View Single Post
Old 2019-01-21, 14:58   #25
Till's Avatar
"Tilman Neumann"
Jan 2016

22·109 Posts

Originally Posted by bsquared View Post
Yes, I did, and unfortunately I didn't see any speed improvement. Well, possibly there was a slight improvement at sizes > 50 bits, past the crossover point with rho.

You would see just that if you test the Lehman with a dataset of semiprimes N with smaller factor > cbrt(N) and parameter doTDivFirst=false ;-) The minimal improvement at bigger sizes would result from the few cases where the "correction loop" is needed, which we put behind the (second) trial division loop.

Anyway, thanks a lot for cross-checking our stuff in C, that is very helpful!
Till is offline   Reply With Quote