1
  2
  3
  4
  5
  6
  7
  8
  9
 10
 11
 12
 13
 14
 15
 16
 17
 18
 19
 20
 21
 22
 23
 24
 25
 26
 27
 28
 29
 30
 31
 32
 33
 34
 35
 36
 37
 38
 39
 40
 41
 42
 43
 44
 45
 46
 47
 48
 49
 50
 51
 52
 53
 54
 55
 56
 57
 58
 59
 60
 61
 62
 63
 64
 65
 66
 67
 68
 69
 70
 71
 72
 73
 74
 75
 76
 77
 78
 79
 80
 81
 82
 83
 84
 85
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155

use crate::Constellation;
use bytemuck::cast;
use crossbeam_channel::bounded;
use nalgebra::{
    allocator::Allocator, distance, DefaultAllocator, DimName, NamedDim, Point, VectorN,
};
use rayon::prelude::*;
use simba::simd::{SimdValue, WideF32x4};
use std::sync::{Arc, RwLock};

pub type Point32<DimX> = Point<WideF32x4, DimX>;

fn make_point<DimX>(point: Vec<f32>) -> Point32<DimX::Name>
where
    DimX: NamedDim,
    DefaultAllocator: Allocator<WideF32x4, DimX::Name>,
{
    let wide_vec: Vec<WideF32x4> = point
        .chunks(4)
        .map(|c| WideF32x4::from(<[f32; 4]>::from([c[0], c[1], c[2], c[3]])))
        .collect();
    VectorN::<WideF32x4, DimX::Name>::from_vec(wide_vec).into()
}

/// A constellation contains lots of points.
pub struct VecSIMDConstellation<DimX>
where
    DimX: NamedDim,
    DefaultAllocator: Allocator<WideF32x4, DimX::Name>,
{
    points: Arc<RwLock<Vec<Point32<DimX::Name>>>>,
}

impl<DimX> Default for VecSIMDConstellation<DimX>
where
    DimX: NamedDim,
    DefaultAllocator: Allocator<WideF32x4, DimX::Name>,
{
    fn default() -> Self {
        VecSIMDConstellation {
            points: Arc::new(RwLock::new(Vec::new())),
        }
    }
}

impl<DimX> Constellation for VecSIMDConstellation<DimX>
where
    DimX: NamedDim + Sync,
    DefaultAllocator: Allocator<WideF32x4, DimX::Name>,
    <DefaultAllocator as Allocator<WideF32x4, DimX::Name>>::Buffer: Send + Sync,
{
    fn add_points(&self, points: Vec<Vec<f32>>) {
        self.points
            .clone()
            .write()
            .unwrap()
            .extend(points.into_iter().map(|p| make_point::<DimX>(p)))
    }

    fn find(&self, point: Vec<f32>, within: f32) -> Box<dyn Iterator<Item = (f32, Vec<f32>)>> {
        let point = make_point::<DimX>(point);
        let (tx, rx) = bounded(100);
        let points = self.points.clone();

        std::thread::Builder::new()
            .name("find_iterate".to_string())
            .spawn(move || {
                points
                    .read()
                    .unwrap()
                    .par_iter()
                    .try_for_each_with(tx.clone(), |tx, p| {
                        let result = distance(&point, &p);
                        let dist: f32 = cast::<_, [f32; 4]>(result.0).iter().sum::<f32>().sqrt();
                        if dist <= within {
                            // This seems absolutely horrible. Is there really not a better way?
                            let flat_coords: Vec<f32> = p
                                .coords
                                .iter()
                                .map(|p| cast::<_, [f32; 4]>(p.0).to_vec())
                                .flatten()
                                .collect();
                            return tx.send((dist, flat_coords));
                        }
                        Ok(())
                    })
                    .ok();
                // This is really important. Without this line there are sporadic stack overflows
                // with the benchmark - this thread doesn't terminate fast enough after `par_iter()`
                // finishes, and threads pile up.
                std::mem::drop(tx);
            })
            .unwrap();
        return Box::new(rx.into_iter());
    }

    fn count(&self) -> usize {
        self.points.read().unwrap().len()
    }

    fn dimensions(&self) -> usize {
        DimX::Name::dim() * WideF32x4::lanes()
    }

    fn memory_size(&self) -> usize {
        std::mem::size_of::<Point32<DimX::Name>>() * self.count()
    }
}

#[cfg(test)]
mod tests {
    // Note this useful idiom: importing names from outer (for mod tests) scope.
    use super::*;
    use typenum::{U1, U2};

    #[test]
    fn test_len() {
        let constellation = VecSIMDConstellation::<U1>::default();
        assert_eq!(constellation.count(), 0);
    }

    #[test]
    fn test_mem_size() {
        let constellation1 = VecSIMDConstellation::<U2>::default();
        constellation1.add_points(vec![vec![1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0]]);
        // Should be exactly 32 bytes
        assert_eq!(constellation1.memory_size(), 32);
    }

    #[test]
    fn test_add_multiple() {
        let constellation = VecSIMDConstellation::<U1>::default();
        let points: Vec<_> = vec![vec![1.0, 1.0, 1.0, 1.0], vec![1.0, 1.0, 1.0, 1.0]];
        constellation.add_points(points);
        assert_eq!(constellation.count(), 2);
    }

    #[test]
    fn test_query() {
        let constellation = VecSIMDConstellation::<U1>::default();
        constellation.add_points(vec![vec![2.0, 2.0, 2.0, 2.0]]);
        let iterator = constellation.find(vec![1.0, 1.0, 1.0, 1.0], 10.);
        let items: Vec<(f32, Vec<f32>)> = iterator.collect();
        assert_eq!(items, vec![(2.0, vec![2.0, 2.0, 2.0, 2.0])]);
    }

    #[test]
    fn test_query_missing() {
        let constellation = VecSIMDConstellation::<U1>::default();
        constellation.add_points(vec![vec![2., 2., 2., 2.]]);
        let iterator = constellation.find(vec![1., 1., 1., 1.], 0.99);
        let items: Vec<(f32, Vec<f32>)> = iterator.collect();
        assert_eq!(items, vec![]);
    }
}