numr 0.5.2

High-performance numerical computing with multi-backend GPU acceleration (CPU/CUDA/WebGPU)
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

//! SIMD-accelerated clamp operation
//!
//! clamp(x, min, max) = min(max(x, min), max)
//!
//! # SIMD Approach
//!
//! - Broadcast min and max values to vectors
//! - Use SIMD max then min operations

#[cfg(target_arch = "x86_64")]
mod avx2;
#[cfg(target_arch = "x86_64")]
mod avx512;

#[cfg(target_arch = "aarch64")]
mod aarch64;

use super::{SimdLevel, detect_simd};

/// Minimum length to justify SIMD overhead
const SIMD_THRESHOLD: usize = 32;

/// SIMD clamp for f32
///
/// # Safety
/// - `a` and `out` must point to `len` elements
#[inline]
pub unsafe fn clamp_f32(a: *const f32, out: *mut f32, len: usize, min_val: f32, max_val: f32) {
    let level = detect_simd();

    if len < SIMD_THRESHOLD || level == SimdLevel::Scalar {
        clamp_scalar_f32(a, out, len, min_val, max_val);
        return;
    }

    #[cfg(target_arch = "x86_64")]
    match level {
        SimdLevel::Avx512 => avx512::clamp_f32(a, out, len, min_val, max_val),
        SimdLevel::Avx2Fma => avx2::clamp_f32(a, out, len, min_val, max_val),
        _ => clamp_scalar_f32(a, out, len, min_val, max_val),
    }

    #[cfg(target_arch = "aarch64")]
    match level {
        SimdLevel::Neon | SimdLevel::NeonFp16 => {
            aarch64::neon::clamp_f32(a, out, len, min_val, max_val)
        }
        _ => clamp_scalar_f32(a, out, len, min_val, max_val),
    }

    #[cfg(not(any(target_arch = "x86_64", target_arch = "aarch64")))]
    clamp_scalar_f32(a, out, len, min_val, max_val);
}

/// SIMD clamp for f64
///
/// # Safety
/// - `a` and `out` must point to `len` elements
#[inline]
pub unsafe fn clamp_f64(a: *const f64, out: *mut f64, len: usize, min_val: f64, max_val: f64) {
    let level = detect_simd();

    if len < SIMD_THRESHOLD || level == SimdLevel::Scalar {
        clamp_scalar_f64(a, out, len, min_val, max_val);
        return;
    }

    #[cfg(target_arch = "x86_64")]
    match level {
        SimdLevel::Avx512 => avx512::clamp_f64(a, out, len, min_val, max_val),
        SimdLevel::Avx2Fma => avx2::clamp_f64(a, out, len, min_val, max_val),
        _ => clamp_scalar_f64(a, out, len, min_val, max_val),
    }

    #[cfg(target_arch = "aarch64")]
    match level {
        SimdLevel::Neon | SimdLevel::NeonFp16 => {
            aarch64::neon::clamp_f64(a, out, len, min_val, max_val)
        }
        _ => clamp_scalar_f64(a, out, len, min_val, max_val),
    }

    #[cfg(not(any(target_arch = "x86_64", target_arch = "aarch64")))]
    clamp_scalar_f64(a, out, len, min_val, max_val);
}

// ============================================================================
// Scalar fallbacks
// ============================================================================

/// Scalar clamp for f32
#[inline]
pub unsafe fn clamp_scalar_f32(
    a: *const f32,
    out: *mut f32,
    len: usize,
    min_val: f32,
    max_val: f32,
) {
    for i in 0..len {
        let val = *a.add(i);
        *out.add(i) = val.max(min_val).min(max_val);
    }
}

/// Scalar clamp for f64
#[inline]
pub unsafe fn clamp_scalar_f64(
    a: *const f64,
    out: *mut f64,
    len: usize,
    min_val: f64,
    max_val: f64,
) {
    for i in 0..len {
        let val = *a.add(i);
        *out.add(i) = val.max(min_val).min(max_val);
    }
}

half_clamp!(clamp, clamp_f32);

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

    #[test]
    fn test_clamp_f32() {
        let len = 128;
        let input: Vec<f32> = (0..len).map(|x| (x as f32) - 64.0).collect();
        let mut out = vec![0.0f32; len];
        let mut out_ref = vec![0.0f32; len];

        let min_val = -10.0f32;
        let max_val = 10.0f32;

        unsafe {
            clamp_f32(input.as_ptr(), out.as_mut_ptr(), len, min_val, max_val);
            clamp_scalar_f32(input.as_ptr(), out_ref.as_mut_ptr(), len, min_val, max_val);
        }

        assert_eq!(out, out_ref);
    }

    #[test]
    fn test_clamp_all_below() {
        let len = 64;
        let input = vec![-100.0f32; len];
        let mut out = vec![0.0f32; len];

        unsafe {
            clamp_f32(input.as_ptr(), out.as_mut_ptr(), len, 0.0, 10.0);
        }

        assert!(out.iter().all(|&x| x == 0.0));
    }

    #[test]
    fn test_clamp_all_above() {
        let len = 64;
        let input = vec![100.0f32; len];
        let mut out = vec![0.0f32; len];

        unsafe {
            clamp_f32(input.as_ptr(), out.as_mut_ptr(), len, 0.0, 10.0);
        }

        assert!(out.iter().all(|&x| x == 10.0));
    }

    #[test]
    fn test_clamp_in_range() {
        let len = 64;
        let input: Vec<f32> = (0..len).map(|x| x as f32).collect();
        let mut out = vec![0.0f32; len];

        unsafe {
            clamp_f32(input.as_ptr(), out.as_mut_ptr(), len, -100.0, 100.0);
        }

        assert_eq!(out, input);
    }
}