benchmarking Poly/ML

[Stephen Weeks]

> Those benchmarks are interesting!
> 
> Why don't you try Poly/ML?  It is a free implementation of Standard ML and
> in my experience it greatly outperforms SML/NJ (50% faster).

I went ahead and ran Poly/ML on the same examples.  Here are the running times.
Note that I had to use a safe Array sub and update for the Poly/ML version since
I didn't know how to use the unsafe ones.  I'm sure that hurt Poly/ML some, but
nowhere near enough to explain the slowdown.

              OCAML  MLton  SML/NJ  Poly/ML
holes           1.8    3.5     3.9     17.5
fov             1.5    1.8     3.2     24.6
intercyl        1.6    2.0     4.3     32.5
snowgoon        2.9    3.2     5.1     65.2
dice            3.9    5.1     8.8     72.5
golf            1.5    2.5     3.1     22.1
cone-fractal    3.7    4.4     6.5     82.7
large           4.3    3.5     6.7     11.1
pipe            5.4    4.7     7.9     74.6
chess          16.0   15.9    21.6    271.5
fractal        12.2   13.8    45.4    109.4
geom-mean       3.6    4.3     7.1     47.8