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
//! Element-wise addition for COO matrices
use super::super::CooData;
use crate::dtype::DType;
use crate::error::{Error, Result};
use crate::runtime::Runtime;
use crate::sparse::{SparseOps, SparseStorage};
impl<R: Runtime<DType = DType>> CooData<R> {
/// Element-wise addition: C = A + B
///
/// Computes the sum of two sparse matrices with the same shape.
///
/// # Arguments
///
/// * `other` - Another COO matrix with the same shape and dtype
///
/// # Returns
///
/// A new COO matrix containing the element-wise sum (sorted by row, then column)
///
/// # Errors
///
/// Returns error if:
/// - Shapes don't match
/// - Dtypes don't match
///
/// # Algorithm
///
/// Concatenates triplets from both matrices, sorts, and merges duplicates.
/// GPU-accelerated when CUDA runtime is used.
///
/// # Performance
///
/// - CPU: O((nnz_a + nnz_b) log(nnz_a + nnz_b)) for sorting
/// - GPU: O((nnz_a + nnz_b) log(nnz_a + nnz_b)) parallel sort-merge
///
/// # Example
///
/// ```
/// # use numr::prelude::*;
/// # #[cfg(feature = "sparse")]
/// # {
/// # use numr::sparse::SparseTensor;
/// # let device = CpuDevice::new();
/// // A: B: C = A + B:
/// // [1, 0] [0, 2] [1, 2]
/// // [0, 3] + [4, 0] = [4, 3]
/// # let a_sp = SparseTensor::<CpuRuntime>::from_coo_slices(&[0, 1], &[0, 1], &[1.0f32, 3.0], [2, 2], &device)?;
/// # let b_sp = SparseTensor::<CpuRuntime>::from_coo_slices(&[0, 1], &[1, 0], &[2.0f32, 4.0], [2, 2], &device)?;
/// # if let numr::sparse::SparseTensor::Coo(a) = a_sp { if let numr::sparse::SparseTensor::Coo(b) = b_sp {
/// let c = a.add(&b)?;
/// # } }
/// # }
/// # Ok::<(), numr::error::Error>(())
/// ```
pub fn add(&self, other: &Self) -> Result<Self>
where
R::Client: SparseOps<R>,
{
// Validate shapes match
if self.shape != other.shape {
return Err(Error::ShapeMismatch {
expected: vec![self.shape[0], self.shape[1]],
got: vec![other.shape[0], other.shape[1]],
});
}
// Validate dtypes match
if self.dtype() != other.dtype() {
return Err(Error::DTypeMismatch {
lhs: self.dtype(),
rhs: other.dtype(),
});
}
let dtype = self.dtype();
let device = self.values.device();
// Get client for runtime dispatch
let client = R::default_client(device);
// Dispatch to runtime-specific implementation
crate::dispatch_dtype!(dtype, T => {
let (out_row_indices, out_col_indices, out_values) = client.add_coo::<T>(
&self.row_indices,
&self.col_indices,
&self.values,
&other.row_indices,
&other.col_indices,
&other.values,
self.shape,
)?;
Ok(Self {
row_indices: out_row_indices,
col_indices: out_col_indices,
values: out_values,
shape: self.shape,
sorted: true, // Backend guarantees sorted output
})
}, "coo_add")
}
}
#[cfg(test)]
mod tests {
use super::*;
use crate::dtype::DType;
use crate::runtime::cpu::CpuRuntime;
#[test]
fn test_coo_add_disjoint() {
let device = <CpuRuntime as Runtime>::Device::default();
// A: B:
// [1, 0] [0, 2]
// [0, 3] [4, 0]
let a = CooData::<CpuRuntime>::from_slices(
&[0i64, 1],
&[0i64, 1],
&[1.0f32, 3.0],
[2, 2],
&device,
)
.unwrap();
let b = CooData::<CpuRuntime>::from_slices(
&[0i64, 1],
&[1i64, 0],
&[2.0f32, 4.0],
[2, 2],
&device,
)
.unwrap();
let c = a.add(&b).unwrap();
// C = A + B:
// [1, 2]
// [4, 3]
assert_eq!(c.shape(), [2, 2]);
assert_eq!(c.nnz(), 4);
assert!(c.is_sorted());
let rows: Vec<i64> = c.row_indices().to_vec();
let cols: Vec<i64> = c.col_indices().to_vec();
let vals: Vec<f32> = c.values().to_vec();
assert_eq!(rows, vec![0, 0, 1, 1]);
assert_eq!(cols, vec![0, 1, 0, 1]);
assert_eq!(vals, vec![1.0, 2.0, 4.0, 3.0]);
}
#[test]
fn test_coo_add_overlapping() {
let device = <CpuRuntime as Runtime>::Device::default();
// A: B:
// [1, 2] [3, 0]
// [0, 0] [0, 4]
let a = CooData::<CpuRuntime>::from_slices(
&[0i64, 0],
&[0i64, 1],
&[1.0f32, 2.0],
[2, 2],
&device,
)
.unwrap();
let b = CooData::<CpuRuntime>::from_slices(
&[0i64, 1],
&[0i64, 1],
&[3.0f32, 4.0],
[2, 2],
&device,
)
.unwrap();
let c = a.add(&b).unwrap();
// C = A + B:
// [4, 2] (1+3=4 at (0,0))
// [0, 4]
assert_eq!(c.nnz(), 3);
let rows: Vec<i64> = c.row_indices().to_vec();
let cols: Vec<i64> = c.col_indices().to_vec();
let vals: Vec<f32> = c.values().to_vec();
assert_eq!(rows, vec![0, 0, 1]);
assert_eq!(cols, vec![0, 1, 1]);
assert_eq!(vals, vec![4.0, 2.0, 4.0]);
}
#[test]
fn test_coo_add_empty_a() {
let device = <CpuRuntime as Runtime>::Device::default();
let a = CooData::<CpuRuntime>::empty([2, 2], DType::F32, &device);
let b = CooData::<CpuRuntime>::from_slices(
&[0i64, 1],
&[0i64, 1],
&[1.0f32, 2.0],
[2, 2],
&device,
)
.unwrap();
let c = a.add(&b).unwrap();
assert_eq!(c.nnz(), 2);
}
#[test]
fn test_coo_add_empty_b() {
let device = <CpuRuntime as Runtime>::Device::default();
let a = CooData::<CpuRuntime>::from_slices(
&[0i64, 1],
&[0i64, 1],
&[1.0f32, 2.0],
[2, 2],
&device,
)
.unwrap();
let b = CooData::<CpuRuntime>::empty([2, 2], DType::F32, &device);
let c = a.add(&b).unwrap();
assert_eq!(c.nnz(), 2);
}
#[test]
fn test_coo_add_shape_mismatch() {
let device = <CpuRuntime as Runtime>::Device::default();
let a = CooData::<CpuRuntime>::empty([2, 3], DType::F32, &device);
let b = CooData::<CpuRuntime>::empty([3, 2], DType::F32, &device);
let result = a.add(&b);
assert!(result.is_err());
}
#[test]
fn test_coo_add_dtype_mismatch() {
let device = <CpuRuntime as Runtime>::Device::default();
let a = CooData::<CpuRuntime>::empty([2, 2], DType::F32, &device);
let b = CooData::<CpuRuntime>::empty([2, 2], DType::F64, &device);
let result = a.add(&b);
assert!(result.is_err());
}
#[test]
fn test_coo_add_f64() {
let device = <CpuRuntime as Runtime>::Device::default();
let a = CooData::<CpuRuntime>::from_slices(&[0i64], &[0i64], &[1.5f64], [2, 2], &device)
.unwrap();
let b = CooData::<CpuRuntime>::from_slices(&[0i64], &[0i64], &[2.5f64], [2, 2], &device)
.unwrap();
let c = a.add(&b).unwrap();
assert_eq!(c.dtype(), DType::F64);
let vals: Vec<f64> = c.values().to_vec();
assert_eq!(vals, vec![4.0]);
}
}