trueno 0.18.0

High-performance SIMD compute library with GPU support, LLM inference engine, and GGUF model loading
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

//! FALSIFY-BLIS contract enforcement tests (blis-gemm-v1).
//!
//! Targets the top coverage gaps: gemm_blis, elementwise, norms,
//! softmax, gemv. Exercises 0%-covered SIMD paths.

use super::super::*;
use crate::blis;

// ═══ Contract: blis-gemm-v1 — GEMM correctness ═══

#[test]
fn falsify_blis_001_gemm_standard_sizes() {
    for (m, n, k) in [(32, 32, 32), (64, 64, 64), (128, 64, 32), (16, 48, 24)] {
        let a = vec![1.0f32; m * k];
        let b = vec![1.0f32; k * n];
        let mut c = vec![0.0f32; m * n];
        blis::gemm_blis(m, n, k, &a, &b, &mut c, None).unwrap();
        for (i, &v) in c.iter().enumerate() {
            assert!((v - k as f32).abs() < 1.0, "001: c[{i}]={v}, exp {k}");
        }
    }
}

#[test]
fn falsify_blis_002_gemm_non_aligned() {
    for (m, n, k) in [(7, 13, 5), (3, 1, 11), (1, 1, 1), (17, 23, 9)] {
        let a: Vec<f32> = (0..m * k).map(|i| (i % 7) as f32 * 0.1).collect();
        let b: Vec<f32> = (0..k * n).map(|i| (i % 5) as f32 * 0.1).collect();
        let mut c_blis = vec![0.0f32; m * n];
        let mut c_ref = vec![0.0f32; m * n];
        blis::gemm_blis(m, n, k, &a, &b, &mut c_blis, None).unwrap();
        reference::gemm_reference(m, n, k, &a, &b, &mut c_ref).unwrap();
        for i in 0..m * n {
            let diff = (c_blis[i] - c_ref[i]).abs();
            assert!(diff < 0.1, "002: [{i}] blis={} ref={}", c_blis[i], c_ref[i]);
        }
    }
}

#[test]
fn falsify_blis_003_elementwise_add() {
    for len in [1, 7, 8, 15, 16, 31, 32, 100, 255, 256, 1000] {
        let a: Vec<f32> = (0..len).map(|i| i as f32).collect();
        let b: Vec<f32> = (0..len).map(|i| (len - i) as f32).collect();
        let result = elementwise::add_alloc(&a, &b);
        assert_eq!(result.len(), len, "003: length mismatch for {len}");
        for i in 0..len {
            assert!((result[i] - len as f32).abs() < 1e-5, "003: add[{i}]={}", result[i]);
        }
    }
}

#[test]
fn falsify_blis_004_rms_norm() {
    let x = vec![1.0f32, 2.0, 3.0, 4.0, 5.0, 6.0, 7.0, 8.0];
    let w = vec![1.0f32; 8];
    let eps = 1e-5;
    let result = norms::rms_norm_alloc(&x, &w, eps);
    assert_eq!(result.len(), 8, "004: output length");
    for (i, &v) in result.iter().enumerate() {
        assert!(v.is_finite(), "004: rms_norm[{i}] not finite");
    }
    assert!(result.iter().map(|x| x.abs()).sum::<f32>() > 0.0, "004: all zeros");
}

#[test]
fn falsify_blis_005_gemv() {
    // gemv(k, n, a, b, c): a[k] is the vector, b[k×n] is the matrix, c[n] is output
    // c[j] = sum_i(a[i] * b[i*n + j])
    for (k, n) in [(8, 16), (32, 32), (64, 128), (3, 4), (1, 1)] {
        let a: Vec<f32> = (0..k).map(|i| (i + 1) as f32 * 0.1).collect();
        let b: Vec<f32> = (0..k * n).map(|i| (i % 10) as f32 * 0.01).collect();
        let mut c = vec![0.0f32; n];
        gemv::gemv(k, n, &a, &b, &mut c);
        for j in 0..n {
            let expected: f32 = (0..k).map(|i| a[i] * b[i * n + j]).sum();
            assert!((c[j] - expected).abs() < 0.01, "005: [{j}] got={} exp={expected}", c[j]);
        }
    }
}

#[test]
fn falsify_blis_006_gemm_large_avx_path() {
    let m = 128;
    let n = 128;
    let k = 128;
    let a: Vec<f32> = (0..m * k).map(|i| ((i * 7 + 3) % 100) as f32 * 0.01).collect();
    let b: Vec<f32> = (0..k * n).map(|i| ((i * 11 + 5) % 100) as f32 * 0.01).collect();
    let mut c = vec![0.0f32; m * n];
    let mut c_ref = vec![0.0f32; m * n];
    blis::gemm_blis(m, n, k, &a, &b, &mut c, None).unwrap();
    reference::gemm_reference(m, n, k, &a, &b, &mut c_ref).unwrap();
    let max_diff: f32 =
        c.iter().zip(c_ref.iter()).map(|(a, b)| (a - b).abs()).fold(0.0f32, f32::max);
    assert!(max_diff < 0.5, "006: large GEMM max_diff={max_diff}");
}

#[test]
fn falsify_blis_007_elementwise_add_inplace() {
    for len in [1, 8, 32, 255, 1024] {
        let mut a: Vec<f32> = (0..len).map(|i| i as f32).collect();
        let b: Vec<f32> = (0..len).map(|_| 1.0).collect();
        elementwise::add_inplace(&mut a, &b).unwrap();
        for i in 0..len {
            assert!((a[i] - (i as f32 + 1.0)).abs() < 1e-5, "007: [{i}]={}", a[i]);
        }
    }
}

#[test]
fn falsify_blis_008_layer_norm() {
    let x: Vec<f32> = (0..64).map(|i| (i as f32) * 0.5 - 16.0).collect();
    let w = vec![1.0f32; 64];
    let b = vec![0.0f32; 64];
    let eps = 1e-5;
    let result = norms::layer_norm_alloc(&x, &w, &b, eps);
    assert_eq!(result.len(), 64, "008: output length");
    for (i, &v) in result.iter().enumerate() {
        assert!(v.is_finite(), "008: layer_norm[{i}] not finite");
    }
    let mean: f32 = result.iter().sum::<f32>() / result.len() as f32;
    assert!(mean.abs() < 0.1, "008: mean={mean}");
}

#[test]
fn falsify_blis_009_softmax() {
    let x: Vec<f32> = (0..32).map(|i| (i as f32) * 0.5 - 8.0).collect();
    let result = softmax::softmax_1d_alloc(&x);
    assert_eq!(result.len(), 32, "009: output length");
    let sum: f32 = result.iter().sum();
    assert!((sum - 1.0).abs() < 1e-5, "009: softmax sum={sum}");
    for (i, &v) in result.iter().enumerate() {
        assert!(v >= 0.0, "009: softmax[{i}]={v} < 0");
    }
}

#[test]
fn falsify_blis_010_relu() {
    let input: Vec<f32> = (-50..50).map(|i| i as f32 * 0.1).collect();
    let result = elementwise::relu_alloc(&input);
    for (&r, &x) in result.iter().zip(input.iter()) {
        if x < 0.0 {
            assert_eq!(r, 0.0, "010: relu({x}) should be 0");
        } else {
            assert_eq!(r, x, "010: relu({x}) should be {x}");
        }
    }
}

#[test]
fn falsify_blis_011_fused_mul_add() {
    let a: Vec<f32> = (0..64).map(|i| i as f32).collect();
    let b: Vec<f32> = vec![2.0f32; 64];
    let c: Vec<f32> = vec![1.0f32; 64];
    let mut out = vec![0.0f32; 64];
    elementwise::fused_mul_add(&a, &b, &c, &mut out).unwrap();
    for i in 0..64 {
        let expected = a[i] * b[i] + c[i];
        assert!((out[i] - expected).abs() < 1e-5, "011: [{i}] got={} exp={expected}", out[i]);
    }
}

#[test]
fn falsify_blis_012_scale_inplace() {
    let mut data: Vec<f32> = (0..100).map(|i| i as f32).collect();
    elementwise::scale_inplace(&mut data, 0.5);
    for i in 0..100 {
        assert!((data[i] - i as f32 * 0.5).abs() < 1e-5, "012: [{i}]={}", data[i]);
    }
}