MCODE: random numbers

[Albert Chan]

(04-20-2019 10:40 PM)RobertM Wrote:  Any thoughts on the how to algorithmically implement the "modulo(n)" portion without a div instruction?

If you don't mind tiny non-biased random modulo, it can be done with a multiply and a shift.

see first version of random_bounded() in https://lemire.me/blog/2016/06/30/fast-r...shuffling/

Non-biased modulo is not much harder, see the second version.
This required some random number samples to be rejected.

Also, I had conversations with Lemire about how divisionless random code work.
On the last post, there is a link to a paper about it.