kiddo 6.0.1

A high-performance, flexible, ergonomic k-d tree library. Ideal for geo- and astro- nearest-neighbour and k-nearest-neighbor queries
Documentation
use std::arch::x86_64::*;

use crate::dist::distance_metric_avx2::{Avx2F32LeafOps, Avx2F64LeafOps};

pub struct SquaredEuclideanAvx2F64LeafOps;

impl Avx2F64LeafOps for SquaredEuclideanAvx2F64LeafOps {
    #[inline(always)]
    unsafe fn dist_k0_f64x4(delta: __m256d) -> __m256d {
        _mm256_mul_pd(delta, delta)
    }

    #[inline(always)]
    unsafe fn dist_kn_f64x4(acc: __m256d, delta: __m256d) -> __m256d {
        _mm256_add_pd(acc, _mm256_mul_pd(delta, delta))
    }

    #[inline(always)]
    unsafe fn dist_k0_f64x2(delta: __m128d) -> __m128d {
        _mm_mul_pd(delta, delta)
    }

    #[inline(always)]
    unsafe fn dist_kn_f64x2(acc: __m128d, delta: __m128d) -> __m128d {
        _mm_add_pd(acc, _mm_mul_pd(delta, delta))
    }

    #[inline(always)]
    fn dist_k0_f64x1(delta: f64) -> f64 {
        delta * delta
    }

    #[inline(always)]
    fn dist_kn_f64x1(acc: f64, delta: f64) -> f64 {
        acc + delta * delta
    }
}

pub struct SquaredEuclideanAvx2F32LeafOps;

impl Avx2F32LeafOps for SquaredEuclideanAvx2F32LeafOps {
    #[inline(always)]
    unsafe fn dist_k0_f32x8(delta: __m256) -> __m256 {
        _mm256_mul_ps(delta, delta)
    }

    #[inline(always)]
    unsafe fn dist_kn_f32x8(acc: __m256, delta: __m256) -> __m256 {
        _mm256_add_ps(acc, _mm256_mul_ps(delta, delta))
    }

    #[inline(always)]
    unsafe fn dist_k0_f32x4(delta: __m128) -> __m128 {
        _mm_mul_ps(delta, delta)
    }

    #[inline(always)]
    unsafe fn dist_kn_f32x4(acc: __m128, delta: __m128) -> __m128 {
        _mm_add_ps(acc, _mm_mul_ps(delta, delta))
    }

    #[inline(always)]
    fn dist_k0_f32x1(delta: f32) -> f32 {
        delta * delta
    }

    #[inline(always)]
    fn dist_kn_f32x1(acc: f32, delta: f32) -> f32 {
        acc + delta * delta
    }
}