Generating random numbers with `rand::rng()` is very slow

[phkeese]

Using this example:

use rand::Rng;

#[cfg(test)]
mod tests {
    use std::hint::black_box;
    use rand::Rng;

    fn generate_random(n: usize) {
        let mut rng = rand::rng();
        for i in 0..n {
            black_box(rng.random_bool(0.5));
        }
    }

    #[test] fn gen_10() { generate_random(10); }
    #[test] fn gen_100() { generate_random(100); }
    #[test] fn gen_1000() { generate_random(1000); }
    #[test] fn gen_10000() { generate_random(10_000); }
    #[test] fn gen_100000() { generate_random(100_000); }
}

to generate an increasing number of random bools takes much longer with Miri than without.
Running without Miri completes in less than a second, with Miri, even just 10000 takes multiple seconds, generating 100,000 does not seem to finish in 30 seconds.
Is there a way of speeding up random value generation?
I tried MIRIFLAGS="-Zmiri-disable-isolation" cargo +nightly miri test but it did not speed up the process.

[RalfJung]

The expected performance of Miri is around 100x-1000x slower than a debug build. This seems to be in that ballpark? I doubt this has anything to do with random numbers. The only way currently available to speed up Miri is to have it check fewer things, I am afraid.

[phkeese]

Thank you, I did some further testing, and it looks like it really is just a constant factor of ~5000 slower with Miri vs. without.
Looks like I will have to #[cfg_attr(miri, ignore)] my large random tests or use a smaller number of runs :)

[RalfJung]

5000x is slower than I expected, but not implausible, yeah. We have an ongoing project to get a better idea of where Miri spends its time, maybe something will come out of that.

What I usually do is to set the n in your case to something like if cfg!(miri) { 100 } else { 10_000 }.