hermes-simd 0.5.0

High-performance, zero-overhead SIMD abstraction library
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

//! Integration tests for `SimdCow` math extensions:
//! norm_sq, norm, normalize, add/sub/mul/div scalar broadcast, div_cow.
//!
//! All tests run against the `Scalar` arch (lane=1) for portability.
//! The scalar fallback is the authoritative correctness reference.

use hermes_simd::{Scalar, SimdCow, SimdError, Unaligned};

type Cow<'a, T> = SimdCow<'a, T, Scalar, Unaligned>;

fn assert_simd_error<T>(result: Result<T, SimdError>, expected: SimdError) {
    match result {
        Err(actual) => assert_eq!(actual, expected),
        Ok(_) => panic!("expected {expected:?}"),
    }
}

// ---------------------------------------------------------------------------
// Scalar-broadcast arithmetic
// ---------------------------------------------------------------------------

#[test]
fn test_add_scalar_cow_f32() {
    let data = vec![1.0f32, 2.0, 3.0, 4.0, 5.0];
    let cow = Cow::<f32>::borrow_slice(&data).unwrap();
    let result = cow.add_scalar_cow(10.0);
    assert_eq!(result.as_ref(), &[11.0f32, 12.0, 13.0, 14.0, 15.0]);
}

#[test]
fn test_sub_scalar_cow_f32() {
    let data = vec![10.0f32, 20.0, 30.0, 40.0];
    let cow = Cow::<f32>::borrow_slice(&data).unwrap();
    let result = cow.sub_scalar_cow(5.0);
    assert_eq!(result.as_ref(), &[5.0f32, 15.0, 25.0, 35.0]);
}

#[test]
fn test_mul_scalar_cow_f32() {
    let data = vec![1.0f32, 2.0, 3.0, 4.0];
    let cow = Cow::<f32>::borrow_slice(&data).unwrap();
    let result = cow.mul_scalar_cow(3.0);
    assert_eq!(result.as_ref(), &[3.0f32, 6.0, 9.0, 12.0]);
}

#[test]
fn test_div_scalar_cow_f32() {
    let data = vec![10.0f32, 20.0, 30.0, 40.0];
    let cow = Cow::<f32>::borrow_slice(&data).unwrap();
    let result = cow.div_scalar_cow(10.0);
    let expected = [1.0f32, 2.0, 3.0, 4.0];
    for (a, b) in result.as_ref().iter().zip(expected.iter()) {
        assert!((a - b).abs() < 1e-6, "got={a}, expected={b}");
    }
}

#[test]
fn test_scalar_broadcast_empty() {
    let data: Vec<f32> = vec![];
    let cow = Cow::<f32>::borrow_slice(&data).unwrap();
    assert!(cow.add_scalar_cow(5.0).is_empty());
    assert!(cow.sub_scalar_cow(5.0).is_empty());
    assert!(cow.mul_scalar_cow(5.0).is_empty());
    assert!(cow.div_scalar_cow(5.0).is_empty());
}

#[test]
fn test_scalar_broadcast_tail_handling() {
    // 5 elements — exercises scalar tail with Scalar arch (lane=1 → no tail)
    // but tests correctness for non-power-of-2 lengths with wider arches.
    let data = vec![1.0f32, 2.0, 3.0, 4.0, 5.0];
    let cow = Cow::<f32>::from_slice(&data);
    let result = cow.add_scalar_cow(100.0);
    assert_eq!(result.as_ref(), &[101.0f32, 102.0, 103.0, 104.0, 105.0]);
}

// ---------------------------------------------------------------------------
// div_cow (elementwise division)
// ---------------------------------------------------------------------------

#[test]
fn test_div_cow_f32() {
    let a_data = vec![10.0f32, 20.0, 30.0, 40.0];
    let b_data = vec![2.0f32, 4.0, 5.0, 8.0];
    let a = Cow::<f32>::borrow_slice(&a_data).unwrap();
    let b = Cow::<f32>::borrow_slice(&b_data).unwrap();
    let result = a.div_cow(&b).unwrap();
    let expected = [5.0f32, 5.0, 6.0, 5.0];
    for (got, exp) in result.as_ref().iter().zip(expected.iter()) {
        assert!((got - exp).abs() < 1e-5, "got={got}, expected={exp}");
    }
}

#[test]
fn test_div_cow_length_mismatch() {
    let a = Cow::<f32>::from_slice(&[1.0, 2.0]);
    let b = Cow::<f32>::from_slice(&[1.0, 2.0, 3.0]);
    assert_simd_error(a.div_cow(&b), SimdError::LengthMismatch);
}

// ---------------------------------------------------------------------------
// norm_sq, norm
// ---------------------------------------------------------------------------

#[test]
fn test_norm_sq_f32() {
    let data = vec![3.0f32, 4.0];
    let cow = Cow::<f32>::borrow_slice(&data).unwrap();
    // ‖(3,4)‖² = 9 + 16 = 25
    let ns = cow.norm_sq();
    assert!((ns - 25.0f32).abs() < 1e-5, "norm_sq={ns}");
}

#[test]
fn test_norm_f32() {
    let data = vec![3.0f32, 4.0];
    let cow = Cow::<f32>::borrow_slice(&data).unwrap();
    // ‖(3,4)‖ = 5
    let n = cow.norm();
    assert!((n - 5.0f32).abs() < 1e-5, "norm={n}");
}

#[test]
fn test_norm_sq_unit_vector() {
    // Unit vector — norm_sq must be 1.0
    let s = 1.0f32 / 2.0f32.sqrt();
    let data = vec![s, s];
    let cow = Cow::<f32>::borrow_slice(&data).unwrap();
    let ns = cow.norm_sq();
    assert!((ns - 1.0f32).abs() < 1e-5, "norm_sq of unit vec = {ns}");
}

#[test]
fn test_norm_zero_vector() {
    let data = vec![0.0f32, 0.0, 0.0, 0.0];
    let cow = Cow::<f32>::borrow_slice(&data).unwrap();
    assert_eq!(cow.norm_sq(), 0.0f32);
    assert_eq!(cow.norm(), 0.0f32);
}

// ---------------------------------------------------------------------------
// normalize
// ---------------------------------------------------------------------------

#[test]
fn test_normalize_f32() {
    let data = vec![3.0f32, 4.0];
    let cow = Cow::<f32>::borrow_slice(&data).unwrap();
    let unit = cow.normalize();
    // Normalized (3,4)/5 → (0.6, 0.8)
    let expected = [0.6f32, 0.8];
    for (got, exp) in unit.as_ref().iter().zip(expected.iter()) {
        assert!((got - exp).abs() < 1e-5, "got={got}, expected={exp}");
    }
    // Re-norm of result should be 1
    let renormed = SimdCow::<f32, Scalar, Unaligned>::borrow_slice(unit.as_ref()).unwrap();
    let n = renormed.norm();
    assert!((n - 1.0f32).abs() < 1e-5, "re-norm={n}");
}

#[test]
fn test_normalize_zero_vector_safe() {
    // Zero vector → normalize returns zeros, no NaN/panic
    let data = vec![0.0f32, 0.0, 0.0];
    let cow = Cow::<f32>::borrow_slice(&data).unwrap();
    let result = cow.normalize();
    assert_eq!(result.as_ref(), &[0.0f32, 0.0, 0.0]);
}

#[test]
fn test_normalize_empty_vector_safe() {
    let data: Vec<f32> = vec![];
    let cow = Cow::<f32>::borrow_slice(&data).unwrap();
    let result = cow.normalize();
    assert!(result.is_empty());
}

#[test]
fn test_normalize_single_positive() {
    // (5.0) → (1.0)
    let cow = Cow::<f32>::from_slice(&[5.0f32]);
    let unit = cow.normalize();
    assert!((unit.as_ref()[0] - 1.0f32).abs() < 1e-5);
}

#[test]
fn test_normalize_owned_returned() {
    // normalize always returns SimdCow::Owned regardless of input variant
    let data = vec![3.0f32, 4.0];
    let cow = Cow::<f32>::borrow_slice(&data).unwrap();
    let result = cow.normalize();
    assert!(matches!(result, SimdCow::Owned(_)));
}

// ---------------------------------------------------------------------------
// ZST size invariants — math extension marker types
// ---------------------------------------------------------------------------

#[test]
fn test_math_fn_zero_runtime_state() {
    use core::mem::size_of;
    // The broadcast_op helper is generic — verify no strategy type adds state.
    // Clamp<T> is the only stateful UnaryOp; all ElementOp ZSTs are size 0.
    use hermes_simd::{Add, Div, Mul, Sub};
    assert_eq!(size_of::<Add>(), 0);
    assert_eq!(size_of::<Sub>(), 0);
    assert_eq!(size_of::<Mul>(), 0);
    assert_eq!(size_of::<Div>(), 0);
}