simdeez 3.0.1

SIMD library to abstract over different instruction sets and widths
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

use crate::prelude::*;

fn reference_adaptive_select_i32(lhs: &[i32], rhs: &[i32]) -> Vec<i32> {
    lhs.iter()
        .zip(rhs.iter())
        .map(|(&left, &right)| {
            let hi = left.max(right);
            let lo = left.min(right);
            let spread = hi - lo;
            let expanded = hi.min(4089) + 7;
            let compressed = lo.max(-4089) - 7;
            if spread > 1024 {
                expanded
            } else {
                compressed
            }
        })
        .collect()
}

simd_unsafe_generate_all!(
    fn adaptive_select_i32(lhs: &[i32], rhs: &[i32], output: &mut [i32]) {
        assert_eq!(lhs.len(), rhs.len());
        assert_eq!(lhs.len(), output.len());

        let mut lhs = lhs;
        let mut rhs = rhs;
        let mut output = output;
        let threshold = S::Vi32::set1(1024);
        let upper_pre_add = S::Vi32::set1(4089);
        let lower_pre_sub = S::Vi32::set1(-4089);

        while lhs.len() >= S::Vi32::WIDTH {
            let left = S::Vi32::load_from_slice(lhs);
            let right = S::Vi32::load_from_slice(rhs);
            let hi = left.max(right);
            let lo = left.min(right);
            let spread = hi - lo;
            let expanded = hi.min(upper_pre_add) + 7;
            let compressed = lo.max(lower_pre_sub) - 7;
            let selected = spread.cmp_gt(threshold).blendv(compressed, expanded);
            selected.copy_to_slice(output);

            lhs = &lhs[S::Vi32::WIDTH..];
            rhs = &rhs[S::Vi32::WIDTH..];
            output = &mut output[S::Vi32::WIDTH..];
        }

        for ((&left, &right), slot) in lhs.iter().zip(rhs.iter()).zip(output.iter_mut()) {
            let hi = left.max(right);
            let lo = left.min(right);
            let spread = hi - lo;
            let expanded = hi.min(4089) + 7;
            let compressed = lo.max(-4089) - 7;
            *slot = if spread > 1024 { expanded } else { compressed };
        }
    }
);

fn assert_adaptive_select_matches_all_backends(lhs: &[i32], rhs: &[i32]) {
    let expected = reference_adaptive_select_i32(lhs, rhs);

    let mut scalar = vec![0; lhs.len()];
    adaptive_select_i32(lhs, rhs, &mut scalar);
    assert_eq!(scalar, expected);

    #[cfg(any(target_arch = "x86_64", target_arch = "x86"))]
    {
        if std::arch::is_x86_feature_detected!("sse2") {
            let mut output = vec![0; lhs.len()];
            unsafe { adaptive_select_i32_sse2(lhs, rhs, &mut output) };
            assert_eq!(output, expected);
        }
        if std::arch::is_x86_feature_detected!("sse4.1") {
            let mut output = vec![0; lhs.len()];
            unsafe { adaptive_select_i32_sse41(lhs, rhs, &mut output) };
            assert_eq!(output, expected);
        }
        if std::arch::is_x86_feature_detected!("avx2") && std::arch::is_x86_feature_detected!("fma")
        {
            let mut output = vec![0; lhs.len()];
            unsafe { adaptive_select_i32_avx2(lhs, rhs, &mut output) };
            assert_eq!(output, expected);
        }
    }
}

// This resembles integer decision stages in codecs and DSP control paths where a fast per-lane
// spread check chooses between compressing noise and preserving strong transients.
#[test]
fn real_world_adaptive_select_i32_matches_reference() {
    let lhs = [
        -5000, -4096, -4090, -2048, -1025, -1024, -17, -1, 0, 1, 9, 17, 1023, 1024, 2047, 4090,
        4096, 5000,
    ]
    .into_iter()
    .cycle()
    .take(61)
    .enumerate()
    .map(|(i, value)| value + ((i % 7) as i32 * 3 - 9))
    .collect::<Vec<_>>();
    let rhs = lhs
        .iter()
        .enumerate()
        .map(|(i, &value)| {
            let delta = match i % 6 {
                0 => -1500,
                1 => -1024,
                2 => -17,
                3 => 17,
                4 => 1024,
                _ => 1500,
            };
            value.wrapping_neg().clamp(-5000, 5000) + delta
        })
        .collect::<Vec<_>>();

    assert_adaptive_select_matches_all_backends(&[], &[]);
    assert_adaptive_select_matches_all_backends(&lhs[..5], &rhs[..5]);
    assert_adaptive_select_matches_all_backends(&lhs[..19], &rhs[..19]);
    assert_adaptive_select_matches_all_backends(&lhs, &rhs);
}

#[test]
fn real_world_adaptive_select_i32_threshold_and_clamp_boundaries_match_reference() {
    let lhs = vec![
        i32::MIN + 2048,
        i32::MIN + 1025,
        -4096,
        -4090,
        -1,
        0,
        4089,
        4090,
        i32::MAX - 1025,
        i32::MAX - 2048,
    ];
    let rhs = vec![
        i32::MIN + 1024,
        i32::MIN + 1,
        -3072,
        -3065,
        1023,
        1025,
        3065,
        3066,
        i32::MAX - 1,
        i32::MAX - 1024,
    ];

    assert_adaptive_select_matches_all_backends(&lhs, &rhs);
}