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
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268

use super::*;
use crate::Vector;

// =========================================================================
// Additional coverage tests for untested paths
// =========================================================================

#[test]
fn test_get_mut_valid() {
    let mut m = Matrix::from_vec(2, 3, vec![1.0, 2.0, 3.0, 4.0, 5.0, 6.0]).unwrap();
    if let Some(val) = m.get_mut(0, 1) {
        *val = 99.0;
    }
    assert_eq!(m.get(0, 1), Some(&99.0));
}

#[test]
fn test_get_mut_out_of_bounds() {
    let mut m = Matrix::from_vec(2, 3, vec![1.0, 2.0, 3.0, 4.0, 5.0, 6.0]).unwrap();
    assert!(m.get_mut(5, 0).is_none());
    assert!(m.get_mut(0, 10).is_none());
    assert!(m.get_mut(10, 10).is_none());
}

#[test]
fn test_matrix_zeros_coverage() {
    let m: Matrix<f32> = Matrix::zeros(3, 4);
    assert_eq!(m.rows(), 3);
    assert_eq!(m.cols(), 4);
    for val in m.as_slice() {
        assert_eq!(*val, 0.0);
    }
}

#[test]
fn test_matrix_identity_coverage() {
    let m: Matrix<f32> = Matrix::identity(3);
    assert_eq!(m.get(0, 0), Some(&1.0));
    assert_eq!(m.get(1, 1), Some(&1.0));
    assert_eq!(m.get(2, 2), Some(&1.0));
    assert_eq!(m.get(0, 1), Some(&0.0));
    assert_eq!(m.get(1, 0), Some(&0.0));
}

#[test]
fn test_get_out_of_bounds_coverage() {
    let m = Matrix::from_vec(2, 3, vec![1.0, 2.0, 3.0, 4.0, 5.0, 6.0]).unwrap();
    assert!(m.get(5, 0).is_none());
    assert!(m.get(0, 10).is_none());
}

// =========================================================================
// Kitchen Sink coverage tests (PMAT-018)
// =========================================================================

/// Test pad method with various configurations
#[test]
fn test_pad_kitchen_sink() {
    let m = Matrix::from_vec(2, 3, vec![1.0, 2.0, 3.0, 4.0, 5.0, 6.0]).unwrap();

    // Symmetric padding
    let padded = m.pad(((1, 1), (1, 1)), 0.0).unwrap();
    assert_eq!(padded.rows(), 4);
    assert_eq!(padded.cols(), 5);
    assert_eq!(padded.get(0, 0), Some(&0.0)); // top-left corner (padding)
    assert_eq!(padded.get(1, 1), Some(&1.0)); // original data starts here

    // Asymmetric padding
    let padded2 = m.pad(((2, 0), (0, 3)), 9.0).unwrap();
    assert_eq!(padded2.rows(), 4);
    assert_eq!(padded2.cols(), 6);
    assert_eq!(padded2.get(0, 0), Some(&9.0)); // padding
    assert_eq!(padded2.get(2, 0), Some(&1.0)); // original data

    // Zero padding
    let padded3 = m.pad(((0, 0), (0, 0)), 0.0).unwrap();
    assert_eq!(padded3.rows(), 2);
    assert_eq!(padded3.cols(), 3);
}

/// Test gather with various axis configurations
#[test]
fn test_gather_kitchen_sink() {
    let m = Matrix::from_vec(3, 4, (0..12).map(|x| x as f32).collect()).unwrap();

    // Gather rows (axis=0)
    let rows = m.gather(&[0, 2], 0).unwrap();
    assert_eq!(rows.rows(), 2);
    assert_eq!(rows.cols(), 4);
    assert_eq!(rows.get(0, 0), Some(&0.0)); // row 0
    assert_eq!(rows.get(1, 0), Some(&8.0)); // row 2

    // Gather columns (axis=1)
    let cols = m.gather(&[1, 3], 1).unwrap();
    assert_eq!(cols.rows(), 3);
    assert_eq!(cols.cols(), 2);
}

/// Test topk with various k values
#[test]
fn test_topk_kitchen_sink() {
    let m = Matrix::from_vec(2, 4, vec![3.0, 1.0, 4.0, 1.0, 5.0, 9.0, 2.0, 6.0]).unwrap();

    // topk operates on flattened data, not per-row
    // k=1 returns top 1 value from all 8 elements
    let (vals, idxs) = m.topk(1).unwrap();
    assert_eq!(vals.len(), 1);
    assert_eq!(idxs.len(), 1);
    assert_eq!(vals[0], 9.0); // max value in matrix

    // k=3 returns top 3 values from all 8 elements
    let (vals2, idxs2) = m.topk(3).unwrap();
    assert_eq!(vals2.len(), 3);
    assert_eq!(idxs2.len(), 3);

    // k=8 (all elements)
    let (vals3, _) = m.topk(8).unwrap();
    assert_eq!(vals3.len(), 8);

    // k=0 edge case
    let (vals4, idxs4) = m.topk(0).unwrap();
    assert_eq!(vals4.len(), 0);
    assert_eq!(idxs4.len(), 0);
}

/// Test pooling operations edge cases
#[test]
fn test_pooling_kitchen_sink() {
    // Exact divisible size
    let m = Matrix::from_vec(4, 4, (0..16).map(|x| x as f32).collect()).unwrap();

    let max_pooled = m.max_pool2d((2, 2), (2, 2)).unwrap();
    assert_eq!(max_pooled.rows(), 2);
    assert_eq!(max_pooled.cols(), 2);

    let avg_pooled = m.avg_pool2d((2, 2), (2, 2)).unwrap();
    assert_eq!(avg_pooled.rows(), 2);
    assert_eq!(avg_pooled.cols(), 2);

    // Non-exact divisible size
    let m2 = Matrix::from_vec(5, 5, (0..25).map(|x| x as f32).collect()).unwrap();
    let max_pooled2 = m2.max_pool2d((2, 2), (2, 2)).unwrap();
    assert_eq!(max_pooled2.rows(), 2); // floor(5/2)
    assert_eq!(max_pooled2.cols(), 2);
}

/// Test vecmat (v @ M) operation
#[test]
fn test_vecmat_kitchen_sink() {
    let v = Vector::from_slice(&[1.0, 2.0, 3.0]);
    let m =
        Matrix::from_vec(3, 4, 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])
            .unwrap();

    // v (1x3) @ M (3x4) -> result (1x4)
    let result = Matrix::vecmat(&v, &m).unwrap();
    assert_eq!(result.len(), 4);
    // Manual calculation: [1*1+2*5+3*9, 1*2+2*6+3*10, 1*3+2*7+3*11, 1*4+2*8+3*12]
    // = [1+10+27, 2+12+30, 3+14+33, 4+16+36] = [38, 44, 50, 56]
    assert!((result.as_slice()[0] - 38.0).abs() < 1e-5);
    assert!((result.as_slice()[1] - 44.0).abs() < 1e-5);
    assert!((result.as_slice()[2] - 50.0).abs() < 1e-5);
    assert!((result.as_slice()[3] - 56.0).abs() < 1e-5);
}

/// Test convolve2d edge cases
#[test]
fn test_convolve2d_kitchen_sink() {
    // 3x3 input with 3x3 kernel (produces 1x1)
    let input = Matrix::from_vec(3, 3, vec![1.0, 2.0, 3.0, 4.0, 5.0, 6.0, 7.0, 8.0, 9.0]).unwrap();

    let kernel = Matrix::from_vec(3, 3, vec![0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]).unwrap();

    let result = input.convolve2d(&kernel).unwrap();
    assert_eq!(result.rows(), 1);
    assert_eq!(result.cols(), 1);
    assert!((result.get(0, 0).unwrap() - 5.0).abs() < 1e-5); // Center value

    // 5x5 input with 3x3 kernel (produces 3x3)
    let input5 = Matrix::from_vec(5, 5, (0..25).map(|x| x as f32).collect()).unwrap();
    let result5 = input5.convolve2d(&kernel).unwrap();
    assert_eq!(result5.rows(), 3);
    assert_eq!(result5.cols(), 3);
}

/// Test embedding lookups
#[test]
fn test_embedding_kitchen_sink() {
    // Embedding table: 5 words x 4 dimensions
    let embeddings = Matrix::from_vec(5, 4, (0..20).map(|x| x as f32).collect()).unwrap();

    // Single lookup
    let result = embeddings.embedding_lookup(&[0]).unwrap();
    assert_eq!(result.rows(), 1);
    assert_eq!(result.cols(), 4);

    // Multiple lookups
    let result2 = embeddings.embedding_lookup(&[0, 2, 4]).unwrap();
    assert_eq!(result2.rows(), 3);
    assert_eq!(result2.cols(), 4);

    // Sparse lookup (returns matrix and indices)
    let (result3_matrix, result3_indices) = embeddings.embedding_lookup_sparse(&[0, 1, 2]).unwrap();
    assert_eq!(result3_matrix.rows(), 3);
    assert_eq!(result3_matrix.cols(), 4);
    assert_eq!(result3_indices.len(), 3);
}

/// Test batched_matmul_4d
#[test]
fn test_batched_matmul_4d_kitchen_sink() {
    // batch=2, heads=2, m=3, k=4, n=5
    let batch = 2;
    let heads = 2;
    let m = 3;
    let k = 4;
    let n = 5;

    // A: [batch, heads, m, k] = 2*2*3*4 = 48 elements
    let a_data: Vec<f32> = (0..48).map(|x| x as f32 * 0.1).collect();

    // B: [batch, heads, k, n] = 2*2*4*5 = 80 elements
    let b_data: Vec<f32> = (0..80).map(|x| x as f32 * 0.1).collect();

    // Result should be [batch, heads, m, n] = 2*2*3*5 = 60 elements
    let result = Matrix::batched_matmul_4d(&a_data, &b_data, batch, heads, m, k, n).unwrap();
    assert_eq!(result.len(), batch * heads * m * n);
}

/// Test matmul with remainder handling (non-aligned sizes)
#[test]
fn test_matmul_remainder_kitchen_sink() {
    // Sizes that don't align with SIMD widths
    for m in [1, 3, 5, 7, 9, 13, 17] {
        for k in [1, 3, 5, 7, 9, 15] {
            for n in [1, 3, 5, 7, 11] {
                let a = Matrix::from_vec(m, k, vec![1.0; m * k]).unwrap();
                let b = Matrix::from_vec(k, n, vec![1.0; k * n]).unwrap();
                let c = a.matmul(&b).unwrap();
                assert_eq!(c.rows(), m);
                assert_eq!(c.cols(), n);
                // Each element should equal k (sum of 1.0 x 1.0, k times)
                assert!((c.get(0, 0).unwrap() - k as f32).abs() < 1e-4);
            }
        }
    }
}

/// Test transpose edge cases
#[test]
fn test_transpose_kitchen_sink() {
    // 1xN
    let row = Matrix::from_vec(1, 5, vec![1.0, 2.0, 3.0, 4.0, 5.0]).unwrap();
    let col = row.transpose();
    assert_eq!(col.rows(), 5);
    assert_eq!(col.cols(), 1);

    // Nx1
    let col2 = Matrix::from_vec(5, 1, vec![1.0, 2.0, 3.0, 4.0, 5.0]).unwrap();
    let row2 = col2.transpose();
    assert_eq!(row2.rows(), 1);
    assert_eq!(row2.cols(), 5);

    // 1x1
    let single = Matrix::from_vec(1, 1, vec![42.0]).unwrap();
    let single_t = single.transpose();
    assert_eq!(single_t.get(0, 0), Some(&42.0));
}