[survived]

Perf difference:

• Time it takes to do 10000 primality tests on random odd 1000-bits integers
  • Nbi: v0.3.2: 0.77-0.84s; m: 3.30-3.34s; this PR: 0.55-0.57s
  • Rug: v0.3.2: 0.32-0.34s; m: 0.32-0.34s; this PR: 0.09s
• Generate 1000-bits prime, 10 times
  • Nbi: v0.3.2: 0.25-0.30s; m: 0.84-1.68s; this PR: 0.19s
  • Rug: v0.3.2: 0.09-0.15s; m: 0.08s-0.12s; this PR: 0.08s
• Generate 1000-bits safe prime, 10 times (with the same RNG seed)
  • Nbi: v0.3.2: 35.86s; m: 185.94s; this PR: 23.63s
  • Rug: v0.3.2: 11.92s; m: 11.92s; this PR: 10.47s

Code used for benchmarks

#[test]
fn is_prime_many_times() {
    for _ in 0..10000 {
        let mut x = Integer::random_bits(1000, &mut rand_core::OsRng);
        x |= 1;
        let _ = std::hint::black_box(x.is_probably_prime(25, &mut rand_core::OsRng));
    }
}

#[test]
fn generate_prime() {
    let mut rng = rand_dev::DevRng::new();

    for _ in 0..10 {
        let _x = std::hint::black_box(Integer::generate_prime(&mut rng, 1000));
    }
}

// run with `export RUST_TESTS_SEED=c020d800c74e122683308880e800397634cbb3823e63aaf5045b9b52d537ef8c`
#[test]
fn generate_safe_prime() {
    let mut rng = rand_dev::DevRng::new();

    for _ in 0..10 {
        let _x = std::hint::black_box(Integer::generate_safe_prime(&mut rng, 1000));
    }
}

[maurges]

I have a suspicion that crypto-primes does a constant-time primality check