In article <···················@liasg5.epfl.ch> ·····@lia.di.epfl.ch (Simon Leinen) writes:
> > Since two different Lisp's are off by a factor of 5 and 10 from
> > CMUCL, the benchmark probably needs a little work.
>
>One might also conclude that the two other Lisps need a little work.
One might conclude all kinds of things. But would one be right?
As has been pointed out a number of times, it can be difficult to
get the best results from several different Common Lisps. Sometimes
declarations that help a lot in one make things worse in another,
and so on. I find it easier to get the best numerical results from
CMU CL than from Lucid or Allegro. This may mean that *something*
about Lucid and Allegro needs a little work, or it may just mean
that I and the CMU CL folk think alike when it comes to this.
Another factor is that the person running the benchmarks may not
know all the Lisps equally well, just as someone running C benchmarks
may not know that "-O6" is the optimization setting to use for
compiler X.
>Maybe the problem is that it is simply very hard to implement all of
>Common Lisp in a maximally efficient manner.
And the ways to get the maximally efficient stuff vary.
BTW, it took a while for C to get as efficient as it now usually(?)
is. And where would many people be without gcc? Unfortunately,
the Lisp equivalent of gcc hasn't yet come along.
-- jeff