simd_itertools

Trait FilterSimd

Source 
pub trait FilterSimd<'a, T>
where
    T: PartialEq + Copy,
{
    // Required method
    fn filter_simd<F>(&self, f: F) -> SimdFilter<'a, T, F>
       where F: Fn(&T) -> bool + 'a;
}

Required Methods§

Source

fn filter_simd<F>(&self, f: F) -> SimdFilter<'a, T, F>
where F: Fn(&T) -> bool + 'a,

Dyn Compatibility§

This trait is not dyn compatible.

In older versions of Rust, dyn compatibility was called "object safety", so this trait is not object safe.

Implementations on Foreign Types§

Source§

impl<'a, T> FilterSimd<'a, T> for Iter<'a, T>
where T: PartialEq + Copy,

Source§

fn filter_simd<F>(&self, f: F) -> SimdFilter<'a, T, F>
where F: Fn(&T) -> bool + 'a,

This is the least optimal of all functions. current implementation relies on sparsity of elems.

This kind of pattern is fast: [0,0,0,0,0,0,0,0,0,0,1,1,0,1,1,0,0,0,0,0,0]

This kind of pattern is slow (similar to scalar speed): [1,0,1,0,1,0,1,0,1,0,1,0,1,0,1,0,1,0,1,0,1]

The speed comes from checking if a chunks contains any wanted element.

(0..10000).collect_vec().iter().filter_simd(|x| *x % 100 == 0).collect::<Vec<i32>>() is ~4x faster on x86 with avx2

(0..10000).collect_vec().iter().filter_simd(|x| *x % 10 == 0).collect::<Vec<i32>>() is ~2x faster on x86 with avx2

(0..10000).collect_vec().iter().filter_simd(|x| *x % 1 == 0).collect::<Vec<i32>>() is 30% slower than scalar on x86 with avx2

Something like this works well on all patterns on x86:

Implementors§