Xoroshiro128

[Chris Rutz]

Seba,

Regarding xoroshiro128 states 0 and 1, I cannot recall anything specific about the difference in randomness between them.

Obviously you have biased more toward the use of state[0] with ** and ++, but is state[1] noticeably less random (e.g., linear complexity, etc.), and if so, by approximately to what extent?

Thanks,

Chris

[Sebastiano Vigna]

Frankly, not. We tested it and it works very well. Since it's const it should be faster.

Ciao,

seba

[Chris Rutz]

Thanks for the reply.

Since you brought up the speed of s0, I am curious if you know of an example where a C compiler will actually use const as a hint in that specific context to generate different code?

The asm dumps I have looked at from all recent gcc and .NET versions (using godbolt.org) show no difference in emitted code with/without 'const' in xoroshiro, and benchmark results also seem to be the same.

Notes I've read on const usage in that context suggest it is still done for 'correctness' even though it has no actual effect.