Nice algorithm! The idea of using vectors never occured to me.

I went back and uncomment the print statements to see what kind of overhead the grep might cause. At N=100000 I end up with:

Benchmark: timing 100 iterations of Ase, Maverick...
       Ase: 142 wallclock secs (141.93 usr +  0.04 sys = 141.97 CPU)
  Maverick: 166 wallclock secs (166.05 usr +  0.00 sys = 166.05 CPU)
[download]

I would have expected us to end up with closer times, but your's ends up being 24 seconds faster? wow. So, I removed the prints and got:

Benchmark: timing 100 iterations of Ase, Maverick...
       Ase: 122 wallclock secs (121.11 usr +  0.00 sys = 121.11 CPU)
  Maverick: 164 wallclock secs (164.65 usr +  0.03 sys = 164.68 CPU)
[download]

I'm running this on a PII 400, using Red Hat Linux and the stock perl RPM (5.005_03). It would seem that vec is much faster under Linux than under windows.

Hmmm...I wonder how these benchmark out on other platforms...

/\/\averick