  Given two
equally fast compilers, one for Ada, and one for Fortran, I'd much
rather use Ada - since it doesn't have to stay compatible to older
versions, its design (and syntax) is much cleaner.