realizar 0.8.4

Pure Rust ML inference engine built from scratch - model serving for GGUF and safetensors
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


// ============================================================================
// Tests for SIMD Helpers (PMAT-802: T-COV-95)
// ============================================================================

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

    // -------------------------------------------------------------------------
    // simd_dot_f32 Tests
    // -------------------------------------------------------------------------

    #[test]
    fn test_simd_dot_f32_basic() {
        let a = vec![1.0f32, 2.0, 3.0, 4.0];
        let b = vec![5.0f32, 6.0, 7.0, 8.0];
        let result = simd_dot_f32(&a, &b);
        // 1*5 + 2*6 + 3*7 + 4*8 = 5 + 12 + 21 + 32 = 70
        assert!((result - 70.0).abs() < 0.001);
    }

    #[test]
    fn test_simd_dot_f32_zeros() {
        let a = vec![0.0f32; 16];
        let b = vec![1.0f32; 16];
        let result = simd_dot_f32(&a, &b);
        assert!((result - 0.0).abs() < 0.001);
    }

    #[test]
    fn test_simd_dot_f32_ones() {
        let a = vec![1.0f32; 16];
        let b = vec![1.0f32; 16];
        let result = simd_dot_f32(&a, &b);
        assert!((result - 16.0).abs() < 0.001);
    }

    #[test]
    fn test_simd_dot_f32_large_vector() {
        // Test with vector size > 8 to exercise AVX2 path
        let n = 64;
        let a: Vec<f32> = (1..=n).map(|i| i as f32).collect();
        let b = vec![1.0f32; n];
        let result = simd_dot_f32(&a, &b);
        // Sum of 1 to 64 = 64*65/2 = 2080
        assert!((result - 2080.0).abs() < 0.001);
    }

    #[test]
    fn test_simd_dot_f32_remainder() {
        // Test with size not divisible by 8 to exercise remainder handling
        let n = 13;
        let a = vec![2.0f32; n];
        let b = vec![3.0f32; n];
        let result = simd_dot_f32(&a, &b);
        // 2*3*13 = 78
        assert!((result - 78.0).abs() < 0.001);
    }

    #[test]
    fn test_simd_dot_f32_small_vector() {
        // Test with size < 8 to exercise scalar fallback
        let a = vec![1.0f32, 2.0, 3.0];
        let b = vec![4.0f32, 5.0, 6.0];
        let result = simd_dot_f32(&a, &b);
        // 1*4 + 2*5 + 3*6 = 4 + 10 + 18 = 32
        assert!((result - 32.0).abs() < 0.001);
    }

    #[test]
    fn test_simd_dot_f32_negative() {
        let a = vec![-1.0f32, 2.0, -3.0, 4.0];
        let b = vec![1.0f32, 1.0, 1.0, 1.0];
        let result = simd_dot_f32(&a, &b);
        // -1 + 2 - 3 + 4 = 2
        assert!((result - 2.0).abs() < 0.001);
    }

    // -------------------------------------------------------------------------
    // simd_add_weighted Tests
    // -------------------------------------------------------------------------

    #[test]
    fn test_simd_add_weighted_basic() {
        let mut out = vec![1.0f32, 2.0, 3.0, 4.0];
        let val = vec![10.0f32, 20.0, 30.0, 40.0];
        simd_add_weighted(&mut out, &val, 0.5);
        // out[i] = out[i] + 0.5 * val[i]
        assert!((out[0] - 6.0).abs() < 0.001); // 1 + 0.5*10 = 6
        assert!((out[1] - 12.0).abs() < 0.001); // 2 + 0.5*20 = 12
        assert!((out[2] - 18.0).abs() < 0.001); // 3 + 0.5*30 = 18
        assert!((out[3] - 24.0).abs() < 0.001); // 4 + 0.5*40 = 24
    }

    #[test]
    fn test_simd_add_weighted_zero_weight() {
        let mut out = vec![1.0f32, 2.0, 3.0, 4.0];
        let val = vec![100.0f32; 4];
        simd_add_weighted(&mut out, &val, 0.0);
        // out should remain unchanged
        assert!((out[0] - 1.0).abs() < 0.001);
        assert!((out[1] - 2.0).abs() < 0.001);
    }

    #[test]
    fn test_simd_add_weighted_large_vector() {
        // Test with vector size > 8 to exercise AVX2 path
        let n = 64;
        let mut out = vec![1.0f32; n];
        let val = vec![2.0f32; n];
        simd_add_weighted(&mut out, &val, 3.0);
        // out[i] = 1 + 3*2 = 7
        for &v in &out {
            assert!((v - 7.0).abs() < 0.001);
        }
    }

    #[test]
    fn test_simd_add_weighted_remainder() {
        // Test with size not divisible by 8
        let n = 11;
        let mut out = vec![0.0f32; n];
        let val: Vec<f32> = (1..=n).map(|i| i as f32).collect();
        simd_add_weighted(&mut out, &val, 2.0);
        // out[i] = 0 + 2*val[i]
        for (i, &v) in out.iter().enumerate() {
            let expected = 2.0 * (i + 1) as f32;
            assert!((v - expected).abs() < 0.001);
        }
    }

    #[test]
    fn test_simd_add_weighted_small_vector() {
        // Test with size < 8 to exercise scalar fallback
        let mut out = vec![5.0f32, 10.0];
        let val = vec![1.0f32, 1.0];
        simd_add_weighted(&mut out, &val, -2.0);
        // out[0] = 5 + (-2)*1 = 3
        // out[1] = 10 + (-2)*1 = 8
        assert!((out[0] - 3.0).abs() < 0.001);
        assert!((out[1] - 8.0).abs() < 0.001);
    }

    #[test]
    fn test_simd_add_weighted_accumulate() {
        // Test multiple accumulations
        let mut out = vec![0.0f32; 16];
        let val1 = vec![1.0f32; 16];
        let val2 = vec![2.0f32; 16];

        simd_add_weighted(&mut out, &val1, 1.0);
        simd_add_weighted(&mut out, &val2, 0.5);
        // out = 0 + 1*1 + 0.5*2 = 2
        for &v in &out {
            assert!((v - 2.0).abs() < 0.001);
        }
    }
}