我测了一次performance，发现u32/u32的速度竟然比f64/f64慢

有没有dalao知道发生了什么吗？

extern crate rayon;
use rayon::prelude::*;
fn main(){
    let b=std::time::Instant::now();
    let r1=(0u32..65536).into_par_iter().chunks(4096).map(|x|{
        x.iter().map(|&i|{
            (1u32..65536).into_iter().map(|j|{
                i/j
            }).sum::<u32>() as usize
        }).sum::<usize>()
    }).sum::<usize>();
    println!("{:?}",b.elapsed());
    let b=std::time::Instant::now();
    let a:Vec<usize>=(0..1).chain((1..2097152).map(|x|((4398046511104_u64+x-1)/x as u64) as usize)).collect();
    let r2=(0usize..65536).into_par_iter().chunks(4096).map(|x|{
        x.iter().map(|&i|{
            (1usize..65536).into_iter().map(|j|{
                i*a[j]>>42
            }).sum::<usize>()
        }).sum::<usize>()
    }).sum::<usize>();
    println!("{:?}",b.elapsed());
    let b=std::time::Instant::now();
    let r3=(0u32..65536).into_par_iter().chunks(4096).map(|x|{
        x.iter().map(|&i|{
            (1u32..65536).into_iter().map(|j|{
                (i as f64/j as f64) as u32
            }).sum::<u32>() as usize
        }).sum::<usize>()
    }).sum::<usize>();
    println!("{:?}",b.elapsed());
    println!("{}",(r1+r2)>>1-r3);
}

测试结果

neutron@Neutron:/me/rust/rayonops$ cargo run --release
   Compiling rayonops v0.1.0 (/me/rust/rayonops)
    Finished release [optimized] target(s) in 0.86s
     Running `target/release/rayonops`
1.739315795s
735.013066ms
957.63335ms
0
neutron@Neutron:/me/rust/rayonops$ cargo run --release
    Finished release [optimized] target(s) in 0.01s
     Running `target/release/rayonops`
1.77189577s
900.427822ms
1.013179198s
0
neutron@Neutron:/me/rust/rayonops$ cargo run --release
    Finished release [optimized] target(s) in 0.01s
     Running `target/release/rayonops`
1.778011042s
859.130591ms
1.003195896s
0
neutron@Neutron:/me/rust/rayonops$ cargo run --release
    Finished release [optimized] target(s) in 0.01s
     Running `target/release/rayonops`
1.53730607s
767.199922ms
1.048112309s
0

并不是rayon的锅：

//不带rayon
fn main(){
    let b=std::time::Instant::now();
    let r1=(0u32..65536).into_iter().map(|i|{
            (1u32..65536).into_iter().map(|j|{
                i/j
            }).sum::<u32>() as usize
    }).sum::<usize>();
    println!("{:?}",b.elapsed());
    let b=std::time::Instant::now();
    let a:Vec<usize>=(0..1).chain((1..2097152).map(|x|((4398046511104_u64+x-1)/x as u64) as usize)).collect();
    let r2=(0usize..65536).into_iter().map(|i|{
            (1usize..65536).into_iter().map(|j|{
                i*a[j]>>42
            }).sum::<usize>()
    }).sum::<usize>();
    println!("{:?}",b.elapsed());
    let b=std::time::Instant::now();
    let r3=(0u32..65536).into_iter().map(|i|{
            (1u32..65536).into_iter().map(|j|{
                (i as f64/j as f64) as u32
            }).sum::<u32>() as usize
    }).sum::<usize>();
    println!("{:?}",b.elapsed());
    println!("{}",(r1+r2)>>1-r3);
}

测试结果

neutron@Neutron:/me/rust/rayonops$ ./main_speed_test 
10.20379502s
3.601069229s
6.565324578s
0