`Exp1` is slower than `Exp`

[vks]

Running the corresponding benchmarks from #1063, I get the following results:

test distr_exp1_general                    ... bench:       4,208 ns/iter (+/- 192) = 1901 MB/s
test distr_exp1_specialized                ... bench:       5,288 ns/iter (+/- 196) = 1512 MB/s

This contradicts our documentation:

Samples floating-point numbers according to the exponential distribution, with rate parameter λ = 1. This is equivalent to Exp::new(1.0) or sampling with -rng.gen::<f64>().ln(), but faster.

[dhardy]

So the doc is wrong (likely depending on optimisations), but as I mentioned in the benchmark PR, there may still be reason to use this: saving space in a derived distribution.

[vks]

I cannot reproduce this on a more modern CPU, therefore I'll close this.

test distr_exp1_general                    ... bench:       2,868 ns/iter (+/- 56) = 2789 MB/s
test distr_exp1_specialized                ... bench:       2,633 ns/iter (+/- 74) = 3038 MB/s

[dhardy]

It could also be changes in the compiler. This is still a Haswell CPU:

test distr_exp1_general                    ... bench:       3,868 ns/iter (+/- 352) = 2068 MB/s
test distr_exp1_specialized                ... bench:       3,691 ns/iter (+/- 189) = 2167 MB/s