TheIronBorn

Doesn't work out for me but maybe for newer machines: https://godbolt.org/g/Wcd9KH ``` test raw_lfsr113 ... bench: 87,112 ns/iter (+/- 55,554) = 752 MB/s test raw_lfsr113_new ... bench: 110,710 ns/iter (+/-...

Here's my benchmarks: ``` test raw_lfsr113_new ... bench: 186,136 ns/iter (+/- 602,228) = 704 MB/s test raw_lfsr113_new2 ... bench: 231,368 ns/iter (+/- 69,544) = 566 MB/s

We've not yet begun considering what PRNG to use for rand's SIMD generation (we may not even end up supporting SIMD at all). My early research suggests SFC in packed...

I'd be interested to see what sort of changes that would require.

You may want to replace your `xoroshiro128+` comparison with the newer `xoroshiro128**` or `xoshiro256**` as they obsoleted `xoroshiro128+`: http://xoshiro.di.unimi.it.

What specifications did you use for `++` generators? I believe the paper only mentions specifications for a `xoroshiro32++` generator.

Also something went wrong with formatting so I can't quite tell if you tested the `**` generators but they should do much better in PractRand

That's odd, my testing has it ~~>16GB~~ **Edit:** >512GB.

Note that's it's not only naming that's incorrect. [`encode.avx512vl.cpp`](https://github.com/WojciechMula/base64simd/blob/6bde46e13d42f962620c0c496f0e5b288043ff5d/encode/encode.avx512vl.cpp) uses AVX512VBMI (`vpermb/vpmultishiftqb`) yet not all CPUs with AVX512VL have AVX512VBMI ([chart](https://en.wikipedia.org/wiki/AVX-512#CPUs_with_AVX-512)). It actually uses no AVX512VL instructions at all (according...

This seems to be related to this note on the Go website https://go.dev/doc/faq#virus