leto 0.45.2

N-dimensional strided arrays, zero-copy views, and storage backends
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

//! Value-semantic tests for elementwise operators (ADR 0004).

#![expect(
    clippy::unwrap_used,
    reason = "test scope: failed precondition = test failure"
)]

use leto::{Array2, Storage};

fn assert_reduced_precision_scalar_arithmetic<T>()
where
    T: leto::ScalarOperand
        + eunomia::FloatElement
        + core::fmt::Debug
        + core::ops::Add<Output = T>
        + core::ops::Mul<Output = T>,
{
    let value = |input| <T as eunomia::FloatElement>::from_f32(input);
    let array = Array2::from_shape_vec([1, 2], vec![value(1.0), value(2.0)]).unwrap();

    assert_eq!(
        (&array + value(2.0)).storage().as_slice(),
        &[value(3.0), value(4.0)]
    );
    assert_eq!(
        (&array * value(2.0)).storage().as_slice(),
        &[value(2.0), value(4.0)]
    );
}

fn seq(len: usize, scale: f64, bias: f64) -> Vec<f64> {
    (0..len).map(|i| i as f64 * scale + bias).collect()
}

fn leto_pair() -> (Array2<f64>, Array2<f64>) {
    let a = Array2::from_shape_vec([3, 4], seq(12, 0.5, 1.0)).unwrap();
    let b = Array2::from_shape_vec([3, 4], seq(12, 0.25, 2.0)).unwrap();
    (a, b)
}

/// Reference elementwise add using plain Rust loops (replaces external oracle).
fn ref_add(a: &[f64], b: &[f64]) -> Vec<f64> {
    a.iter().zip(b.iter()).map(|(x, y)| x + y).collect()
}

fn ref_sub(a: &[f64], b: &[f64]) -> Vec<f64> {
    a.iter().zip(b.iter()).map(|(x, y)| x - y).collect()
}

fn ref_mul(a: &[f64], b: &[f64]) -> Vec<f64> {
    a.iter().zip(b.iter()).map(|(x, y)| x * y).collect()
}

fn ref_div(a: &[f64], b: &[f64]) -> Vec<f64> {
    a.iter().zip(b.iter()).map(|(x, y)| x / y).collect()
}

fn ref_scalar_add(a: &[f64], s: f64) -> Vec<f64> {
    a.iter().map(|x| x + s).collect()
}

fn ref_scalar_sub(a: &[f64], s: f64) -> Vec<f64> {
    a.iter().map(|x| x - s).collect()
}

fn ref_scalar_mul(a: &[f64], s: f64) -> Vec<f64> {
    a.iter().map(|x| x * s).collect()
}

fn ref_scalar_div(a: &[f64], s: f64) -> Vec<f64> {
    a.iter().map(|x| x / s).collect()
}

fn ref_neg(a: &[f64]) -> Vec<f64> {
    a.iter().map(|x| -x).collect()
}

#[test]
fn array_array_operators_match_reference() {
    let (la, lb) = leto_pair();
    let a_slice = la.storage().as_slice();
    let b_slice = lb.storage().as_slice();

    assert_eq!((&la + &lb).storage().as_slice(), ref_add(a_slice, b_slice));
    assert_eq!((&la - &lb).storage().as_slice(), ref_sub(a_slice, b_slice));
    assert_eq!((&la * &lb).storage().as_slice(), ref_mul(a_slice, b_slice));
    assert_eq!((&la / &lb).storage().as_slice(), ref_div(a_slice, b_slice));
}

#[test]
fn mul_is_elementwise_not_matmul() {
    //  is the Hadamard product: shape is preserved and each element is the
    // product of the corresponding inputs (ADR 0004).
    let a = Array2::from_shape_vec([2, 2], vec![1.0, 2.0, 3.0, 4.0]).unwrap();
    let b = Array2::from_shape_vec([2, 2], vec![5.0, 6.0, 7.0, 8.0]).unwrap();
    let c = &a * &b;
    assert_eq!(c.shape(), [2, 2]);
    assert_eq!(c.storage().as_slice(), &[5.0, 12.0, 21.0, 32.0]);
}

#[test]
fn scalar_operators_match_reference() {
    let (la, _) = leto_pair();
    let a_slice = la.storage().as_slice();

    assert_eq!(
        (&la + 2.0).storage().as_slice(),
        ref_scalar_add(a_slice, 2.0)
    );
    assert_eq!(
        (&la - 1.5).storage().as_slice(),
        ref_scalar_sub(a_slice, 1.5)
    );
    assert_eq!(
        (&la * 3.0).storage().as_slice(),
        ref_scalar_mul(a_slice, 3.0)
    );
    assert_eq!(
        (&la / 2.0).storage().as_slice(),
        ref_scalar_div(a_slice, 2.0)
    );
}

#[test]
fn eunomia_reduced_precision_types_are_scalar_operands() {
    assert_reduced_precision_scalar_arithmetic::<eunomia::F16>();
    assert_reduced_precision_scalar_arithmetic::<eunomia::Bf16>();
}

#[test]
fn neg_matches_reference() {
    let (la, _) = leto_pair();
    assert_eq!(
        (-&la).storage().as_slice(),
        ref_neg(la.storage().as_slice())
    );
}

#[test]
fn operators_compose() {
    //  exercises chaining of scalar, binary, and unary ops.
    let (la, lb) = leto_pair();
    let a_slice = la.storage().as_slice();
    let b_slice = lb.storage().as_slice();
    let leto = -&(&(&la * 2.0) + &lb);
    let expected = ref_neg(&ref_add(&ref_scalar_mul(a_slice, 2.0), b_slice));
    assert_eq!(leto.storage().as_slice(), &expected);
}

#[test]
fn integer_arrays_support_operators() {
    // Operators are generic over the element type, not f64-specific.
    let a = Array2::from_shape_vec([2, 2], vec![1_i32, 2, 3, 4]).unwrap();
    let b = Array2::from_shape_vec([2, 2], vec![10_i32, 20, 30, 40]).unwrap();
    assert_eq!((&a + &b).storage().as_slice(), &[11, 22, 33, 44]);
    assert_eq!((&a * 2).storage().as_slice(), &[2, 4, 6, 8]);
}

#[test]
#[should_panic(expected = "equal shapes")]
fn shape_mismatch_panics() {
    let a = Array2::from_shape_vec([2, 2], vec![1.0, 2.0, 3.0, 4.0]).unwrap();
    let b = Array2::from_shape_vec([2, 3], vec![1.0, 2.0, 3.0, 4.0, 5.0, 6.0]).unwrap();
    let _ = &a + &b;
}

#[test]
#[should_panic(expected = "equal shapes")]
fn owned_lhs_shape_mismatch_panics() {
    let a = Array2::from_shape_vec([2, 2], vec![1.0, 2.0, 3.0, 4.0]).unwrap();
    let b = Array2::from_shape_vec([2, 3], vec![1.0, 2.0, 3.0, 4.0, 5.0, 6.0]).unwrap();
    let _ = a + &b;
}

#[test]
fn owned_lhs_operators_match_borrowed_lhs() {
    // The owned receiver writes into its own allocation; the observable value
    // must equal the allocating borrowed form element for element.
    let (la, lb) = leto_pair();
    let a_slice = la.storage().as_slice().to_vec();
    let b_slice = lb.storage().as_slice().to_vec();

    assert_eq!(
        (la.clone() + &lb).storage().as_slice(),
        ref_add(&a_slice, &b_slice)
    );
    assert_eq!(
        (la.clone() - &lb).storage().as_slice(),
        ref_sub(&a_slice, &b_slice)
    );
    assert_eq!(
        (la.clone() * &lb).storage().as_slice(),
        ref_mul(&a_slice, &b_slice)
    );
    assert_eq!(
        (la.clone() / &lb).storage().as_slice(),
        ref_div(&a_slice, &b_slice)
    );

    // Both operands owned: the left allocation is kept, the right dropped.
    assert_eq!(
        (la.clone() + lb.clone()).storage().as_slice(),
        ref_add(&a_slice, &b_slice)
    );
}

#[test]
fn owned_lhs_scalar_and_neg_match_borrowed_lhs() {
    let (la, _) = leto_pair();
    let a_slice = la.storage().as_slice().to_vec();

    assert_eq!(
        (la.clone() + 2.0).storage().as_slice(),
        ref_scalar_add(&a_slice, 2.0)
    );
    assert_eq!(
        (la.clone() - 1.5).storage().as_slice(),
        ref_scalar_sub(&a_slice, 1.5)
    );
    assert_eq!(
        (la.clone() * 3.0).storage().as_slice(),
        ref_scalar_mul(&a_slice, 3.0)
    );
    assert_eq!(
        (la.clone() / 2.0).storage().as_slice(),
        ref_scalar_div(&a_slice, 2.0)
    );
    assert_eq!((-la).storage().as_slice(), ref_neg(&a_slice));
}

#[test]
fn owned_lhs_reuses_the_left_allocation() {
    // The buffer-reuse contract, observed directly: the owned operator returns
    // the same heap allocation it consumed, so a chain adds no allocation.
    let (la, lb) = leto_pair();
    let before = la.storage().as_slice().as_ptr();
    let sum = la + &lb;
    assert_eq!(
        sum.storage().as_slice().as_ptr(),
        before,
        "owned lhs must write into its existing allocation"
    );
}

#[test]
fn owned_lhs_accepts_a_strided_right_operand() {
    // Right operand is a transposed (non-C-contiguous) source, so the owned
    // route walks the logical-order element iterator rather than a slice.
    let a = Array2::from_shape_vec([2, 3], vec![1.0, 2.0, 3.0, 4.0, 5.0, 6.0]).unwrap();
    let b = Array2::from_shape_vec([3, 2], vec![10.0, 20.0, 30.0, 40.0, 50.0, 60.0]).unwrap();
    let b_t = b.transpose([1, 0]).unwrap();
    let expected: Vec<f64> = a
        .storage()
        .as_slice()
        .iter()
        .zip(b_t.iter())
        .map(|(x, y)| x + y)
        .collect();
    let owned = a + &b_t.to_contiguous();
    assert_eq!(owned.storage().as_slice(), &expected);
}

#[test]
fn chained_operators_compose_without_intermediate_refs() {
    // The owned receiver makes the natural chain syntax type-check: the middle
    // term is an owned temporary reused in place rather than re-borrowed.
    let (la, lb) = leto_pair();
    let lc = Array2::from_shape_vec([3, 4], seq(12, 0.125, 0.5)).unwrap();
    let a_slice = la.storage().as_slice().to_vec();
    let b_slice = lb.storage().as_slice().to_vec();
    let c_slice = lc.storage().as_slice().to_vec();

    let chained = &la + &lb + &lc;
    let expected = ref_add(&ref_add(&a_slice, &b_slice), &c_slice);
    assert_eq!(chained.storage().as_slice(), &expected);
}