Re: improve strlen

I had a quick play with the second algo on the AMD and changing the
unroll count back to 3 made it faster. On the PIV the magic unroll rate
was 3, 7 or 15 where on the AMD it seems to be best at 3. As a
production algo 3 appears to be the best across the two typical forms
of current hardware so that is the direction I will go with it.

With tongue in cheek as usual, welcome to the world of mixed model
cross hardware development. :)

Regards,

hutch at movsd dot com

.