trueno 0.16.4

High-performance SIMD compute library with GPU support for matrix operations
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
210
211
212

//! Coverage tests for `validate_gemm_dims` (compute.rs:143) and
//! `gemm_blis_parallel` (parallel.rs:67).
//!
//! Targets: 15 uncovered lines in validate_gemm_dims, 19 in gemm_blis_parallel.

use super::super::*;

// ========================================================================
// validate_gemm_dims — exercised via gemm_blis error paths
// ========================================================================

/// A-size mismatch: a.len() != m * k
#[test]
fn test_validate_gemm_dims_a_mismatch() {
    let a = vec![1.0f32; 10]; // expected m*k = 3*4 = 12
    let b = vec![1.0f32; 20]; // k*n = 4*5 = 20
    let mut c = vec![0.0f32; 15]; // m*n = 3*5 = 15

    let result = gemm_blis(3, 5, 4, &a, &b, &mut c, None);
    assert!(result.is_err());
    let err = result.unwrap_err();
    let msg = format!("{}", err);
    assert!(msg.contains("A size mismatch"), "Got: {}", msg);
}

/// B-size mismatch: b.len() != k * n
#[test]
fn test_validate_gemm_dims_b_mismatch() {
    let a = vec![1.0f32; 12]; // m*k = 3*4 = 12 OK
    let b = vec![1.0f32; 19]; // expected k*n = 4*5 = 20
    let mut c = vec![0.0f32; 15]; // m*n = 3*5 = 15

    let result = gemm_blis(3, 5, 4, &a, &b, &mut c, None);
    assert!(result.is_err());
    let msg = format!("{}", result.unwrap_err());
    assert!(msg.contains("B size mismatch"), "Got: {}", msg);
}

/// C-size mismatch: c.len() != m * n
#[test]
fn test_validate_gemm_dims_c_mismatch() {
    let a = vec![1.0f32; 12]; // m*k = 3*4 = 12 OK
    let b = vec![1.0f32; 20]; // k*n = 4*5 = 20 OK
    let mut c = vec![0.0f32; 14]; // expected m*n = 3*5 = 15

    let result = gemm_blis(3, 5, 4, &a, &b, &mut c, None);
    assert!(result.is_err());
    let msg = format!("{}", result.unwrap_err());
    assert!(msg.contains("C size mismatch"), "Got: {}", msg);
}

/// All dimensions correct — should succeed.
#[test]
fn test_validate_gemm_dims_all_correct() {
    let m = 4;
    let n = 3;
    let k = 5;
    let a = vec![1.0f32; m * k];
    let b = vec![1.0f32; k * n];
    let mut c = vec![0.0f32; m * n];

    let result = gemm_blis(m, n, k, &a, &b, &mut c, None);
    assert!(result.is_ok());
}

/// Zero-dimension m: should return Ok immediately (early exit).
#[test]
fn test_gemm_blis_zero_m() {
    let a: Vec<f32> = vec![];
    let b = vec![1.0f32; 20]; // k*n = 4*5
    let mut c: Vec<f32> = vec![];

    let result = gemm_blis(0, 5, 4, &a, &b, &mut c, None);
    assert!(result.is_ok());
}

/// Zero-dimension n: should return Ok immediately.
#[test]
fn test_gemm_blis_zero_n() {
    let a = vec![1.0f32; 12]; // m*k = 3*4
    let b: Vec<f32> = vec![];
    let mut c: Vec<f32> = vec![];

    let result = gemm_blis(3, 0, 4, &a, &b, &mut c, None);
    assert!(result.is_ok());
}

/// Zero-dimension k: should return Ok immediately.
#[test]
fn test_gemm_blis_zero_k() {
    let a: Vec<f32> = vec![];
    let b: Vec<f32> = vec![];
    let mut c = vec![0.0f32; 15]; // m*n = 3*5

    let result = gemm_blis(3, 5, 0, &a, &b, &mut c, None);
    assert!(result.is_ok());
}

/// Small matrix (m*n*k < 4096) falls through to reference GEMM.
#[test]
fn test_gemm_blis_falls_to_reference() {
    // 4*4*4 = 64 < 4096 => reference path
    let a =
        vec![1.0, 2.0, 3.0, 4.0, 5.0, 6.0, 7.0, 8.0, 9.0, 10.0, 11.0, 12.0, 13.0, 14.0, 15.0, 16.0];
    let b = vec![1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0];
    let mut c = vec![0.0f32; 16];

    gemm_blis(4, 4, 4, &a, &b, &mut c, None).unwrap();
    // Multiplying by identity should give A back
    for i in 0..16 {
        assert!((c[i] - a[i]).abs() < 1e-5, "c[{}] = {}, expected {}", i, c[i], a[i]);
    }
}

// ========================================================================
// gemm_blis_parallel — exercises dimension mismatch and large-matrix path
// ========================================================================

/// Parallel GEMM: a-size mismatch.
#[test]
fn test_gemm_blis_parallel_a_mismatch() {
    let a = vec![1.0f32; 5]; // wrong: expected m*k = 2*3 = 6
    let b = vec![1.0f32; 12]; // k*n = 3*4 = 12
    let mut c = vec![0.0f32; 8]; // m*n = 2*4 = 8

    let result = gemm_blis_parallel(2, 4, 3, &a, &b, &mut c);
    assert!(result.is_err());
}

/// Parallel GEMM: b-size mismatch.
#[test]
fn test_gemm_blis_parallel_b_mismatch() {
    let a = vec![1.0f32; 6]; // m*k = 2*3 = 6 OK
    let b = vec![1.0f32; 11]; // wrong: expected k*n = 3*4 = 12
    let mut c = vec![0.0f32; 8]; // m*n = 2*4 = 8

    let result = gemm_blis_parallel(2, 4, 3, &a, &b, &mut c);
    assert!(result.is_err());
}

/// Parallel GEMM: c-size mismatch.
#[test]
fn test_gemm_blis_parallel_c_mismatch() {
    let a = vec![1.0f32; 6]; // m*k = 2*3 = 6 OK
    let b = vec![1.0f32; 12]; // k*n = 3*4 = 12 OK
    let mut c = vec![0.0f32; 7]; // wrong: expected m*n = 2*4 = 8

    let result = gemm_blis_parallel(2, 4, 3, &a, &b, &mut c);
    assert!(result.is_err());
}

/// Parallel GEMM: small matrix falls through to single-threaded gemm_blis.
#[test]
fn test_gemm_blis_parallel_small_falls_to_sequential() {
    // m*n*k = 8*8*8 = 512 < 1_000_000 => sequential path
    let n = 8;
    let a: Vec<f32> = (0..n * n).map(|i| (i % 7) as f32).collect();
    let b: Vec<f32> = (0..n * n).map(|i| ((i + 3) % 5) as f32).collect();
    let mut c_par = vec![0.0f32; n * n];
    let mut c_ref = vec![0.0f32; n * n];

    gemm_blis_parallel(n, n, n, &a, &b, &mut c_par).unwrap();
    gemm_reference(n, n, n, &a, &b, &mut c_ref).unwrap();

    for i in 0..n * n {
        assert!(
            (c_par[i] - c_ref[i]).abs() < 1e-3,
            "Mismatch at {}: par={}, ref={}",
            i,
            c_par[i],
            c_ref[i]
        );
    }
}

/// Parallel GEMM: large matrix exercises the parallel Rayon path.
/// m*n*k = 128*128*128 = 2_097_152 > 1_000_000 => parallel path (when `parallel` feature is on).
#[test]
fn test_gemm_blis_parallel_large_matrix() {
    let n = 128;
    let a: Vec<f32> = (0..n * n).map(|i| ((i % 11) as f32) * 0.1).collect();
    let b: Vec<f32> = (0..n * n).map(|i| ((i % 7) as f32) * 0.1).collect();
    let mut c_par = vec![0.0f32; n * n];
    let mut c_ref = vec![0.0f32; n * n];

    gemm_blis_parallel(n, n, n, &a, &b, &mut c_par).unwrap();
    gemm_reference(n, n, n, &a, &b, &mut c_ref).unwrap();

    let mut max_diff = 0.0f32;
    for i in 0..n * n {
        let diff = (c_par[i] - c_ref[i]).abs();
        max_diff = max_diff.max(diff);
    }
    assert!(max_diff < 1e-1, "Max diff: {}", max_diff);
}

/// Parallel GEMM: rectangular matrix (M > N).
#[test]
fn test_gemm_blis_parallel_rectangular_tall() {
    let m = 200;
    let n = 50;
    let k = 60;
    // m*n*k = 600_000 < 1M, sequential path; but tests validation logic
    let a: Vec<f32> = (0..m * k).map(|i| ((i % 9) as f32) * 0.1).collect();
    let b: Vec<f32> = (0..k * n).map(|i| ((i % 5) as f32) * 0.1).collect();
    let mut c = vec![0.0f32; m * n];

    let result = gemm_blis_parallel(m, n, k, &a, &b, &mut c);
    assert!(result.is_ok());
    // Verify at least one non-zero output
    assert!(c.iter().any(|&v| v != 0.0));
}