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
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

//! Tests for BrickTuner evolution methods, CalibrationResult, and bandit edge cases.

use super::super::super::*;

#[test]
fn online_learner_is_converging_after_large_error() {
    let mut learner = OnlineLearner::new();
    let features = TunerFeatures::builder().model_params_b(1.5).batch_size(4).build();
    let vec = features.to_vector();

    // Observe with very different target to create large error
    learner.observe(&vec, 10000.0);
    // After a large error, ema_loss should be > 0.15
    // Whether it exceeds threshold depends on exact prediction vs target gap
    // Just verify the method runs without panic
    let _ = learner.is_converging();
}

// ============================================================================
// BrickTuner Evolution Methods Tests
// ============================================================================

#[test]
fn brick_tuner_with_pretrained_has_weights() {
    let tuner = BrickTuner::with_pretrained();
    assert!(tuner.version().contains("pretrained"));
    assert!(tuner.throughput_sample_count() > 0);
    assert!(tuner.throughput_mape() > 0.0);
}

#[test]
fn brick_tuner_with_pretrained_mape() {
    let tuner = BrickTuner::with_pretrained();
    assert!((tuner.throughput_mape() - 0.082).abs() < 0.001, "Pretrained MAPE should be 8.2%");
}

#[test]
fn brick_tuner_with_pretrained_sample_count() {
    let tuner = BrickTuner::with_pretrained();
    assert_eq!(tuner.throughput_sample_count(), 10_000);
}

#[test]
fn brick_tuner_online_learner_inherits_weights() {
    let tuner = BrickTuner::new();
    let learner = tuner.online_learner();
    assert_eq!(learner.weights().len(), TunerFeatures::DIM + 1);
    assert_eq!(learner.num_updates(), 0);
}

#[test]
fn brick_tuner_apply_online_updates_no_updates() {
    let mut tuner = BrickTuner::new();
    let original_version = tuner.version().to_string();
    let learner = OnlineLearner::new();
    tuner.apply_online_updates(&learner);
    // No updates => version should not change
    assert_eq!(tuner.version(), original_version);
}

#[test]
fn brick_tuner_apply_online_updates_with_observations() {
    let mut tuner = BrickTuner::new();
    let mut learner = tuner.online_learner();

    let features = TunerFeatures::builder().model_params_b(1.5).batch_size(4).build();
    let vec = features.to_vector();

    learner.observe(&vec, 100.0);
    learner.observe(&vec, 120.0);

    tuner.apply_online_updates(&learner);
    assert!(tuner.version().contains("online-2"));
}

#[test]
fn brick_tuner_kernel_bandit_returns_new_bandit() {
    let tuner = BrickTuner::new();
    let bandit = tuner.kernel_bandit();
    assert_eq!(bandit.total_pulls, 0);
    assert_eq!(bandit.arms.len(), KernelBandit::NUM_KERNELS);
}

#[test]
fn brick_tuner_recommend_with_exploration_exploit() {
    let tuner = BrickTuner::with_pretrained();
    let bandit = KernelBandit::new();
    let features = TunerFeatures::builder().model_params_b(1.5).batch_size(4).build();

    // With explore_prob = 0.0, should always exploit (model prediction)
    let rec = tuner.recommend_kernel_with_exploration(&features, &bandit, 0.0);
    assert!(rec.confidence >= 0.0);
}

#[test]
fn brick_tuner_recommend_with_exploration_explore() {
    let tuner = BrickTuner::with_pretrained();
    let bandit = KernelBandit::new();
    let features = TunerFeatures::builder().model_params_b(1.5).batch_size(4).build();

    // With explore_prob = 1.0, should always explore
    let rec = tuner.recommend_kernel_with_exploration(&features, &bandit, 1.0);
    // When exploring, confidence should be 0.5
    // (May or may not trigger depending on hash; test no panic)
    assert!(rec.confidence >= 0.0);
}

// ============================================================================
// CalibrationResult Tests
// ============================================================================

#[test]
fn calibration_result_fields_accessible() {
    let result = CalibrationResult {
        throughput_weights: vec![0.1, 0.2, 0.3],
        local_mape: 0.05,
        improvement_pct: 15.0,
        hardware_id: "RTX4090".to_string(),
        duration_secs: 2.5,
        num_benchmarks: 24,
    };

    assert_eq!(result.throughput_weights.len(), 3);
    assert!((result.local_mape - 0.05).abs() < 1e-6);
    assert!((result.improvement_pct - 15.0).abs() < 1e-6);
    assert_eq!(result.hardware_id, "RTX4090");
    assert!((result.duration_secs - 2.5).abs() < 1e-6);
    assert_eq!(result.num_benchmarks, 24);
}

#[test]
fn calibration_result_clone() {
    let result = CalibrationResult {
        throughput_weights: vec![0.1, 0.2],
        local_mape: 0.08,
        improvement_pct: 10.0,
        hardware_id: "A100".to_string(),
        duration_secs: 1.0,
        num_benchmarks: 12,
    };
    let cloned = result.clone();
    assert_eq!(cloned.throughput_weights, result.throughput_weights);
    assert_eq!(cloned.hardware_id, result.hardware_id);
}

#[test]
fn calibration_result_debug_format() {
    let result = CalibrationResult {
        throughput_weights: vec![0.1],
        local_mape: 0.05,
        improvement_pct: 20.0,
        hardware_id: "test".to_string(),
        duration_secs: 0.5,
        num_benchmarks: 6,
    };
    let debug = format!("{:?}", result);
    assert!(debug.contains("CalibrationResult"));
    assert!(debug.contains("local_mape"));
}

#[test]
fn calibration_result_serde_roundtrip() {
    let result = CalibrationResult {
        throughput_weights: vec![0.1, 0.2, 0.3],
        local_mape: 0.07,
        improvement_pct: 12.5,
        hardware_id: "RTX3090".to_string(),
        duration_secs: 3.0,
        num_benchmarks: 24,
    };
    let json = serde_json::to_string(&result).expect("serialize should succeed");
    let deserialized: CalibrationResult =
        serde_json::from_str(&json).expect("deserialize should succeed");
    assert_eq!(deserialized.throughput_weights, result.throughput_weights);
    assert_eq!(deserialized.hardware_id, result.hardware_id);
    assert_eq!(deserialized.num_benchmarks, result.num_benchmarks);
}

// ============================================================================
// KernelBandit Edge Cases
// ============================================================================

#[test]
fn kernel_bandit_select_after_single_update() {
    let mut bandit = KernelBandit::new();
    bandit.update(KernelType::TiledQ4K, 0.5);

    // After one update, all other arms have INFINITY UCB, so they should be preferred
    let selected = bandit.select();
    assert_ne!(selected, KernelType::TiledQ4K, "Should explore unexplored arms first");
}

#[test]
fn kernel_bandit_thompson_after_updates() {
    let mut bandit = KernelBandit::with_thompson_sampling();
    // Give one arm high reward
    for _ in 0..20 {
        bandit.update(KernelType::BatchedQ4K, 0.95);
    }
    // Give another arm low reward
    for _ in 0..20 {
        bandit.update(KernelType::TiledQ4K, 0.1);
    }

    // Thompson sampling should generally prefer the high-reward arm
    // (but is stochastic, so we just check it doesn't panic)
    let selected = bandit.select();
    let _ = selected;
}

#[test]
fn kernel_bandit_all_arms_explored_select_best() {
    let mut bandit = KernelBandit::new();
    // Explore all 12 arms with varying rewards
    for i in 0..KernelBandit::NUM_KERNELS {
        let kernel = KernelType::from_index(i);
        let reward = if i == 3 { 0.95 } else { 0.2 };
        // Pull each arm multiple times to reduce UCB bonus
        for _ in 0..50 {
            bandit.update(kernel, reward);
        }
    }

    // Best kernel should be index 3 (BatchedQ4K with highest reward)
    let best = bandit.best_kernel();
    assert_eq!(best, KernelType::BatchedQ4K);
}

#[test]
fn kernel_bandit_exploration_rate_partial() {
    let mut bandit = KernelBandit::new();
    // 6 pulls on one arm, 4 on another
    for _ in 0..6 {
        bandit.update(KernelType::TiledQ4K, 0.5);
    }
    for _ in 0..4 {
        bandit.update(KernelType::BatchedQ4K, 0.7);
    }
    // best_pulls = 6, total = 10, rate = 1 - 6/10 = 0.4
    assert!((bandit.exploration_rate() - 0.4).abs() < 1e-6);
}

// ============================================================================
// OnlineLearner Edge Cases
// ============================================================================

#[test]
fn online_learner_weights_length_matches_dim() {
    let learner = OnlineLearner::new();
    // DIM features + 1 bias
    assert_eq!(learner.weights().len(), TunerFeatures::DIM + 1);
}

#[test]
fn online_learner_predict_with_all_zeros() {
    let learner = OnlineLearner::new();
    let zeros = vec![0.0; TunerFeatures::DIM];
    let prediction = learner.predict(&zeros);
    // Should just return the bias (clamped to >= 0)
    assert!(prediction >= 0.0);
    // The bias is the first pretrained weight (0.36)
    assert!((prediction - 0.36).abs() < 0.01);
}

#[test]
fn online_learner_predict_with_all_ones() {
    let learner = OnlineLearner::new();
    let ones = vec![1.0; TunerFeatures::DIM];
    let prediction = learner.predict(&ones);
    assert!(prediction >= 0.0, "Prediction must be non-negative");
}

#[test]
fn online_learner_observe_multiple_different_targets() {
    let mut learner = OnlineLearner::new().with_learning_rate(0.001);

    let features_small = TunerFeatures::builder().model_params_b(1.5).batch_size(1).build();
    let features_large = TunerFeatures::builder().model_params_b(13.0).batch_size(8).build();

    let vec_small = features_small.to_vector();
    let vec_large = features_large.to_vector();

    learner.observe(&vec_small, 200.0);
    learner.observe(&vec_large, 50.0);

    assert_eq!(learner.num_updates(), 2);
    assert!(learner.ema_loss() > 0.0);
}

// ============================================================================
// BrickTuner with_pretrained + Evolution Integration
// ============================================================================

#[test]
fn pretrained_then_online_workflow() {
    let mut tuner = BrickTuner::with_pretrained();
    let mut learner = tuner.online_learner();

    let features = TunerFeatures::builder().model_params_b(7.0).batch_size(1).build();
    let vec = features.to_vector();

    // Simulate online learning loop
    for i in 0..5 {
        let target = 140.0 + (i as f32) * 2.0;
        learner.observe(&vec, target);
    }

    tuner.apply_online_updates(&learner);
    assert!(tuner.version().contains("online-5"));
    assert_eq!(tuner.throughput_sample_count(), 10_005); // pretrained 10000 + 5
}

#[test]
fn pretrained_then_bandit_workflow() {
    let tuner = BrickTuner::with_pretrained();
    let mut bandit = tuner.kernel_bandit();

    // Simulate bandit exploration
    let features = TunerFeatures::builder().model_params_b(1.5).batch_size(4).build();

    let kernel = tuner.recommend_kernel_with_exploration(&features, &bandit, 0.3).top_kernel;
    bandit.update(kernel, 0.8);

    assert_eq!(bandit.total_pulls, 1);
}

#[test]
fn kernel_from_index_roundtrip() {
    for i in 0..KernelBandit::NUM_KERNELS {
        let kernel = KernelType::from_index(i);
        let back = kernel.to_index();
        assert_eq!(back, i, "KernelType::from_index({}).to_index() should be {}", i, i);
    }
}