numr 0.5.2

High-performance numerical computing with multi-backend GPU acceleration (CPU/CUDA/WebGPU)
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

//! Sparse format conversion implementations (COO↔CSR↔CSC) for CPU

use super::CpuRuntime;
use crate::dtype::Element;
use crate::error::Result;
use crate::tensor::Tensor;

pub fn coo_to_csr<T: Element>(
    row_indices: &Tensor<CpuRuntime>,
    col_indices: &Tensor<CpuRuntime>,
    values: &Tensor<CpuRuntime>,
    shape: [usize; 2],
) -> Result<(Tensor<CpuRuntime>, Tensor<CpuRuntime>, Tensor<CpuRuntime>)> {
    let [nrows, _ncols] = shape;
    let nnz = values.numel();
    let device = values.device();

    let row_idx: Vec<i64> = row_indices.to_vec();
    let col_idx: Vec<i64> = col_indices.to_vec();
    let vals: Vec<T> = values.to_vec();

    let mut perm: Vec<usize> = (0..nnz).collect();
    perm.sort_by_key(|&i| (row_idx[i], col_idx[i]));

    let mut sorted_col_indices = Vec::with_capacity(nnz);
    let mut sorted_values = Vec::with_capacity(nnz);
    for &i in &perm {
        sorted_col_indices.push(col_idx[i]);
        sorted_values.push(vals[i]);
    }

    let mut row_ptrs = vec![0i64; nrows + 1];
    for &i in &perm {
        let row = row_idx[i] as usize;
        row_ptrs[row + 1] += 1;
    }
    for i in 1..=nrows {
        row_ptrs[i] += row_ptrs[i - 1];
    }

    let row_ptrs_tensor = Tensor::from_slice(&row_ptrs, &[nrows + 1], device);
    let col_indices_tensor = Tensor::from_slice(&sorted_col_indices, &[nnz], device);
    let values_tensor = Tensor::from_slice(&sorted_values, &[nnz], device);

    Ok((row_ptrs_tensor, col_indices_tensor, values_tensor))
}

pub fn coo_to_csc<T: Element>(
    row_indices: &Tensor<CpuRuntime>,
    col_indices: &Tensor<CpuRuntime>,
    values: &Tensor<CpuRuntime>,
    shape: [usize; 2],
) -> Result<(Tensor<CpuRuntime>, Tensor<CpuRuntime>, Tensor<CpuRuntime>)> {
    let [_nrows, ncols] = shape;
    let nnz = values.numel();
    let device = values.device();

    let row_idx: Vec<i64> = row_indices.to_vec();
    let col_idx: Vec<i64> = col_indices.to_vec();
    let vals: Vec<T> = values.to_vec();

    let mut perm: Vec<usize> = (0..nnz).collect();
    perm.sort_by_key(|&i| (col_idx[i], row_idx[i]));

    let mut sorted_row_indices = Vec::with_capacity(nnz);
    let mut sorted_values = Vec::with_capacity(nnz);
    for &i in &perm {
        sorted_row_indices.push(row_idx[i]);
        sorted_values.push(vals[i]);
    }

    let mut col_ptrs = vec![0i64; ncols + 1];
    for &i in &perm {
        let col = col_idx[i] as usize;
        col_ptrs[col + 1] += 1;
    }
    for i in 1..=ncols {
        col_ptrs[i] += col_ptrs[i - 1];
    }

    let col_ptrs_tensor = Tensor::from_slice(&col_ptrs, &[ncols + 1], device);
    let row_indices_tensor = Tensor::from_slice(&sorted_row_indices, &[nnz], device);
    let values_tensor = Tensor::from_slice(&sorted_values, &[nnz], device);

    Ok((col_ptrs_tensor, row_indices_tensor, values_tensor))
}

pub fn csr_to_coo<T: Element>(
    row_ptrs: &Tensor<CpuRuntime>,
    col_indices: &Tensor<CpuRuntime>,
    values: &Tensor<CpuRuntime>,
    shape: [usize; 2],
) -> Result<(Tensor<CpuRuntime>, Tensor<CpuRuntime>, Tensor<CpuRuntime>)> {
    let [nrows, _ncols] = shape;
    let nnz = values.numel();
    let device = values.device();

    let ptrs: Vec<i64> = row_ptrs.to_vec();

    let mut row_indices = Vec::with_capacity(nnz);
    for row in 0..nrows {
        let start = ptrs[row] as usize;
        let end = ptrs[row + 1] as usize;
        for _ in start..end {
            row_indices.push(row as i64);
        }
    }

    let row_indices_tensor = Tensor::from_slice(&row_indices, &[nnz], device);
    Ok((row_indices_tensor, col_indices.clone(), values.clone()))
}

pub fn csc_to_coo<T: Element>(
    col_ptrs: &Tensor<CpuRuntime>,
    row_indices: &Tensor<CpuRuntime>,
    values: &Tensor<CpuRuntime>,
    shape: [usize; 2],
) -> Result<(Tensor<CpuRuntime>, Tensor<CpuRuntime>, Tensor<CpuRuntime>)> {
    let [_nrows, ncols] = shape;
    let nnz = values.numel();
    let device = values.device();

    let ptrs: Vec<i64> = col_ptrs.to_vec();

    let mut col_indices_vec = Vec::with_capacity(nnz);
    for col in 0..ncols {
        let start = ptrs[col] as usize;
        let end = ptrs[col + 1] as usize;
        for _ in start..end {
            col_indices_vec.push(col as i64);
        }
    }

    let col_indices_tensor = Tensor::from_slice(&col_indices_vec, &[nnz], device);
    Ok((row_indices.clone(), col_indices_tensor, values.clone()))
}

pub fn csr_to_csc<T: Element>(
    row_ptrs: &Tensor<CpuRuntime>,
    col_indices: &Tensor<CpuRuntime>,
    values: &Tensor<CpuRuntime>,
    shape: [usize; 2],
) -> Result<(Tensor<CpuRuntime>, Tensor<CpuRuntime>, Tensor<CpuRuntime>)> {
    let [nrows, ncols] = shape;
    let nnz = values.numel();
    let device = values.device();

    let row_ptr: Vec<i64> = row_ptrs.to_vec();
    let col_idx: Vec<i64> = col_indices.to_vec();
    let vals: Vec<T> = values.to_vec();

    let mut col_counts = vec![0usize; ncols];
    for &col in &col_idx {
        col_counts[col as usize] += 1;
    }

    let mut col_ptrs = vec![0i64; ncols + 1];
    for col in 0..ncols {
        col_ptrs[col + 1] = col_ptrs[col] + col_counts[col] as i64;
    }

    let mut new_row_indices = vec![0i64; nnz];
    let mut new_values = vec![T::from_f64(0.0); nnz];
    let mut col_positions = col_ptrs[..ncols].to_vec();

    for row in 0..nrows {
        let start = row_ptr[row] as usize;
        let end = row_ptr[row + 1] as usize;
        for idx in start..end {
            let col = col_idx[idx] as usize;
            let pos = col_positions[col] as usize;
            new_row_indices[pos] = row as i64;
            new_values[pos] = vals[idx];
            col_positions[col] += 1;
        }
    }

    let col_ptrs_tensor = Tensor::from_slice(&col_ptrs, &[ncols + 1], device);
    let row_indices_tensor = Tensor::from_slice(&new_row_indices, &[nnz], device);
    let values_tensor = Tensor::from_slice(&new_values, &[nnz], device);

    Ok((col_ptrs_tensor, row_indices_tensor, values_tensor))
}

pub fn csc_to_csr<T: Element>(
    col_ptrs: &Tensor<CpuRuntime>,
    row_indices: &Tensor<CpuRuntime>,
    values: &Tensor<CpuRuntime>,
    shape: [usize; 2],
) -> Result<(Tensor<CpuRuntime>, Tensor<CpuRuntime>, Tensor<CpuRuntime>)> {
    let [nrows, ncols] = shape;
    let nnz = values.numel();
    let device = values.device();

    let col_ptr: Vec<i64> = col_ptrs.to_vec();
    let row_idx: Vec<i64> = row_indices.to_vec();
    let vals: Vec<T> = values.to_vec();

    let mut row_counts = vec![0usize; nrows];
    for &row in &row_idx {
        row_counts[row as usize] += 1;
    }

    let mut row_ptrs = vec![0i64; nrows + 1];
    for row in 0..nrows {
        row_ptrs[row + 1] = row_ptrs[row] + row_counts[row] as i64;
    }

    let mut new_col_indices = vec![0i64; nnz];
    let mut new_values = vec![T::from_f64(0.0); nnz];
    let mut row_positions = row_ptrs[..nrows].to_vec();

    for col in 0..ncols {
        let start = col_ptr[col] as usize;
        let end = col_ptr[col + 1] as usize;
        for idx in start..end {
            let row = row_idx[idx] as usize;
            let pos = row_positions[row] as usize;
            new_col_indices[pos] = col as i64;
            new_values[pos] = vals[idx];
            row_positions[row] += 1;
        }
    }

    let row_ptrs_tensor = Tensor::from_slice(&row_ptrs, &[nrows + 1], device);
    let col_indices_tensor = Tensor::from_slice(&new_col_indices, &[nnz], device);
    let values_tensor = Tensor::from_slice(&new_values, &[nnz], device);

    Ok((row_ptrs_tensor, col_indices_tensor, values_tensor))
}

pub fn extract_diagonal_csr<T: Element>(
    row_ptrs: &Tensor<CpuRuntime>,
    col_indices: &Tensor<CpuRuntime>,
    values: &Tensor<CpuRuntime>,
    shape: [usize; 2],
) -> Result<Tensor<CpuRuntime>> {
    let [nrows, ncols] = shape;
    let n = nrows.min(ncols);
    let device = values.device();

    if n == 0 {
        return Ok(Tensor::empty(&[0], values.dtype(), device));
    }

    let row_ptrs_data: Vec<i64> = row_ptrs.to_vec();
    let col_indices_data: Vec<i64> = col_indices.to_vec();
    let values_data: Vec<T> = values.to_vec();

    let mut diag = vec![T::zero(); n];
    for row in 0..n {
        let start = row_ptrs_data[row] as usize;
        let end = row_ptrs_data[row + 1] as usize;
        for pos in start..end {
            if col_indices_data[pos] as usize == row {
                diag[row] = values_data[pos];
                break;
            }
        }
    }

    Ok(Tensor::from_slice(&diag, &[n], device))
}