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

use crate::algebra::parallel;
use crate::algebra::scalar::KrystScalar;
use crate::error::KError;
use std::sync::atomic::{AtomicPtr, Ordering};

use super::{IluCsr, Real};

pub fn tri_solve_serial(pc: &IluCsr, x: &[Real], y: &mut [Real]) -> Result<(), KError> {
    let n = pc.n();
    let lr = pc.l_row();
    let lc = pc.l_col();
    let lv = pc.l_val();
    let ur = pc.u_row();
    let uc = pc.u_col();
    let uv = pc.u_val();
    let di = pc.u_diag_ix();

    if y.len() != n || x.len() != n {
        return Err(KError::InvalidInput("tri_solve: dimension mismatch".into()));
    }

    // forward: L y = x  (unit diagonal)
    // Reuse y as the solution vector
    for i in 0..n {
        let mut s = x[i];
        for p in lr[i]..lr[i + 1] {
            let j = lc[p]; // j < i
            s -= lv[p] * y[j];
        }
        y[i] = s;
    }

    // backward: U z = y -> y <- z (output)
    for i in (0..n).rev() {
        let mut s = y[i];
        for p in ur[i]..ur[i + 1] {
            let j = uc[p];
            if j > i {
                s -= uv[p] * y[j];
            }
        }
        let d = uv[di[i]];
        y[i] = s / d;
    }
    Ok(())
}

pub fn tri_solve_level_scheduled(pc: &IluCsr, x: &[Real], y: &mut [Real]) -> Result<(), KError> {
    // Fallback to serial if no levels computed
    if pc.buckets_fwd().is_empty() || pc.buckets_bwd().is_empty() {
        return tri_solve_serial(pc, x, y);
    }

    let n = pc.n();
    if x.len() != n || y.len() != n {
        return Err(KError::InvalidInput("tri_solve: dimension mismatch".into()));
    }

    let lr = pc.l_row();
    let lc = pc.l_col();
    let lv = pc.l_val();
    let ur = pc.u_row();
    let uc = pc.u_col();
    let uv = pc.u_val();
    let di = pc.u_diag_ix();

    // Forward: L y = x (unit diagonal). Per-level parallel, disjoint writes.
    y.fill(Real::zero());
    for bucket in pc.buckets_fwd() {
        let y_ptr = AtomicPtr::new(y.as_mut_ptr());
        parallel::par_for_each_index(bucket.len(), move |k| unsafe {
            let i = *bucket.get_unchecked(k);
            let mut s = *x.get_unchecked(i);
            let rs = *lr.get_unchecked(i);
            let re = *lr.get_unchecked(i + 1);
            let lc_p = lc.as_ptr();
            let lv_p = lv.as_ptr();
            let y_ptr = y_ptr.load(Ordering::Relaxed);
            for p in rs..re {
                let j = *lc_p.add(p);
                s -= *lv_p.add(p) * *y_ptr.add(j);
            }
            *y_ptr.add(i) = s;
        });
    }

    // Backward: U z = y → write z into y. Per-level parallel, disjoint writes.
    for bucket in pc.buckets_bwd() {
        let y_ptr = AtomicPtr::new(y.as_mut_ptr());
        parallel::par_for_each_index(bucket.len(), move |k| unsafe {
            let i = *bucket.get_unchecked(k);
            let y_ptr = y_ptr.load(Ordering::Relaxed);
            let mut s = *y_ptr.add(i);
            let rs = *ur.get_unchecked(i);
            let re = *ur.get_unchecked(i + 1);
            let uc_p = uc.as_ptr();
            let uv_p = uv.as_ptr();
            for p in rs..re {
                let j = *uc_p.add(p);
                if j > i {
                    s -= *uv_p.add(p) * *y_ptr.add(j);
                }
            }
            let d = *uv_p.add(*di.get_unchecked(i));
            *y_ptr.add(i) = s / d;
        });
    }

    Ok(())
}

pub fn tri_solve_transpose_serial(
    pc: &IluCsr,
    ut_row: &[usize],
    ut_col: &[usize],
    ut_val: &[Real],
    lt_row: &[usize],
    lt_col: &[usize],
    lt_val: &[Real],
    x: &[Real],
    y: &mut [Real],
) -> Result<(), KError> {
    let n = pc.n();
    if x.len() != n || y.len() != n {
        return Err(KError::InvalidInput("tri_solve: dimension mismatch".into()));
    }

    // Forward solve with U^T (lower with diag)
    for i in 0..n {
        let mut s = x[i];
        for p in ut_row[i]..ut_row[i + 1] {
            let j = ut_col[p];
            if j < i {
                s -= ut_val[p] * y[j];
            }
        }
        let d = pc.u_val()[pc.u_diag_ix()[i]];
        y[i] = s / d;
    }

    // Backward solve with L^T (upper unit)
    for i in (0..n).rev() {
        let mut s = y[i];
        for p in lt_row[i]..lt_row[i + 1] {
            let j = lt_col[p];
            if j > i {
                s -= lt_val[p] * y[j];
            }
        }
        y[i] = s;
    }

    Ok(())
}