trueno 0.17.5

High-performance SIMD compute library with GPU support for matrix operations
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
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
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
//! Embedding Lookup Contract Falsification Tests
//!
//! Popperian falsification of embedding-lookup-v1.yaml claims:
//!   - FALSIFY-EM-001: Output shape = (seq_len, d_model)
//!   - FALSIFY-EM-002: OOB token IDs produce controlled error (no panic)
//!   - FALSIFY-EM-003: Deterministic output (identical W + ids → identical result)
//!   - FALSIFY-EM-004: Finite output when embedding table is finite
//!
//! Five-Whys (PMAT-354):
//!   Why #1: trueno has 9 embedding_lookup tests but zero FALSIFY-EM-* tagged tests
//!   Why #2: existing tests verify examples, not provable-contract YAML claims
//!   Why #3: trueno was built before the embedding-lookup-v1.yaml contract
//!   Why #4: no systematic link between contract YAML and Rust test names
//!   Why #5: no cross-repo falsification naming convention existed
//!
//! References:
//!   - Mikolov et al. (2013) Efficient Estimation of Word Representations
//!   - provable-contracts/contracts/embedding-lookup-v1.yaml
//!   - src/matrix/ops/ml_ops/mod.rs

use super::super::Matrix;

// ============================================================================
// FALSIFY-EM-001: Output shape correctness
// Contract: output.shape = (seq_len, d_model) for any valid seq_len
// ============================================================================

#[test]
fn falsify_em_001_output_shape_single() {
    let table = Matrix::from_vec(10, 4, vec![1.0; 40]).unwrap();
    let result = table.embedding_lookup(&[3]).unwrap();

    assert_eq!(
        result.rows(),
        1,
        "FALSIFIED EM-001: single lookup rows={}, expected 1",
        result.rows()
    );
    assert_eq!(
        result.cols(),
        4,
        "FALSIFIED EM-001: single lookup cols={}, expected 4",
        result.cols()
    );
}

#[test]
fn falsify_em_001_output_shape_batch() {
    let vocab_size = 50;
    let d_model = 16;
    let data: Vec<f32> = (0..vocab_size * d_model).map(|i| i as f32).collect();
    let table = Matrix::from_vec(vocab_size, d_model, data).unwrap();

    for seq_len in [1, 2, 5, 10, 49] {
        let indices: Vec<usize> = (0..seq_len).collect();
        let result = table.embedding_lookup(&indices).unwrap();

        assert_eq!(
            result.rows(),
            seq_len,
            "FALSIFIED EM-001: seq_len={seq_len}, got rows={}",
            result.rows()
        );
        assert_eq!(
            result.cols(),
            d_model,
            "FALSIFIED EM-001: seq_len={seq_len}, got cols={}",
            result.cols()
        );
    }
}

#[test]
fn falsify_em_001_output_shape_empty() {
    let table = Matrix::from_vec(10, 4, vec![1.0; 40]).unwrap();
    let result = table.embedding_lookup(&[]).unwrap();

    assert_eq!(result.rows(), 0, "FALSIFIED EM-001: empty indices should produce 0 rows");
    assert_eq!(result.cols(), 4, "FALSIFIED EM-001: empty indices should preserve d_model=4");
}

// ============================================================================
// FALSIFY-EM-002: Out-of-bounds safety
// Contract: OOB token IDs produce controlled error, not panic
// ============================================================================

#[test]
fn falsify_em_002_oob_returns_error() {
    let table = Matrix::from_vec(5, 3, vec![1.0; 15]).unwrap();

    // Exactly at boundary
    let result = table.embedding_lookup(&[5]);
    assert!(result.is_err(), "FALSIFIED EM-002: index=5 should error for vocab_size=5");
}

#[test]
fn falsify_em_002_oob_error_message() {
    let table = Matrix::from_vec(5, 3, vec![1.0; 15]).unwrap();
    let err = table.embedding_lookup(&[100]).unwrap_err();

    assert!(
        err.to_string().contains("out of bounds"),
        "FALSIFIED EM-002: error message should mention 'out of bounds', got: {}",
        err
    );
}

#[test]
fn falsify_em_002_boundary_valid() {
    let table = Matrix::from_vec(5, 3, vec![1.0; 15]).unwrap();

    // Last valid index
    let result = table.embedding_lookup(&[4]);
    assert!(result.is_ok(), "FALSIFIED EM-002: index=4 should succeed for vocab_size=5");
}

#[test]
fn falsify_em_002_mixed_valid_invalid() {
    let table = Matrix::from_vec(5, 3, vec![1.0; 15]).unwrap();

    // Valid + invalid mix: should fail (first invalid contaminates)
    let result = table.embedding_lookup(&[0, 2, 10]);
    assert!(result.is_err(), "FALSIFIED EM-002: mixed valid+invalid should error");
}

// ============================================================================
// FALSIFY-EM-003: Deterministic output
// Contract: two calls with identical W and token_ids produce bit-identical output
// ============================================================================

#[test]
fn falsify_em_003_determinism() {
    let data: Vec<f32> = (0..200).map(|i| (i as f32 * 0.37).sin()).collect();
    let table = Matrix::from_vec(10, 20, data).unwrap();
    let indices = vec![3, 7, 1, 9, 0, 5];

    let r1 = table.embedding_lookup(&indices).unwrap();
    let r2 = table.embedding_lookup(&indices).unwrap();

    // Bit-identical comparison (not approximate)
    assert_eq!(r1.data, r2.data, "FALSIFIED EM-003: embedding lookup is non-deterministic");
}

#[test]
fn falsify_em_003_repeated_index_determinism() {
    let data: Vec<f32> = (0..60).map(|i| i as f32).collect();
    let table = Matrix::from_vec(5, 12, data).unwrap();
    let indices = vec![2, 2, 2];

    let result = table.embedding_lookup(&indices).unwrap();

    // All three rows must be identical (same index)
    for col in 0..12 {
        let v0 = result.get(0, col).unwrap();
        let v1 = result.get(1, col).unwrap();
        let v2 = result.get(2, col).unwrap();
        assert_eq!(v0, v1, "FALSIFIED EM-003: repeated index produced different rows");
        assert_eq!(v1, v2, "FALSIFIED EM-003: repeated index produced different rows");
    }
}

// ============================================================================
// FALSIFY-EM-004: Finite output
// Contract: all output elements are finite when W elements are finite
// ============================================================================

#[test]
fn falsify_em_004_finite_output() {
    let data: Vec<f32> = (0..500).map(|i| (i as f32 * 0.123).sin() * 100.0).collect();
    let table = Matrix::from_vec(25, 20, data).unwrap();
    let indices: Vec<usize> = (0..25).collect();

    let result = table.embedding_lookup(&indices).unwrap();

    for (i, val) in result.data.iter().enumerate() {
        assert!(val.is_finite(), "FALSIFIED EM-004: output[{i}] = {val} is not finite");
    }
}

#[test]
fn falsify_em_004_no_nan_no_inf() {
    let data: Vec<f32> = (0..120).map(|i| (i as f32) * 0.001).collect();
    let table = Matrix::from_vec(10, 12, data).unwrap();
    let indices = vec![0, 5, 9];

    let result = table.embedding_lookup(&indices).unwrap();

    let nan_count = result.data.iter().filter(|v| v.is_nan()).count();
    let inf_count = result.data.iter().filter(|v| v.is_infinite()).count();

    assert_eq!(nan_count, 0, "FALSIFIED EM-004: output contains {} NaN values", nan_count);
    assert_eq!(inf_count, 0, "FALSIFIED EM-004: output contains {} Inf values", inf_count);
}

// ============================================================================
// FALSIFY-EM-005: Embedding value correctness (extractive)
// Contract: output[i] = W[token_ids[i]] (row copy, not transformation)
// ============================================================================

#[test]
fn falsify_em_005_value_correctness() {
    let data: Vec<f32> = (0..40).map(|i| i as f32).collect();
    let table = Matrix::from_vec(10, 4, data).unwrap();
    let indices = vec![3, 7, 0];

    let result = table.embedding_lookup(&indices).unwrap();

    // Row 0 should be table row 3: [12, 13, 14, 15]
    assert_eq!(result.get(0, 0), Some(&12.0));
    assert_eq!(result.get(0, 3), Some(&15.0));

    // Row 1 should be table row 7: [28, 29, 30, 31]
    assert_eq!(result.get(1, 0), Some(&28.0));
    assert_eq!(result.get(1, 3), Some(&31.0));

    // Row 2 should be table row 0: [0, 1, 2, 3]
    assert_eq!(result.get(2, 0), Some(&0.0));
    assert_eq!(result.get(2, 3), Some(&3.0));
}

// ============================================================================
// FALSIFY-EMB-001: Lookup determinism
// Contract: same table + same indices = same output, always
//
// Five-Whys (PMAT-354):
//   Why 1: trueno had EMB-005 but no EMB-001 (determinism is in embedding-algebra-v1)
//   Why 2: EM-003 tests determinism for lookup, but EMB-001 is the algebra contract
//   Why 3: EMB-001 was only tested in aprender
//   Why 4: trueno treated embedding as "just lookup" — no algebra coverage
//   Why 5: nobody mapped embedding-algebra-v1 claims to trueno tests
// ============================================================================

#[test]
fn falsify_emb_001_lookup_determinism() {
    let data: Vec<f32> = (0..200).map(|i| (i as f32 * 0.37).sin()).collect();
    let table = Matrix::from_vec(10, 20, data).unwrap();
    let indices = vec![3, 7, 1, 9, 0, 5];

    let result1 = table.embedding_lookup(&indices).unwrap();
    let result2 = table.embedding_lookup(&indices).unwrap();

    assert_eq!(
        result1.data, result2.data,
        "FALSIFIED EMB-001: identical lookup produced different results"
    );
}

// ============================================================================
// FALSIFY-EMB-002: Shape preservation
// Contract: embed(token_id).shape = [d_model] for any valid token_id
// ============================================================================

#[test]
fn falsify_emb_002_shape_preservation() {
    let dims = [4, 16, 64, 128];
    for &d_model in &dims {
        let table = Matrix::from_vec(50, d_model, vec![1.0; 50 * d_model]).unwrap();

        for &token_id in &[0, 25, 49] {
            let result = table.embedding_lookup(&[token_id]).unwrap();
            assert_eq!(
                result.cols(),
                d_model,
                "FALSIFIED EMB-002: embed({token_id}).cols={}, expected d_model={d_model}",
                result.cols()
            );
            assert_eq!(
                result.rows(),
                1,
                "FALSIFIED EMB-002: embed({token_id}).rows={}, expected 1",
                result.rows()
            );
        }
    }
}

// ============================================================================
// FALSIFY-EMB-004: Vocabulary bounds — OOB IDs rejected
// Contract: token_id >= vocab_size produces error, not panic/garbage
// ============================================================================

#[test]
fn falsify_emb_004_vocabulary_bounds() {
    let vocab_size = 10;
    let table = Matrix::from_vec(vocab_size, 4, vec![1.0; 40]).unwrap();

    // At the boundary: vocab_size-1 is valid
    let result = table.embedding_lookup(&[vocab_size - 1]);
    assert!(result.is_ok(), "FALSIFIED EMB-004: valid index {} rejected", vocab_size - 1);

    // Past the boundary: vocab_size is invalid
    let result = table.embedding_lookup(&[vocab_size]);
    assert!(result.is_err(), "FALSIFIED EMB-004: OOB index {vocab_size} was not rejected");

    // Way past the boundary
    let result = table.embedding_lookup(&[999]);
    assert!(result.is_err(), "FALSIFIED EMB-004: OOB index 999 was not rejected");
}

// ============================================================================
// FALSIFY-EMB-005: Non-zero embeddings
// Contract: embedding table with non-zero values produces non-zero output
//
// Five-Whys (PMAT-354):
//   Why 1: trueno had 0 FALSIFY-EMB-* tests
//   Why 2: EMB contract covers algebra, not just lookup mechanics
//   Why 3: EMB-005 (non-zero) was only tested in aprender
//   Why 4: no cross-stack EMB mapping existed before PMAT-354
//   Why 5: embedding_lookup was treated as "just slicing" — no algebra tests
// ============================================================================

#[test]
fn falsify_emb_005_non_zero_output() {
    // Non-zero embedding table must produce non-zero lookup results
    let data: Vec<f32> = (0..200).map(|i| (i as f32 * 0.37).sin()).collect();
    let table = Matrix::from_vec(10, 20, data).unwrap();
    let indices = vec![3, 7, 1];

    let result = table.embedding_lookup(&indices).unwrap();

    // At least some output values must be non-zero (not a degenerate table)
    let l2_norm: f32 = result.data.iter().map(|v| v * v).sum::<f32>().sqrt();
    assert!(
        l2_norm > 1e-6,
        "FALSIFIED EMB-005: embedding lookup produced all-zero output (L2={l2_norm})"
    );
}

#[test]
fn falsify_emb_005_per_row_non_zero() {
    // Each looked-up row should be non-zero when the source row is non-zero
    let data: Vec<f32> = (1..=60).map(|i| i as f32).collect();
    let table = Matrix::from_vec(5, 12, data).unwrap();

    for idx in 0..5 {
        let result = table.embedding_lookup(&[idx]).unwrap();
        let row_l2: f32 = result.data.iter().map(|v| v * v).sum::<f32>().sqrt();
        assert!(row_l2 > 1e-6, "FALSIFIED EMB-005: row {idx} is all-zero (L2={row_l2})");
    }
}

// ============================================================================
// PROPTEST FALSIFY: Embedding property-based falsification per YAML directives
//
// Five-Whys (PMAT-354, Phase 8):
//   Why 1: YAML calls for "proptest with random..." for EM and EMB claims
//   Why 2: trueno had 12 EM + 5 EMB deterministic tests but zero proptest
//   Why 3: deterministic tests cover hand-picked exemplars, not input space
//   Why 4: proptest generates adversarial edge cases for matrix ops
//   Why 5: Popperian falsification demands maximally adversarial generation
// ============================================================================

mod proptest_falsify {
    use super::*;
    use proptest::prelude::*;

    // FALSIFY-EM-001-prop: Shape correctness for random dimensions
    proptest! {
        #![proptest_config(ProptestConfig::with_cases(100))]
        #[test]
        fn falsify_em_001_prop_output_shape(
            vocab in 10_usize..200,
            d_model in prop::sample::select(vec![4_usize, 16, 32, 64]),
            num_tokens in 1_usize..20,
        ) {
            let data: Vec<f32> = (0..vocab * d_model)
                .map(|i| (i as f32 * 0.13).sin())
                .collect();
            let table = Matrix::from_vec(vocab, d_model, data).unwrap();
            let indices: Vec<usize> = (0..num_tokens).map(|i| i % vocab).collect();
            let result = table.embedding_lookup(&indices).unwrap();
            prop_assert_eq!(result.rows(), num_tokens,
                "FALSIFIED EM-001-prop: rows={} != n_tokens={}", result.rows(), num_tokens);
            prop_assert_eq!(result.cols(), d_model,
                "FALSIFIED EM-001-prop: cols={} != d_model={}", result.cols(), d_model);
        }
    }

    // FALSIFY-EMB-001-prop: Lookup determinism for random indices
    proptest! {
        #![proptest_config(ProptestConfig::with_cases(200))]
        #[test]
        fn falsify_emb_001_prop_determinism(
            idx in 0_usize..99,
        ) {
            let data: Vec<f32> = (0..100 * 32)
                .map(|i| (i as f32 * 0.37).sin())
                .collect();
            let table = Matrix::from_vec(100, 32, data).unwrap();
            let r1 = table.embedding_lookup(&[idx]).unwrap();
            let r2 = table.embedding_lookup(&[idx]).unwrap();
            prop_assert_eq!(r1.data, r2.data,
                "FALSIFIED EMB-001-prop: lookup({}) non-deterministic", idx);
        }
    }

    // FALSIFY-EMB-004-prop: Vocabulary bounds for random vocab sizes
    proptest! {
        #![proptest_config(ProptestConfig::with_cases(100))]
        #[test]
        fn falsify_emb_004_prop_bounds(
            vocab in 5_usize..100,
        ) {
            let data: Vec<f32> = (0..vocab * 8)
                .map(|i| (i as f32 * 0.1).cos())
                .collect();
            let table = Matrix::from_vec(vocab, 8, data).unwrap();

            // Last valid index succeeds
            let valid = table.embedding_lookup(&[vocab - 1]);
            prop_assert!(valid.is_ok(), "FALSIFIED EMB-004-prop: valid idx {} rejected", vocab - 1);

            // First OOB index fails
            let oob = table.embedding_lookup(&[vocab]);
            prop_assert!(oob.is_err(), "FALSIFIED EMB-004-prop: OOB idx {} not rejected", vocab);
        }
    }
}