Deterministic Rayon monte carlo example

[masonk]

It wasn't straightforward to understand how to use Rayon in conjunction with Rand to construct a monte-carlo simulation that

1. Given the same seed, always produces the same output
2. Is efficient. In particular, does not require creating a new RNG per work item.

Spawned from this discussion: #399

[masonk]

Sadly, the ST version of this program using ChaCha8 runs in 686ms, whereas the MT version calling set_word_pos runs in 480ms. Hardly an impressive speedup on my 16-core machine.

For comparison, the MT version without set_word_pos runs in 50ms, giving me a 14x speedup factor.

set_word_pos increases runtime by about 10x.

[dhardy]

This example doesn't need to be deterministic — in fact, you don't

[dhardy]

Mis-click.

...you don't even discuss determinism here — a little comment would be good.

[masonk]

It looks like there's a problem with simd on unbuntu-latest. Could it be due to the rayon dep in dev-dependencies, or is that already failing?

[dhardy]

The problem is that packed_simd_2 breaks periodically since it uses nightly features.

Can you bump its version to the latest (0.3.8)?

[dhardy]

Excellent documentation, thanks!