Peter Galbavy <peter@wonderland.org> writes:

>
> Charles you are one of those people that cannot resist a challenge :)

Yah, well.  I confess to being somewhat annoyed that someone removed
all the work I did to optimize the previous version for 486s by using
Duff's device in the main loop.  It obviously wasn't tested very
thoroughly.

That said, I just found a way to make my own version 14-44% faster on
Pentiums (again, depending on the mbuf layout), but unfortunately it's
slightly slower on 486s, much slower on 386s, and probably about the
same on P6s (but I don't have any P6s to test on).  I suppose I could
include multiple routines and use a function pointer, or something.
Gag.