kmeans 2.0.2

Small and fast library for k-means clustering calculations.
Documentation 
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
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229

use crate::helpers;
use core::fmt;
use rand::distributions::uniform::SampleUniform;
use std::ops::{Index, IndexMut};
use std::simd::num::SimdFloat;
use std::simd::{Simd, SimdElement};
use std::{iter, ops};

pub trait Primitive:
    'static
    + SimdElement
    + SampleUniform
    + Default
    + fmt::Display
    + fmt::Debug
    + fmt::LowerExp
    + Send
    + Sync
    + iter::Sum
    + num::Float
    + ops::Add<Output = Self>
    + ops::Sub<Output = Self>
    + ops::Mul<Output = Self>
    + ops::Div<Output = Self>
    + ops::AddAssign
    + ops::SubAssign
    + for<'a> ops::AddAssign<&'a Self>
    + for<'a> ops::Sub<&'a Self>
{
}

impl<T> Primitive for T where
    T: 'static
        + SimdElement
        + SampleUniform
        + Default
        + fmt::Display
        + fmt::Debug
        + fmt::LowerExp
        + Send
        + Sync
        + iter::Sum
        + num::Float
        + ops::Add<Output = Self>
        + ops::Sub<Output = Self>
        + ops::Mul<Output = Self>
        + ops::Div<Output = Self>
        + ops::AddAssign
        + ops::SubAssign
        + for<'a> ops::AddAssign<&'a Self>
        + for<'a> ops::Sub<&'a Self>
{
}

// ##################################################################

pub trait SupportedSimdArray<T: Primitive, const LANES: usize>:
    ops::Sub<Output = Self> + ops::Add<Output = Self> + ops::Mul<Output = Self> + ops::Div<Output = Self> + iter::Sum + SimdFloat<Scalar = T>
{
}

impl<T: Primitive, const LANES: usize> SupportedSimdArray<T, LANES> for Simd<T, LANES> where
    Simd<T, LANES>: ops::Sub<Output = Simd<T, LANES>>
        + ops::Add<Output = Simd<T, LANES>>
        + ops::Mul<Output = Simd<T, LANES>>
        + ops::Div<Output = Simd<T, LANES>>
        + iter::Sum
        + SimdFloat<Scalar = T>
{
}

// ##################################################################

pub(crate) type SIMDAlignBuffer<T> = aligned_vec::AVec<T, aligned_vec::RuntimeAlign>;
pub(crate) struct SIMDAlignBufferHelper;
impl SIMDAlignBufferHelper {
    pub fn create<T: Primitive, const LANES: usize>(size: usize) -> SIMDAlignBuffer<T> {
        aligned_vec::avec_rt!([ LANES * std::mem::size_of::<T>() ]| Default::default(); size)
    }
}

// ##################################################################

/// Buffer storing strided multi-dimensional elements, like centroids.
///
/// # Details
/// kmeans internally uses SIMD instructions. To do that efficiently, it is easiest
/// to pad all handled data to multiples of the used LANES size. This datastructure
/// internally handles what is padding and what is not, and allows easy access to
/// the contained (padded) elements for users.
#[derive(Clone, Debug)]
pub struct StrideBuffer<T> {
    pub(crate) bfr: SIMDAlignBuffer<T>,
    pub(crate) stride: usize,
    pub(crate) centroid_cnt: usize,
    pub(crate) centroid_dim: usize,
}
impl<T: Primitive> StrideBuffer<T> {
    pub fn new<const LANES: usize>(centroid_cnt: usize, centroid_dim: usize) -> Self {
        let stride = helpers::multiple_roundup(centroid_dim, LANES);
        Self {
            bfr: SIMDAlignBufferHelper::create::<T, LANES>(stride * centroid_cnt),
            stride,
            centroid_cnt,
            centroid_dim,
        }
    }

    pub fn from_slice<const LANES: usize>(centroid_dim: usize, vec: &[T]) -> Self {
        assert_eq!(vec.len() % centroid_dim, 0);
        let stride = helpers::multiple_roundup(centroid_dim, LANES);
        let centroid_cnt = vec.len() / centroid_dim;

        let mut bfr = Self {
            bfr: SIMDAlignBufferHelper::create::<T, LANES>(stride * centroid_cnt),
            stride,
            centroid_cnt,
            centroid_dim,
        };
        for i in 0..centroid_cnt {
            bfr.set_nth_from_iter(i, vec[i * centroid_dim..].iter().cloned());
        }
        bfr
    }

    /// Convert the centroids contained in this CentroidBuffer to a tightly packed
    /// Vec with the memory layout `[c0_0, c0_1, ... c0_n, c1_0, c1_1, ... c1_n, ...]`
    /// where `cx_y` is the `y`'th dimension of the `x`'th centroid.
    pub fn to_vec(&self) -> Vec<T> {
        let mut res = Vec::with_capacity(self.centroid_cnt * self.centroid_dim);
        res.extend(self.iter().flat_map(|c| c.iter()));
        res
    }

    /// Set the nth stride from the given iterator
    pub fn set_nth_from_iter(&mut self, index: usize, iter: impl IntoIterator<Item = T>) {
        self.index_mut(index).iter_mut().zip(iter).for_each(|(dst, src)| {
            *dst = src;
        });
    }

    /// Iterate over all contained centroids. Every item in the iterator
    /// corresponds to one centroid.
    pub fn iter(&self) -> impl Iterator<Item = &[T]> { self.bfr.chunks_exact(self.stride).map(|c| &c[..self.centroid_dim]) }

    /// Iterate over all contained centroids. Every item in the iterator
    /// corresponds to one centroid.
    pub fn iter_mut(&mut self) -> impl Iterator<Item = &mut [T]> {
        self.bfr.chunks_exact_mut(self.stride).map(|c| &mut c[..self.centroid_dim])
    }

    // #############################################
    // INTERNAL

    #[inline(always)]
    pub(crate) fn nth_stride(&self, index: usize) -> &[T] {
        let offset = index * self.stride;
        &self.bfr[offset..offset + self.stride]
    }

    #[inline(always)]
    pub(crate) fn nth_stride_mut(&mut self, index: usize) -> &mut [T] {
        let offset = index * self.stride;
        &mut self.bfr[offset..offset + self.stride]
    }

    #[inline(always)]
    pub(crate) fn chunks_exact_stride(&self) -> impl Iterator<Item = &[T]> { self.bfr.chunks_exact(self.stride) }

    #[inline(always)]
    pub(crate) fn chunks_exact_stride_mut(&mut self) -> impl Iterator<Item = &mut [T]> { self.bfr.chunks_exact_mut(self.stride) }
}
impl<T> Index<usize> for StrideBuffer<T> {
    type Output = [T];

    #[inline(always)]
    fn index(&self, index: usize) -> &Self::Output {
        let offset = index * self.stride;
        &self.bfr[offset..offset + self.centroid_dim]
    }
}
impl<T> IndexMut<usize> for StrideBuffer<T> {
    #[inline(always)]
    fn index_mut(&mut self, index: usize) -> &mut Self::Output {
        let offset = index * self.stride;
        &mut self.bfr[offset..offset + self.centroid_dim]
    }
}

#[cfg(test)]
mod tests {
    use super::StrideBuffer;

    #[test]
    fn stride_buffer() {
        for centroid_dim in 1..8 {
            let bfr: StrideBuffer<f32> = StrideBuffer::new::<8>(5, centroid_dim);
            assert_eq!(bfr.stride, 8);
            assert_eq!(bfr.bfr.alignment(), 8 * std::mem::size_of::<f32>());
            assert_eq!(bfr.centroid_dim, centroid_dim);
            assert_eq!(bfr.centroid_cnt, 5);
        }

        {
            let mut bfr: StrideBuffer<f64> = StrideBuffer::new::<8>(3, 9);
            assert_eq!(bfr.stride, 16);
            assert_eq!(bfr.bfr.alignment(), 8 * std::mem::size_of::<f64>());
            assert_eq!(bfr.centroid_dim, 9);
            assert_eq!(bfr.centroid_cnt, 3);
            bfr.set_nth_from_iter(1, [1.0, 2.0, 3.0, 4.0, 5.0, 6.0, 7.0, 8.0, 9.0, 10.0]);
            assert_eq!(bfr.bfr.to_vec(), vec![
                0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 2.0, 3.0, 4.0, 5.0, 6.0, 7.0, 8.0,
                9.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0,
            ]);
        }

        {
            let src_bfr = vec![1.0, 2.0, 3.0, 4.0, 5.0, 6.0, 7.0, 8.0, 9.0, 10.0, 11.0, 12.0];
            let bfr: StrideBuffer<f32> = StrideBuffer::from_slice::<4>(3, &src_bfr);
            assert_eq!(bfr.centroid_cnt, 4);
            assert_eq!(bfr.centroid_dim, 3);
            assert_eq!(bfr.stride, 4);
            assert_eq!(bfr.bfr.to_vec(), vec![
                1.0, 2.0, 3.0, 0.0, 4.0, 5.0, 6.0, 0.0, 7.0, 8.0, 9.0, 0.0, 10.0, 11.0, 12.0, 0.0,
            ]);
            assert_eq!(bfr.to_vec(), src_bfr);
        }
    }
}