Re: Sisal?

wchang@genome.lbl.gov (William Chang) writes:
>Where might I find discussions/critiques/rebuttals regarding Sisal?
>Is this dataflow/single-assignment language "functional", and how does it
>manage to beat Fortran?


Interested parties can join the sisal-info mailing list or ask generic
questions to (mail to SISAL-INFO-REQUEST@SISAL.LLNL.GOV or me at
PATMILLER@LLNL.GOV). And the Sisal language is applicative (functional
semantics and expressions, but no higher order functions). A follow-on
language, Sisal '90 does have some higher order features.


preston@tera.com (Preston Briggs) writes:
|> In one of the Sisal papers, they discuss the language and an
|> implementation....


This is probably "Retire FORTRAN? A Debate Rekindled" by David Cann in
the August '92 CACM -- Its probably the best paper that is easy to
find. We have many other papers with more details available. Ask at one
of the addresses above.


|> However, the compilers were not equivalent, in that the Sisal compiler
|> examined the entire program, performing significant interprocedural
|> optimizations (and eventually emitting C to be compiled by the native
|> C compiler). The Fortran compiler, on the other hand, used the
|> traditional approach to separate compilation...


True to a point, but the SISAL compiler vectorizes better, and the
FORTRAN compiler cannot (because of semantic considerations) do the
global analysis in less than time exponentially proportional to the
size of the program anyway. The paper showed SISAL beating small to
medium sized FORTRAN programs. We have found greater winning margins
for larger codes (No really huge SISAL programs have been written to
compare).


|> practice, we would also expect the Fortran compiler to be much more
|> usable, since it doesn't require recompilation of the entire program
|> after every edit.


This is not a requirement, but is only necessary for optimal runs.
Turning on the full range of optimizations is a last step even for
FORTRAN development.


|> Nevertheless, the results are impressive and have attracted attention


Thanks!


--
Dr. Patrick Miller
patmiller@llnl.gov
--

Return to the comp.compilers page.
Search the comp.compilers archives again.