On Tuesday, 10 June 2014 at 21:02:54 UTC, bearophile wrote:
Sorry, the R250/521 idea and the strongly pure idea are unrelated to each other.
Ah, good. That makes things simpler. I'll implement R250/521 for you, though.
For the strongly pure random generator we should choose a generator with a small internal state (let's say less than 5 CPU words, they get passed by immutable value).
We might be able to rework the Xorshift generators in this way -- they all rely on a very small internal state. It'd be interesting to see if this has any speed implications.