Implemented XoShiro128 psuedo random number generator.

[thelearnerofcode]

Still have to add tests and documentation, but this is a good start.

See here for more info: http://xoshiro.di.unimi.it/.

[pitdicker]

Hi @thelearnerofcode. Thank you for the PR!

We have plans to move all specific PRNG implementations out of Rand, so I don't think adding any xoshiro variants is what we want at the moment. And we have a couple of other ones higher up the list in terms of priority, which are already postponed until after Rand 0.5 gets released.

I have not read the details of this new RNG yet, it is only released very recently. Maybe it should prove itself for a couple of months. And I am curious how it holds up under TestU01 and PractRand, where Xoroshiro128+ performed worse than what was advertised.

[thelearnerofcode]

Ok! Should I close the PR?

[pitdicker]

Maybe for now, but you can also leave it open for some discussion. I am sure @vks is interested, he mentioned the same RNG yesterday in dhardy#60 (comment), and is the author of the xoroshiro crate, which implements more work by Vigna.

[vks]

I worked on a crate for xoshiro128** yesterday. It can be partially vectorized with SIMD, resulting in a nice speedup.

[dhardy]

Cool! Sounds to me like we should start discussing how we want to present extra RNGs to users (dhardy#58) soon (in fact, if someone wants to open a new issue with a proposal, go ahead; I think companion crates are probably the best option but don't know how many). However, I don't think we should merge anything before the 0.5 release (very soon, hopefully).

This PR looks very nice but I don't think we should merge new PRNGs before sorting out the above. Also since this is part of a family of new PRNGs I don't know if we should pick names like Xoshiro128SS.

PS nice SIMD work @vks. But how are we supposed to differentiate between [u64; 2] and [u32; 4] versions of Xoshiro128?

[pitdicker]

Sounds to me like we should start discussing how we want to present extra RNGs to users (dhardy#58) soon

Opened #431.

[vks]

@dhardy Not sure what you mean. There is only one xoshiro128** and xoshiro128+ version, both use [u32; 4]. Only xoroshiro128 uses [u64; 2].

[vks]

This PR does not implement xoshiro128** as claimed by the code, but rather xoroshiro128**!

[dhardy]

Lets close this for now. We don't want it in its current form, and if we do decide to add this PRNG we can refer back to this PR.