Re: KISS 64-bit pseudo-random number generator

On Mon, 9 Sep 2024 8:55:14 +0000, Anton Ertl wrote:

1320>> : RANDOM        seed  $107465 *  $234567 +  ( -- u )
1321:                  9 ROL DUP TO seed ;

>
So this is a linear congruential generator enhanded with the 9 ROL.
LCGs have known weaknesses that are relevant even for
non-cryptographic applications.  Maybe the ROL fixes those; have you
run it through ransomness testers?

The ROL fixes the problem that the lower bits are "less random" than
the higher ones (which was npticeable in my graphics applications). With
9 ROL the generator passes Marsaglia's DIEHARD tests (I showed these
in an ancient post). However, I see a comment in the source that
suggests it did not pass a better test than DIEHARD, so that is why
there are RANF, RAN-NEXT, KISS, pseudo-DES, ran0, ran1, ran2, ran3,
random3, wurst-rng, isaac, lehmer, mersenne-twister, and some I can't
find right now.
-marcel