[Haskell-cafe] OCaml list sees abysmal Language Shootout results

[Greg Buchholz]

Malcolm Wallace wrote:
> Here are the results, in seconds, on my machine (2.4GHz x86/Linux)
> for the suggested input (N=500) from the shootout site.  All Haskell
> versions were compiled with ghc-5.04.2 -O2.

    I thought I'd take a stab at timing a few of the examples with
different compiler versions to see what difference that would make
(ghc-6.2.1 vs. ghc-6.3.20040928).  I compared Kevin Everets' version with
the two Tomasz Zielonka versions.  I ran the test with N=2500 (i.e. 2500
copies of the original file, which is what is apparently used in the
shootout) on my AthlonXP 1600 under x86/Linux.

			6.2.1		6.3.20040928
			-------		-------
Kevin			3.615s		3.156s                  
Kevin  (+RTS -G1)	1.666s		1.405s
Tomasz (pretty) 	0.725s		0.481s
Tomasz (fast)   	0.403s		0.430s

    Interesting to see the speed increase going from 6.2.1 to 6.3 for
Tomasz' pretty example.  Anyone have an explaination for the 2x speed
increase for running Kevin's version with '+RTS -G1'?

(And for reference, here's the results on my machine for the perl and
gcc versions of the test and gnu/wc)

perl-5.8.4		0.535s
gcc-3.4.2		0.102s
gnu/wc			0.435s

Greg Buchholz