yeah, we're just better then awk!

[Jonathan Coupe]

----- Original Message -----
From: Pierpaolo BERNARDI <bernardp@REDACTED>

> Hmmm...  I have been surprised by Erlang in a test
> which probably is as far from its strengths as possible.
>
> I rewrote in Erlang a Common Lisp program which,
> essentially, accesses repeatedly a big 2d array,
> and does some manipulations of short lists.
>
> I obtain these times:
>
> Clisp: 11/12s
> HCL: 1.5/2s
> Erlang using gb_trees as arrays: 32/34s
> Erlang using an ets table as array: 7/8s
>
> I was expecting much worse results for Erlang.

The benchmarks I have put Clisp at about 1/4-1/8 the speed of ACL,
usually considred the fastest CL except possibly for CMUCL. (In the Scheme
world Gambit is much faster again at this sort of benchmark.) ACL had at
best 1/4 the speed of C for floating point intensive code the last time I
checked, assuming you really optimize to death. (The manufacturer's
optimization guy couldn't do any better than that.) So that puts Erlang at a
~ 1/10 the speed of C, using the best possible(?) data struture - which is
what I though Ericson claimed for its performance in general, until
threading overhead becomes significant.

Did you try using a guard to specify that the data was floating point,
 as Björn Gustavsson  suggested recently? I tried to get the code from the
link you gave (http://space.tin.it/romebern/erlab.zip) but it was broken.

- Jonathan Coupe

PS Is there a site for HCL? Although I've already decided I prefer Erlang to
any sort of Lisp.