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
//! Element-wise division for CSC matrices
use super::super::CscData;
use crate::dtype::{DType, Element};
use crate::error::{Error, Result};
use crate::runtime::Runtime;
use crate::sparse::SparseStorage;
use crate::tensor::Tensor;
impl<R: Runtime<DType = DType>> CscData<R> {
/// Element-wise division: C = A ./ B
///
/// Computes the element-wise quotient of two sparse matrices with the same shape.
/// Only positions where BOTH matrices have non-zero values will be non-zero
/// in the result.
///
/// # Arguments
///
/// * `other` - Another CSC matrix with the same shape and dtype (divisor)
///
/// # Returns
///
/// A new CSC matrix containing the element-wise quotient
///
/// # Errors
///
/// Returns error if:
/// - Shapes don't match
/// - Dtypes don't match
///
/// # Algorithm
///
/// Column-by-column intersection of sorted row indices, dividing values
/// at matching positions.
///
/// # Performance
///
/// O(nnz_a + nnz_b) - linear merge since rows are sorted within columns.
/// Result has at most min(nnz_a, nnz_b) non-zeros.
///
/// # Example
///
/// ```
/// # use numr::prelude::*;
/// # #[cfg(feature = "sparse")]
/// # {
/// # use numr::sparse::SparseTensor;
/// # let device = CpuDevice::new();
/// // A: B: C = A ./ B:
/// // [8, 3] [4, 0] [2, 0]
/// // [0, 10] ./ [6, 2] = [0, 5]
/// # let a_sp = SparseTensor::<CpuRuntime>::from_coo_slices(&[0, 0, 1], &[0, 1, 1], &[8.0f32, 3.0, 10.0], [2, 2], &device)?.to_csc()?;
/// # let b_sp = SparseTensor::<CpuRuntime>::from_coo_slices(&[0, 1, 1], &[0, 0, 1], &[4.0f32, 6.0, 2.0], [2, 2], &device)?.to_csc()?;
/// # if let numr::sparse::SparseTensor::Csc(a) = a_sp { if let numr::sparse::SparseTensor::Csc(b) = b_sp {
/// let c = a.div(&b)?;
/// # } }
/// # }
/// # Ok::<(), numr::error::Error>(())
/// ```
pub fn div(&self, other: &Self) -> Result<Self> {
// 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();
let [_nrows, ncols] = self.shape;
// Handle empty cases - if either is empty, result is empty
if self.nnz() == 0 || other.nnz() == 0 {
return Ok(Self::empty(self.shape, dtype, device));
}
// Read CSC data from both matrices
let col_ptrs_a: Vec<i64> = self.col_ptrs.to_vec();
let row_indices_a: Vec<i64> = self.row_indices.to_vec();
let col_ptrs_b: Vec<i64> = other.col_ptrs.to_vec();
let row_indices_b: Vec<i64> = other.row_indices.to_vec();
// Dispatch on dtype to merge and divide values
crate::dispatch_dtype!(dtype, T => {
let vals_a: Vec<T> = self.values.to_vec();
let vals_b: Vec<T> = other.values.to_vec();
// Build result arrays column by column
let mut result_col_ptrs: Vec<i64> = Vec::with_capacity(ncols + 1);
let mut result_rows: Vec<i64> = Vec::new();
let mut result_vals: Vec<T> = Vec::new();
result_col_ptrs.push(0);
for col in 0..ncols {
let start_a = col_ptrs_a[col] as usize;
let end_a = col_ptrs_a[col + 1] as usize;
let start_b = col_ptrs_b[col] as usize;
let end_b = col_ptrs_b[col + 1] as usize;
// Intersect sorted rows for this column
let mut i = start_a;
let mut j = start_b;
while i < end_a && j < end_b {
let row_a = row_indices_a[i];
let row_b = row_indices_b[j];
if row_a < row_b {
// A only - skip (result is 0)
i += 1;
} else if row_b < row_a {
// B only - skip (result is 0)
j += 1;
} else {
// Same row - divide values
let quotient = vals_a[i].to_f64() / vals_b[j].to_f64();
result_rows.push(row_a);
result_vals.push(T::from_f64(quotient));
i += 1;
j += 1;
}
}
result_col_ptrs.push(result_rows.len() as i64);
}
let col_ptrs_tensor = Tensor::from_slice(&result_col_ptrs, &[result_col_ptrs.len()], device);
let row_indices_tensor = Tensor::from_slice(&result_rows, &[result_rows.len()], device);
let values_tensor = Tensor::from_slice(&result_vals, &[result_vals.len()], device);
return Self::new(col_ptrs_tensor, row_indices_tensor, values_tensor, self.shape);
}, "CSC element-wise div");
}
}
#[cfg(test)]
mod tests {
use super::*;
use crate::dtype::DType;
use crate::runtime::cpu::CpuRuntime;
#[test]
fn test_csc_div_overlapping() {
let device = <CpuRuntime as Runtime>::Device::default();
// A: B:
// [8, 0] [2, 5]
// [0, 35] [7, 7]
// CSC for A: col_ptrs=[0,1,2], row_indices=[0,1], values=[8,35]
// CSC for B: col_ptrs=[0,2,4], row_indices=[0,1,0,1], values=[2,7,5,7]
let a = CscData::<CpuRuntime>::from_slices(
&[0i64, 1, 2],
&[0i64, 1],
&[8.0f32, 35.0],
[2, 2],
&device,
)
.unwrap();
let b = CscData::<CpuRuntime>::from_slices(
&[0i64, 2, 4],
&[0i64, 1, 0, 1],
&[2.0f32, 7.0, 5.0, 7.0],
[2, 2],
&device,
)
.unwrap();
let c = a.div(&b).unwrap();
// Only positions where both have values: (0,0) and (1,1)
// 8/2=4, 35/7=5
assert_eq!(c.nnz(), 2);
let col_ptrs: Vec<i64> = c.col_ptrs().to_vec();
let row_indices: Vec<i64> = c.row_indices().to_vec();
let vals: Vec<f32> = c.values().to_vec();
assert_eq!(col_ptrs, vec![0, 1, 2]);
assert_eq!(row_indices, vec![0, 1]);
assert_eq!(vals, vec![4.0, 5.0]);
}
#[test]
fn test_csc_div_disjoint() {
let device = <CpuRuntime as Runtime>::Device::default();
// A: B:
// [1, 0] [0, 2]
// [0, 3] [4, 0]
// CSC for A: col_ptrs=[0,1,2], row_indices=[0,1], values=[1,3]
// CSC for B: col_ptrs=[0,1,2], row_indices=[1,0], values=[4,2]
let a = CscData::<CpuRuntime>::from_slices(
&[0i64, 1, 2],
&[0i64, 1],
&[1.0f32, 3.0],
[2, 2],
&device,
)
.unwrap();
let b = CscData::<CpuRuntime>::from_slices(
&[0i64, 1, 2],
&[1i64, 0],
&[4.0f32, 2.0],
[2, 2],
&device,
)
.unwrap();
let c = a.div(&b).unwrap();
// Result is empty since no positions overlap
assert_eq!(c.nnz(), 0);
}
#[test]
fn test_csc_div_empty() {
let device = <CpuRuntime as Runtime>::Device::default();
let a = CscData::<CpuRuntime>::from_slices(
&[0i64, 1, 2],
&[0i64, 1],
&[1.0f32, 2.0],
[2, 2],
&device,
)
.unwrap();
let b = CscData::<CpuRuntime>::empty([2, 2], DType::F32, &device);
// Divide with empty matrix gives empty result
let c = a.div(&b).unwrap();
assert_eq!(c.nnz(), 0);
let c2 = b.div(&a).unwrap();
assert_eq!(c2.nnz(), 0);
}
#[test]
fn test_csc_div_shape_mismatch() {
let device = <CpuRuntime as Runtime>::Device::default();
let a = CscData::<CpuRuntime>::empty([2, 3], DType::F32, &device);
let b = CscData::<CpuRuntime>::empty([3, 2], DType::F32, &device);
let result = a.div(&b);
assert!(result.is_err());
}
#[test]
fn test_csc_div_same_positions() {
let device = <CpuRuntime as Runtime>::Device::default();
// Both have values at exactly the same positions
// Matrix layout:
// [10, 0] [2, 0]
// [18, 28] [6, 7]
// CSC for A: col_ptrs=[0,2,3], row_indices=[0,1,1], values=[10,18,28]
// CSC for B: col_ptrs=[0,2,3], row_indices=[0,1,1], values=[2,6,7]
let a = CscData::<CpuRuntime>::from_slices(
&[0i64, 2, 3],
&[0i64, 1, 1],
&[10.0f32, 18.0, 28.0],
[2, 2],
&device,
)
.unwrap();
let b = CscData::<CpuRuntime>::from_slices(
&[0i64, 2, 3],
&[0i64, 1, 1],
&[2.0f32, 6.0, 7.0],
[2, 2],
&device,
)
.unwrap();
let c = a.div(&b).unwrap();
assert_eq!(c.nnz(), 3);
let vals: Vec<f32> = c.values().to_vec();
assert_eq!(vals, vec![5.0, 3.0, 4.0]); // 10/2, 18/6, 28/7
}
}