shape_runtime/simd_rolling/
clip.rs

1//! Clip operations (value clamping)
2
3use wide::f64x4;
4
5use super::SIMD_THRESHOLD;
6
7// ===== Public API - Feature-gated SIMD selection =====
8
9/// Clip values (auto-selects SIMD or scalar)
10#[cfg(feature = "simd")]
11#[inline]
12pub fn clip(data: &mut [f64], min: f64, max: f64) {
13    clip_simd(data, min, max)
14}
15
16#[cfg(not(feature = "simd"))]
17#[inline]
18pub fn clip(data: &mut [f64], min: f64, max: f64) {
19    clip_scalar(data, min, max)
20}
21
22// ===== Internal implementations =====
23
24/// SIMD clip: clamp values to [min, max] range
25fn clip_simd(data: &mut [f64], min: f64, max: f64) {
26    if data.len() < SIMD_THRESHOLD {
27        for val in data.iter_mut() {
28            *val = val.clamp(min, max);
29        }
30        return;
31    }
32
33    let min_vec = f64x4::splat(min);
34    let max_vec = f64x4::splat(max);
35
36    let chunks = data.len() / 4;
37    for chunk in 0..chunks {
38        let i = chunk * 4;
39        let vals = f64x4::new([data[i], data[i + 1], data[i + 2], data[i + 3]]);
40        let clamped = vals.max(min_vec).min(max_vec);
41        let arr = clamped.to_array();
42        data[i..i + 4].copy_from_slice(&arr);
43    }
44
45    // Remainder
46    for i in (chunks * 4)..data.len() {
47        data[i] = data[i].clamp(min, max);
48    }
49}
50
51/// Scalar clip (for non-SIMD builds)
52#[cfg(not(feature = "simd"))]
53fn clip_scalar(data: &mut [f64], min: f64, max: f64) {
54    for val in data.iter_mut() {
55        *val = val.clamp(min, max);
56    }
57}
58
59#[cfg(test)]
60mod tests {
61    use super::*;
62
63    #[test]
64    fn test_clip() {
65        let mut data = vec![-10.0, 5.0, 15.0, 20.0, 0.0];
66        clip(&mut data, 0.0, 10.0);
67
68        assert_eq!(data[0], 0.0); // clamped to min
69        assert_eq!(data[1], 5.0); // unchanged
70        assert_eq!(data[2], 10.0); // clamped to max
71        assert_eq!(data[3], 10.0); // clamped to max
72        assert_eq!(data[4], 0.0); // unchanged
73    }
74}
shape_runtime/simd_rolling/clip.rs

shape_runtime/simd_rolling/
clip.rs
