In the message that I posted above, I'd known that since the intermediate values would be computed modulo a value involving maybe twice as many bits, each squared value would be perhaps four times as large as a ordinary LL sequence value.
Another detraction that didn't occur to me so quickly is that the modulo step would no longer be involving a value one less than a power of two. So it might require a true divide rather than anything like convenient method currently used.
Last fiddled with by only_human on 20030929 at 14:59
