kryst 3.2.1

Krylov subspace and preconditioned iterative solvers for dense and sparse linear systems, with shared and distributed memory parallelism.
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

#[allow(unused_imports)]
use crate::algebra::blas::{dot_conj, nrm2};
#[allow(unused_imports)]
use crate::algebra::prelude::*;
use crate::matrix::sparse::CsrMatrix;

#[derive(Clone, Debug)]
pub struct Strength {
    pub row_ptr: Vec<usize>,
    pub col_idx: Vec<usize>,
}

impl Strength {
    pub fn from_csr(a: &CsrMatrix<f64>, theta: f64, normalize: bool) -> Self {
        strength_csr(a, theta, normalize)
    }

    /// Optional per-row cap: keep the first `k` columns in ascending column order.
    pub fn capped(&self, k: usize) -> Strength {
        if k == 0 {
            return self.clone();
        }
        let n = self.row_ptr.len() - 1;
        let mut row_ptr = Vec::with_capacity(n + 1);
        row_ptr.push(0);
        let mut col_idx = Vec::with_capacity(self.col_idx.len());
        for i in 0..n {
            let rs = self.row_ptr[i];
            let re = self.row_ptr[i + 1];
            let take = (re - rs).min(k);
            col_idx.extend_from_slice(&self.col_idx[rs..rs + take]);
            row_ptr.push(col_idx.len());
        }
        Strength { row_ptr, col_idx }
    }

    /// Symmetrize the strength graph treating it as undirected.
    pub fn symmetrize(&self) -> Strength {
        let n = self.row_ptr.len() - 1;
        let mut deg = vec![0usize; n];
        for i in 0..n {
            for &j in &self.col_idx[self.row_ptr[i]..self.row_ptr[i + 1]] {
                deg[i] += 1;
                deg[j] += 1;
            }
        }
        let mut row_ptr = Vec::with_capacity(n + 1);
        row_ptr.push(0);
        for i in 0..n {
            row_ptr.push(row_ptr[i] + deg[i]);
        }
        let mut col_idx = vec![usize::MAX; row_ptr[n]];
        let mut next = vec![0usize; n];
        for i in 0..n {
            next[i] = row_ptr[i];
        }
        for i in 0..n {
            for &j in &self.col_idx[self.row_ptr[i]..self.row_ptr[i + 1]] {
                let pi = next[i];
                col_idx[pi] = j;
                next[i] += 1;
                let pj = next[j];
                col_idx[pj] = i;
                next[j] += 1;
            }
        }
        for i in 0..n {
            let rs = row_ptr[i];
            let re = row_ptr[i + 1];
            col_idx[rs..re].sort_unstable();
            let mut write = rs;
            let mut last = None;
            for idx in rs..re {
                let v = col_idx[idx];
                if Some(v) != last {
                    col_idx[write] = v;
                    write += 1;
                    last = Some(v);
                }
            }
            let len = write - rs;
            col_idx.drain(write..re);
            let delta = (re - rs) - len;
            if delta > 0 {
                for r in (i + 1)..=n {
                    row_ptr[r] -= delta;
                }
            }
        }
        Strength { row_ptr, col_idx }
    }
}

pub fn strength_csr(a: &CsrMatrix<f64>, theta: f64, normalize: bool) -> Strength {
    let n = a.nrows();
    let rp = a.row_ptr();
    let cj = a.col_idx();
    let vv = a.values();

    let mut row_ptr = Vec::with_capacity(n + 1);
    let mut col_idx: Vec<usize> = Vec::with_capacity(a.nnz());
    row_ptr.push(0);

    if normalize {
        // |a_ij| / sqrt(|a_ii||a_jj|)
        let mut diag = vec![0.0f64; n];
        for i in 0..n {
            let rs = rp[i];
            let re = rp[i + 1];
            let mut aii = 0.0;
            for p in rs..re {
                if cj[p] == i {
                    aii = vv[p].abs();
                    break;
                }
            }
            diag[i] = aii;
        }
        for i in 0..n {
            let rs = rp[i];
            let re = rp[i + 1];
            let mut count = 0usize;
            for p in rs..re {
                let j = cj[p];
                if j == i {
                    continue;
                }
                let denom = (diag[i] * diag[j]).sqrt();
                if denom > 0.0 {
                    let s = vv[p].abs() / denom;
                    if s >= theta {
                        col_idx.push(j);
                        count += 1;
                    }
                }
            }
            row_ptr.push(row_ptr.last().unwrap() + count);
        }
    } else {
        // |a_ij| >= theta * max_off_i
        for i in 0..n {
            let rs = rp[i];
            let re = rp[i + 1];
            let mut max_off = 0.0f64;
            for p in rs..re {
                let j = cj[p];
                if j != i {
                    max_off = max_off.max(vv[p].abs());
                }
            }
            let thr = theta * max_off;
            let mut count = 0usize;
            for p in rs..re {
                let j = cj[p];
                if j != i && vv[p].abs() >= thr {
                    col_idx.push(j);
                    count += 1;
                }
            }
            row_ptr.push(row_ptr.last().unwrap() + count);
        }
    }

    Strength { row_ptr, col_idx }
}