> -----Original Message-----
> From: Manuel M. T. Chakravarty [mailto:[EMAIL PROTECTED]]
> Sent: Wednesday, September 22, 1999 10:06 AM
> ...
>
> "R.S. Nikhil" <[EMAIL PROTECTED]> wrote,
>
> > ...
> > But it DID offer an important new feature relative to
> > the original Fortran programs it was trying to
> > displace -- completely automatic parallelization
> > for the Cray vector machines that were the main
> > workhorses at Livermore and similar labs.
>
> Hmm, good point. So, did they succeed in reaching this
> goal, compared to the awful guessing and trial & error you
> have to go through before you get any decent performance out
> of a vectorising Fortran compiler?
In the following paper, the Sisal team describes their technical
successes, including performance numbers.
@article{
Cann92a,
author = "Cann, David C.",
title = "Retire Fortran? A Debate Rekindled",
journal = "CACM",
volume = 35,
number = 8,
year = 1992,
month = "August",
pages = "81-89",
}
Regarding why Sisal research has died: there may be technical reasons,
but there is also a political dimension that is not very flattering
to the functional languages community of the time (1980's, early
1990's).
Sisal researchers adopted a very practical and laudable strategy.
Their primary goal was to wean their physicist colleagues off from
Fortran and on to something more disciplined, semantically sound,
analyzable and compilable. Competitive runtime performance was an
ABSOLUTE requirement. Instead of going for all the features of
functional programming, they chose to limit their focus to a
restricted language that did the job. They deliberatly chose to avoid
higher-order functions, polymorphism, laziness, etc., and focused on
just those features that they could compile into code that was
competitive with Fortran. Incidentally, this is exactly the
philosophy that Claus Reinke articulated in another message in this
thread when he said, about SAC:
"This bottom-up approach to functional language design and
implementation was chosen for SAC exactly because of the
difficulties of the standard, top-down approach".
In another decision aimed at making the language more familiar to
their Fortran-writing potential customers, the Sisal designers chose a
syntax that was Pascal-like (but semantically it was a pure functional
language).
Unfortunately, these choices won them no respect in the FP community
(for which, my commentary, shame on the FP community), who chose to
look down their noses at Sisal for what were essentially trivial and
shallow reasons (Pascal syntax, focus on those "dirty" arrays instead
of those "cool" lists, no polymorphism, no higher-order functions,
...). This made it MUCH harder for the Sisal team to sell the
language to their Fortran-writing colleagues, who kept hearing
negative opinions about Sisal from the FP community.
This may not have been the exclusive reason why Sisal did not take
off, but I have heard members of the Sisal team say, more than once,
that this was a major contributor. Anyway, eventually funding for
this project at Livermore ran out, and the team has dispersed and is
doing other things.
Nikhil
(opinions my own;
my employer will thrice deny any complicity or concurrence)
