numrs2 0.3.0

A Rust implementation inspired by NumPy for numerical computing (NumRS2)
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

//! Tests for GPU operations

use numrs2::array::Array;
use numrs2::gpu::{new_context, ops, GpuArray};

#[tokio::test]
async fn test_add() {
    let context = new_context().expect("Failed to create GPU context");

    let a = Array::from_vec(vec![1.0f32, 2.0, 3.0, 4.0]).reshape(&[4]);
    let b = Array::from_vec(vec![5.0f32, 6.0, 7.0, 8.0]).reshape(&[4]);

    let a_gpu = GpuArray::from_array_with_context(&a, context.clone())
        .expect("Failed to create GPU array");
    let b_gpu = GpuArray::from_array_with_context(&b, context.clone())
        .expect("Failed to create GPU array");

    let c_gpu = ops::add(&a_gpu, &b_gpu).expect("Failed to add");
    let c = c_gpu.to_array().expect("Failed to convert to CPU array");

    assert_eq!(c.shape(), &[4]);
    assert!((c.get(&[0]).expect("Invalid index") - 6.0).abs() < 1e-5);
    assert!((c.get(&[1]).expect("Invalid index") - 8.0).abs() < 1e-5);
    assert!((c.get(&[2]).expect("Invalid index") - 10.0).abs() < 1e-5);
    assert!((c.get(&[3]).expect("Invalid index") - 12.0).abs() < 1e-5);
}

#[tokio::test]
async fn test_subtract() {
    let context = new_context().expect("Failed to create GPU context");

    let a = Array::from_vec(vec![10.0f32, 8.0, 6.0, 4.0]).reshape(&[4]);
    let b = Array::from_vec(vec![1.0f32, 2.0, 3.0, 4.0]).reshape(&[4]);

    let a_gpu = GpuArray::from_array_with_context(&a, context.clone())
        .expect("Failed to create GPU array");
    let b_gpu = GpuArray::from_array_with_context(&b, context.clone())
        .expect("Failed to create GPU array");

    let c_gpu = ops::subtract(&a_gpu, &b_gpu).expect("Failed to subtract");
    let c = c_gpu.to_array().expect("Failed to convert to CPU array");

    assert_eq!(c.shape(), &[4]);
    assert!((c.get(&[0]).expect("Invalid index") - 9.0).abs() < 1e-5);
    assert!((c.get(&[1]).expect("Invalid index") - 6.0).abs() < 1e-5);
    assert!((c.get(&[2]).expect("Invalid index") - 3.0).abs() < 1e-5);
    assert!((c.get(&[3]).expect("Invalid index") - 0.0).abs() < 1e-5);
}

#[tokio::test]
async fn test_multiply() {
    let context = new_context().expect("Failed to create GPU context");

    let a = Array::from_vec(vec![2.0f32, 3.0, 4.0, 5.0]).reshape(&[4]);
    let b = Array::from_vec(vec![1.0f32, 2.0, 3.0, 4.0]).reshape(&[4]);

    let a_gpu = GpuArray::from_array_with_context(&a, context.clone())
        .expect("Failed to create GPU array");
    let b_gpu = GpuArray::from_array_with_context(&b, context.clone())
        .expect("Failed to create GPU array");

    let c_gpu = ops::multiply(&a_gpu, &b_gpu).expect("Failed to multiply");
    let c = c_gpu.to_array().expect("Failed to convert to CPU array");

    assert_eq!(c.shape(), &[4]);
    assert!((c.get(&[0]).expect("Invalid index") - 2.0).abs() < 1e-5);
    assert!((c.get(&[1]).expect("Invalid index") - 6.0).abs() < 1e-5);
    assert!((c.get(&[2]).expect("Invalid index") - 12.0).abs() < 1e-5);
    assert!((c.get(&[3]).expect("Invalid index") - 20.0).abs() < 1e-5);
}

#[tokio::test]
async fn test_divide() {
    let context = new_context().expect("Failed to create GPU context");

    let a = Array::from_vec(vec![10.0f32, 20.0, 30.0, 40.0]).reshape(&[4]);
    let b = Array::from_vec(vec![2.0f32, 4.0, 5.0, 8.0]).reshape(&[4]);

    let a_gpu = GpuArray::from_array_with_context(&a, context.clone())
        .expect("Failed to create GPU array");
    let b_gpu = GpuArray::from_array_with_context(&b, context.clone())
        .expect("Failed to create GPU array");

    let c_gpu = ops::divide(&a_gpu, &b_gpu).expect("Failed to divide");
    let c = c_gpu.to_array().expect("Failed to convert to CPU array");

    assert_eq!(c.shape(), &[4]);
    assert!((c.get(&[0]).expect("Invalid index") - 5.0).abs() < 1e-5);
    assert!((c.get(&[1]).expect("Invalid index") - 5.0).abs() < 1e-5);
    assert!((c.get(&[2]).expect("Invalid index") - 6.0).abs() < 1e-5);
    assert!((c.get(&[3]).expect("Invalid index") - 5.0).abs() < 1e-5);
}

#[tokio::test]
async fn test_gpu_array_creation() {
    let context = new_context().expect("Failed to create GPU context");

    let a = Array::from_vec(vec![1.0f32, 2.0, 3.0, 4.0, 5.0, 6.0]).reshape(&[2, 3]);
    let a_gpu = GpuArray::from_array_with_context(&a, context.clone())
        .expect("Failed to create GPU array");

    assert_eq!(a_gpu.shape(), &[2, 3]);
    assert_eq!(a_gpu.size(), 6);
}

#[tokio::test]
async fn test_gpu_array_round_trip() {
    let context = new_context().expect("Failed to create GPU context");

    let original = Array::from_vec(vec![1.0f32, 2.0, 3.0, 4.0, 5.0, 6.0]).reshape(&[2, 3]);

    let gpu_array = GpuArray::from_array_with_context(&original, context.clone())
        .expect("Failed to create GPU array");

    let result = gpu_array.to_array().expect("Failed to convert back to CPU");

    assert_eq!(result.shape(), original.shape());
    for i in 0..2 {
        for j in 0..3 {
            let orig_val = original.get(&[i, j]).expect("Invalid index");
            let result_val = result.get(&[i, j]).expect("Invalid index");
            assert!((orig_val - result_val).abs() < 1e-5);
        }
    }
}

#[tokio::test]
async fn test_gpu_array_new_with_shape() {
    let context = new_context().expect("Failed to create GPU context");

    let gpu_array = GpuArray::<f32>::new_with_shape(&[3, 4], context)
        .expect("Failed to create GPU array with shape");

    assert_eq!(gpu_array.shape(), &[3, 4]);
    assert_eq!(gpu_array.size(), 12);
}

#[tokio::test]
async fn test_incompatible_shapes() {
    let context = new_context().expect("Failed to create GPU context");

    let a = Array::from_vec(vec![1.0f32, 2.0, 3.0]).reshape(&[3]);
    let b = Array::from_vec(vec![1.0f32, 2.0, 3.0, 4.0]).reshape(&[4]);

    let a_gpu = GpuArray::from_array_with_context(&a, context.clone())
        .expect("Failed to create GPU array");
    let b_gpu = GpuArray::from_array_with_context(&b, context.clone())
        .expect("Failed to create GPU array");

    // Operations on arrays with different sizes should fail
    let result = ops::add(&a_gpu, &b_gpu);
    assert!(result.is_err());
}

#[tokio::test]
async fn test_multidimensional_arrays() {
    let context = new_context().expect("Failed to create GPU context");

    // Create 2D arrays
    let a = Array::from_vec(vec![1.0f32, 2.0, 3.0, 4.0]).reshape(&[2, 2]);
    let b = Array::from_vec(vec![5.0f32, 6.0, 7.0, 8.0]).reshape(&[2, 2]);

    let a_gpu = GpuArray::from_array_with_context(&a, context.clone())
        .expect("Failed to create GPU array");
    let b_gpu = GpuArray::from_array_with_context(&b, context.clone())
        .expect("Failed to create GPU array");

    let c_gpu = ops::add(&a_gpu, &b_gpu).expect("Failed to add");
    let c = c_gpu.to_array().expect("Failed to convert to CPU array");

    assert_eq!(c.shape(), &[2, 2]);
    assert!((c.get(&[0, 0]).expect("Invalid index") - 6.0).abs() < 1e-5);
    assert!((c.get(&[0, 1]).expect("Invalid index") - 8.0).abs() < 1e-5);
    assert!((c.get(&[1, 0]).expect("Invalid index") - 10.0).abs() < 1e-5);
    assert!((c.get(&[1, 1]).expect("Invalid index") - 12.0).abs() < 1e-5);
}

#[tokio::test]
async fn test_large_array() {
    let context = new_context().expect("Failed to create GPU context");

    // Create a larger array to test GPU performance benefits
    let size = 1000;
    let data_a: Vec<f32> = (0..size).map(|i| i as f32).collect();
    let data_b: Vec<f32> = (0..size).map(|i| (i as f32) * 2.0).collect();

    let a = Array::from_vec(data_a.clone()).reshape(&[size]);
    let b = Array::from_vec(data_b.clone()).reshape(&[size]);

    let a_gpu = GpuArray::from_array_with_context(&a, context.clone())
        .expect("Failed to create GPU array");
    let b_gpu = GpuArray::from_array_with_context(&b, context.clone())
        .expect("Failed to create GPU array");

    let c_gpu = ops::add(&a_gpu, &b_gpu).expect("Failed to add");
    let c = c_gpu.to_array().expect("Failed to convert to CPU array");

    assert_eq!(c.shape(), &[size]);

    // Verify a few values
    for i in [0, size / 2, size - 1] {
        let expected = data_a[i] + data_b[i];
        let actual = c.get(&[i]).expect("Invalid index");
        assert!((actual - expected).abs() < 1e-5);
    }
}

#[tokio::test]
async fn test_unary_operations() {
    let context = new_context().expect("Failed to create GPU context");

    let a = Array::from_vec(vec![1.0f32, 2.0, 3.0, 4.0]).reshape(&[4]);
    let a_gpu = GpuArray::from_array_with_context(&a, context.clone())
        .expect("Failed to create GPU array");

    // Test exp
    let exp_result = ops::exp(&a_gpu).expect("Failed to compute exp");
    let exp_cpu = exp_result.to_array().expect("Failed to convert to CPU");
    assert!((exp_cpu.get(&[0]).expect("Invalid index") - 1.0f32.exp()).abs() < 1e-5);

    // Test sqrt
    let sqrt_result = ops::sqrt(&a_gpu).expect("Failed to compute sqrt");
    let sqrt_cpu = sqrt_result.to_array().expect("Failed to convert to CPU");
    assert!((sqrt_cpu.get(&[0]).expect("Invalid index") - 1.0f32.sqrt()).abs() < 1e-5);

    // Test sin
    let sin_result = ops::sin(&a_gpu).expect("Failed to compute sin");
    let sin_cpu = sin_result.to_array().expect("Failed to convert to CPU");
    assert!((sin_cpu.get(&[0]).expect("Invalid index") - 1.0f32.sin()).abs() < 1e-5);

    // Test cos
    let cos_result = ops::cos(&a_gpu).expect("Failed to compute cos");
    let cos_cpu = cos_result.to_array().expect("Failed to convert to CPU");
    assert!((cos_cpu.get(&[0]).expect("Invalid index") - 1.0f32.cos()).abs() < 1e-5);
}

#[tokio::test]
async fn test_reduction_operations() {
    let context = new_context().expect("Failed to create GPU context");

    let data = vec![1.0f32, 2.0, 3.0, 4.0, 5.0];
    let a = Array::from_vec(data.clone()).reshape(&[5]);
    let a_gpu = GpuArray::from_array_with_context(&a, context.clone())
        .expect("Failed to create GPU array");

    // Test sum
    let sum = ops::sum_f32(&a_gpu).expect("Failed to compute sum");
    let expected_sum: f32 = data.iter().sum();
    assert!((sum - expected_sum).abs() < 1e-5);

    // Test mean
    let mean = ops::mean_f32(&a_gpu).expect("Failed to compute mean");
    let expected_mean = expected_sum / data.len() as f32;
    assert!((mean - expected_mean).abs() < 1e-5);

    // Test max
    let max_val = ops::max_f32(&a_gpu).expect("Failed to compute max");
    let expected_max = data.iter().cloned().fold(f32::NEG_INFINITY, f32::max);
    assert!((max_val - expected_max).abs() < 1e-5);

    // Test min
    let min_val = ops::min_f32(&a_gpu).expect("Failed to compute min");
    let expected_min = data.iter().cloned().fold(f32::INFINITY, f32::min);
    assert!((min_val - expected_min).abs() < 1e-5);
}

#[tokio::test]
async fn test_copy_operation() {
    let context = new_context().expect("Failed to create GPU context");

    let a = Array::from_vec(vec![1.0f32, 2.0, 3.0, 4.0]).reshape(&[4]);
    let a_gpu = GpuArray::from_array_with_context(&a, context.clone())
        .expect("Failed to create GPU array");

    let b_gpu = ops::copy_with_format(&a_gpu).expect("Failed to copy array");
    let b = b_gpu.to_array().expect("Failed to convert to CPU");

    assert_eq!(a.shape(), b.shape());
    for i in 0..4 {
        let a_val = a.get(&[i]).expect("Invalid index");
        let b_val = b.get(&[i]).expect("Invalid index");
        assert!((a_val - b_val).abs() < 1e-10);
    }
}

#[tokio::test]
async fn test_transpose_operation() {
    let context = new_context().expect("Failed to create GPU context");

    let a = Array::from_vec(vec![1.0f32, 2.0, 3.0, 4.0, 5.0, 6.0]).reshape(&[2, 3]);
    let a_gpu = GpuArray::from_array_with_context(&a, context.clone())
        .expect("Failed to create GPU array");

    let b_gpu = ops::transpose(&a_gpu).expect("Failed to transpose array");
    assert_eq!(b_gpu.shape(), &[3, 2]);

    let b = b_gpu.to_array().expect("Failed to convert to CPU");

    // Check a few values
    assert!((b.get(&[0, 0]).expect("Invalid index") - a.get(&[0, 0]).expect("Invalid index")).abs() < 1e-5);
    assert!((b.get(&[0, 1]).expect("Invalid index") - a.get(&[1, 0]).expect("Invalid index")).abs() < 1e-5);
    assert!((b.get(&[1, 0]).expect("Invalid index") - a.get(&[0, 1]).expect("Invalid index")).abs() < 1e-5);
}

#[tokio::test]
async fn test_broadcast_shapes_compatible() {
    // Test broadcasting shape calculation
    // This is implicitly tested by broadcast_add, but we test the logic here

    let context = new_context().expect("Failed to create GPU context");

    // Same shapes - should work
    let a = Array::from_vec(vec![1.0f32, 2.0, 3.0, 4.0]).reshape(&[4]);
    let b = Array::from_vec(vec![5.0f32, 6.0, 7.0, 8.0]).reshape(&[4]);

    let a_gpu = GpuArray::from_array_with_context(&a, context.clone())
        .expect("Failed to create GPU array");
    let b_gpu = GpuArray::from_array_with_context(&b, context.clone())
        .expect("Failed to create GPU array");

    // For now, broadcast operations with same shape should fall back to regular ops
    let c_gpu = ops::broadcast_add(&a_gpu, &b_gpu).expect("Failed to broadcast add");
    let c = c_gpu.to_array().expect("Failed to convert to CPU");

    assert_eq!(c.shape(), &[4]);
}

#[tokio::test]
async fn test_pow_operation() {
    let context = new_context().expect("Failed to create GPU context");

    let a = Array::from_vec(vec![2.0f32, 3.0, 4.0, 5.0]).reshape(&[4]);
    let b = Array::from_vec(vec![2.0f32, 2.0, 2.0, 2.0]).reshape(&[4]);

    let a_gpu = GpuArray::from_array_with_context(&a, context.clone())
        .expect("Failed to create GPU array");
    let b_gpu = GpuArray::from_array_with_context(&b, context.clone())
        .expect("Failed to create GPU array");

    let c_gpu = ops::pow(&a_gpu, &b_gpu).expect("Failed to compute pow");
    let c = c_gpu.to_array().expect("Failed to convert to CPU");

    assert_eq!(c.shape(), &[4]);
    assert!((c.get(&[0]).expect("Invalid index") - 4.0).abs() < 1e-5);
    assert!((c.get(&[1]).expect("Invalid index") - 9.0).abs() < 1e-5);
    assert!((c.get(&[2]).expect("Invalid index") - 16.0).abs() < 1e-5);
    assert!((c.get(&[3]).expect("Invalid index") - 25.0).abs() < 1e-5);
}

#[tokio::test]
async fn test_abs_and_neg_operations() {
    let context = new_context().expect("Failed to create GPU context");

    let a = Array::from_vec(vec![-1.0f32, 2.0, -3.0, 4.0]).reshape(&[4]);
    let a_gpu = GpuArray::from_array_with_context(&a, context.clone())
        .expect("Failed to create GPU array");

    // Test abs
    let abs_result = ops::abs(&a_gpu).expect("Failed to compute abs");
    let abs_cpu = abs_result.to_array().expect("Failed to convert to CPU");
    assert!((abs_cpu.get(&[0]).expect("Invalid index") - 1.0).abs() < 1e-5);
    assert!((abs_cpu.get(&[2]).expect("Invalid index") - 3.0).abs() < 1e-5);

    // Test neg
    let neg_result = ops::neg(&a_gpu).expect("Failed to compute neg");
    let neg_cpu = neg_result.to_array().expect("Failed to convert to CPU");
    assert!((neg_cpu.get(&[0]).expect("Invalid index") - 1.0).abs() < 1e-5);
    assert!((neg_cpu.get(&[1]).expect("Invalid index") + 2.0).abs() < 1e-5);
}