deltarho[1859] wrote:paul doe wrote:However, that's not mine, I think it's xorshift128+
Yes, it is although I used 23, 18 and 5 as recommended by Vigna. V8 JavaScript uses 23, 17 and 26. It fails PractRand at 64GB and, from what I have read, it fails Big Crush as well.
thanks!
I read that PCG isn't crypto-strength. And then learning of xorShift, read an article about reverse-engineering for prediction. Having only grazed the article
https://blog.securityevaluators.com/xor ... 3365dc0c17 i 'jumped' to several conclusions
1. you can only reverse engineer, to an extent (with today's hardware) if you know the original algorithm.
2. I don't yet know much about crypto, but cross-referencing my experience reading about reversible RNGs with what I read from the article, if i wanted to make a crypto RNG, I would begin by toying with the idea of non-reversibility.
[update]
i think I get it. non-reversibility isn't possible. At least with our current instruction set. Multiply is reversible via 'mod inverse' .. you'd need something non-linear .. logarithmic .. lol .. elliptical curve maybe? (i've heard the term)
this digs into the foundation of mathematics