splitmix32a info about ranoise HOT 1 OPEN

Thanks for a bit of interesting info.

I'm not really sure what the meaning of the test result difference is for the splitmix32 variations. Several hash functions known to be better (from the hash prospector project did worse than splitmix32a when I tested them, and I had to try several increment constant variations before finding one that didn't give a disappointingly bad result for among the very best hash functions. It's possible that splitmix32a actually does a bit better because in some ways it's a bit worse. An ideal hash function is a perfect shuffle, an ideal PRNG function more messy in some ways.

That said, and more broadly (I'm thinking of all my own efforts as of now), from what I've read as a beginner in this area, these statistical tests are all a little limited in some ways, and results best taken with a grain of salt when it's about small differences which you're not sure of the cause for. Basically, "chance things" may influence the estimation of chance, and when tweaking a PRNG to get a better result while not really knowing what one is doing, it's not always clear when the limits tested in the process are more those of the PRNG or those of the test (though if better testing preserves a good result, it's probably the PRNG which is better).

from ranoise.