simd_aligned 0.6.1

Safe and fast SIMD-aligned data structures with easy and transparent 'flat' access.
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

use std::ops::{Deref, DerefMut, Index, IndexMut};

use crate::traits::Simd;

use super::{
    conversion::{simd_container_flat_slice, simd_container_flat_slice_mut},
    packed::PackedMxN,
};

/// A dynamic (heap allocated) vector aligned for fast and safe SIMD access that also provides a
/// flat view on its data.
///
/// # Example
///
/// ```rust
///
/// // Create a vector of f64x__ elements that, in total, will hold space
/// // for at least 4 f64 values. Internally this might be one f64x4, two f64x2,
/// // or one f64x8 where the 2nd half is hidden, depending on the current architecture.
/// use simd_aligned::{VecSimd, arch::f64x4};
///
/// let mut v = VecSimd::<f64x4>::with(0_f64, 4);
///
/// // Get a 'flat view' (&[f64]) into the SIMD vectors and fill it.
/// v.flat_mut().clone_from_slice(&[0.0, 1.0, 2.0, 3.0]);
/// ```

#[derive(Clone, Debug)]
pub struct VecSimd<T>
where
    T: Simd + Default + Clone,
{
    pub(crate) simd_rows: PackedMxN<T>,
}

impl<T> VecSimd<T>
where
    T: Simd + Default + Clone,
{
    /// Produce a [`VecSimd`] with the given element `t` as default and a flat size of `size`.
    #[inline]
    pub fn with(t: T::Element, size: usize) -> Self {
        Self {
            simd_rows: PackedMxN::with(T::splat(t), 1, size),
        }
    }

    /// Get a flat view for this [`VecSimd`].
    #[inline]
    #[must_use]
    pub fn flat(&self) -> &[T::Element] {
        simd_container_flat_slice(&self.simd_rows.data[..], self.simd_rows.row_length)
    }

    /// Get a flat, mutable view for this [`VecSimd`].
    #[inline]
    pub fn flat_mut(&mut self) -> &mut [T::Element] {
        simd_container_flat_slice_mut(&mut self.simd_rows.data[..], self.simd_rows.row_length)
    }
}

impl<T> Index<usize> for VecSimd<T>
where
    T: Simd + Default + Clone,
{
    type Output = T;

    #[inline]
    fn index(&self, index: usize) -> &Self::Output {
        &self.simd_rows.data[index]
    }
}

impl<T> IndexMut<usize> for VecSimd<T>
where
    T: Simd + Default + Clone,
{
    #[inline]
    fn index_mut(&mut self, index: usize) -> &mut Self::Output {
        &mut self.simd_rows.data[index]
    }
}

impl<T> Deref for VecSimd<T>
where
    T: Simd + Default + Clone,
{
    type Target = [T];

    fn deref(&self) -> &[T] {
        &self.simd_rows.data[..]
    }
}

impl<T> DerefMut for VecSimd<T>
where
    T: Simd + Default + Clone,
{
    fn deref_mut(&mut self) -> &mut [T] {
        &mut self.simd_rows.data[..]
    }
}

#[cfg(test)]
mod test {
    use super::VecSimd;
    use crate::arch::f32x4;

    #[test]
    fn allocation_size() {
        let v_1 = VecSimd::<f32x4>::with(0.0f32, 4);
        let v_2 = VecSimd::<f32x4>::with(0.0f32, 5);

        assert_eq!(v_1.simd_rows.data.len(), 1);
        assert_eq!(v_2.simd_rows.data.len(), 2);
    }

    #[test]
    fn flat() {
        let mut v = VecSimd::<f32x4>::with(10.0f32, 16);
        let r_m = v.flat_mut();

        assert_eq!(r_m.len(), 16);

        for x in r_m {
            *x = 1.0
        }

        let mut sum = 0.0;
        let r = v.flat();

        assert_eq!(r.len(), 16);

        for x in r {
            sum += x;
        }

        assert!((sum - 16.0).abs() <= std::f32::EPSILON);
    }

    #[test]
    fn deref() {
        let v = VecSimd::<f32x4>::with(0.0f32, 16);
        assert_eq!(&v[0], &v[0]);
    }
}