Linnorm benchmarks

Linnorm.java

public final class Linnorm {
    private static long state = 0L;
    private static long random()
    {
        long z = (state = state * 0x41C64E6DL + 1L);
        z = (z ^ z >>> 28) * 0xAEF17502108EF2D9L;
        return z ^ z >>> 30;
    }
    public static void main(String[] args) {
        long out = 0L;
        for (int i = 0; i < 1000000000; i++) {
            out += random(); // Java is inlining this; manually inlining slows it down
        }
        final long startTime = System.nanoTime();
        for (int a = 0; a < 20; a++) {
            for (int i = 0; i < 1000000000; i++) {
                out += random();
            }
        }
        System.out.println(System.nanoTime() - startTime);
        System.out.println(out); // may be negative, Rust uses an unsigned type so convert before checking equality
    }
}

main.rs

extern crate time;
use std::num::Wrapping as W;

#[inline]
fn linnorm(state: &mut W<u64>) -> W<u64> {
    *state = *state * W(0x41C64E6Du64) + W(1u64);
    let mut z = *state;
    z = (z ^ z >> 28) * W(0xAEF17502108EF2D9u64);
    z ^ z >> 30
}

fn main() {
    let mut state: W<u64> = W(0u64);
    let mut out: W<u64> = W(0u64);
    let t = time::precise_time_ns();
    for _ in 0..1_000_000_000 {
        out += linnorm(&mut state);
    }
    for _ in 0..20 {
        for _ in 0..1_000_000_000 {
            out += linnorm(&mut state);
        }
    }
    println!("{}", time::precise_time_ns() - t);
    println!("{}", out.0);
}