lax/
tridiagonal.rs

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
//! Implement linear solver using LU decomposition
//! for tridiagonal matrix

use crate::{error::*, layout::*, *};
use cauchy::*;
use num_traits::Zero;
use std::ops::{Index, IndexMut};

/// Represents a tridiagonal matrix as 3 one-dimensional vectors.
///
/// ```text
/// [d0, u1,  0,   ...,       0,
///  l1, d1, u2,            ...,
///   0, l2, d2,
///  ...           ...,  u{n-1},
///   0,  ...,  l{n-1},  d{n-1},]
/// ```
#[derive(Clone, PartialEq, Eq)]
pub struct Tridiagonal<A: Scalar> {
    /// layout of raw matrix
    pub l: MatrixLayout,
    /// (n-1) sub-diagonal elements of matrix.
    pub dl: Vec<A>,
    /// (n) diagonal elements of matrix.
    pub d: Vec<A>,
    /// (n-1) super-diagonal elements of matrix.
    pub du: Vec<A>,
}

impl<A: Scalar> Tridiagonal<A> {
    fn opnorm_one(&self) -> A::Real {
        let mut col_sum: Vec<A::Real> = self.d.iter().map(|val| val.abs()).collect();
        for i in 0..col_sum.len() {
            if i < self.dl.len() {
                col_sum[i] += self.dl[i].abs();
            }
            if i > 0 {
                col_sum[i] += self.du[i - 1].abs();
            }
        }
        let mut max = A::Real::zero();
        for &val in &col_sum {
            if max < val {
                max = val;
            }
        }
        max
    }
}

/// Represents the LU factorization of a tridiagonal matrix `A` as `A = P*L*U`.
#[derive(Clone, PartialEq)]
pub struct LUFactorizedTridiagonal<A: Scalar> {
    /// A tridiagonal matrix which consists of
    /// - l : layout of raw matrix
    /// - dl: (n-1) multipliers that define the matrix L.
    /// - d : (n) diagonal elements of the upper triangular matrix U.
    /// - du: (n-1) elements of the first super-diagonal of U.
    pub a: Tridiagonal<A>,
    /// (n-2) elements of the second super-diagonal of U.
    pub du2: Vec<A>,
    /// The pivot indices that define the permutation matrix `P`.
    pub ipiv: Pivot,

    a_opnorm_one: A::Real,
}

impl<A: Scalar> Index<(i32, i32)> for Tridiagonal<A> {
    type Output = A;
    #[inline]
    fn index(&self, (row, col): (i32, i32)) -> &A {
        let (n, _) = self.l.size();
        assert!(
            std::cmp::max(row, col) < n,
            "ndarray: index {:?} is out of bounds for array of shape {}",
            [row, col],
            n
        );
        match row - col {
            0 => &self.d[row as usize],
            1 => &self.dl[col as usize],
            -1 => &self.du[row as usize],
            _ => panic!(
                "ndarray-linalg::tridiagonal: index {:?} is not tridiagonal element",
                [row, col]
            ),
        }
    }
}

impl<A: Scalar> Index<[i32; 2]> for Tridiagonal<A> {
    type Output = A;
    #[inline]
    fn index(&self, [row, col]: [i32; 2]) -> &A {
        &self[(row, col)]
    }
}

impl<A: Scalar> IndexMut<(i32, i32)> for Tridiagonal<A> {
    #[inline]
    fn index_mut(&mut self, (row, col): (i32, i32)) -> &mut A {
        let (n, _) = self.l.size();
        assert!(
            std::cmp::max(row, col) < n,
            "ndarray: index {:?} is out of bounds for array of shape {}",
            [row, col],
            n
        );
        match row - col {
            0 => &mut self.d[row as usize],
            1 => &mut self.dl[col as usize],
            -1 => &mut self.du[row as usize],
            _ => panic!(
                "ndarray-linalg::tridiagonal: index {:?} is not tridiagonal element",
                [row, col]
            ),
        }
    }
}

impl<A: Scalar> IndexMut<[i32; 2]> for Tridiagonal<A> {
    #[inline]
    fn index_mut(&mut self, [row, col]: [i32; 2]) -> &mut A {
        &mut self[(row, col)]
    }
}

/// Wraps `*gttrf`, `*gtcon` and `*gttrs`
pub trait Tridiagonal_: Scalar + Sized {
    /// Computes the LU factorization of a tridiagonal `m x n` matrix `a` using
    /// partial pivoting with row interchanges.
    fn lu_tridiagonal(a: Tridiagonal<Self>) -> Result<LUFactorizedTridiagonal<Self>>;

    fn rcond_tridiagonal(lu: &LUFactorizedTridiagonal<Self>) -> Result<Self::Real>;

    fn solve_tridiagonal(
        lu: &LUFactorizedTridiagonal<Self>,
        bl: MatrixLayout,
        t: Transpose,
        b: &mut [Self],
    ) -> Result<()>;
}

macro_rules! impl_tridiagonal {
    (@real, $scalar:ty, $gttrf:path, $gtcon:path, $gttrs:path) => {
        impl_tridiagonal!(@body, $scalar, $gttrf, $gtcon, $gttrs, iwork);
    };
    (@complex, $scalar:ty, $gttrf:path, $gtcon:path, $gttrs:path) => {
        impl_tridiagonal!(@body, $scalar, $gttrf, $gtcon, $gttrs, );
    };
    (@body, $scalar:ty, $gttrf:path, $gtcon:path, $gttrs:path, $($iwork:ident)*) => {
        impl Tridiagonal_ for $scalar {
            fn lu_tridiagonal(mut a: Tridiagonal<Self>) -> Result<LUFactorizedTridiagonal<Self>> {
                let (n, _) = a.l.size();
                let mut du2 = unsafe { vec_uninit( (n - 2) as usize) };
                let mut ipiv = unsafe { vec_uninit( n as usize) };
                // We have to calc one-norm before LU factorization
                let a_opnorm_one = a.opnorm_one();
                let mut info = 0;
                unsafe { $gttrf(n, &mut a.dl, &mut a.d, &mut a.du, &mut du2, &mut ipiv, &mut info,) };
                info.as_lapack_result()?;
                Ok(LUFactorizedTridiagonal {
                    a,
                    du2,
                    ipiv,
                    a_opnorm_one,
                })
            }

            fn rcond_tridiagonal(lu: &LUFactorizedTridiagonal<Self>) -> Result<Self::Real> {
                let (n, _) = lu.a.l.size();
                let ipiv = &lu.ipiv;
                let mut work = unsafe { vec_uninit( 2 * n as usize) };
                $(
                let mut $iwork = unsafe { vec_uninit( n as usize) };
                )*
                let mut rcond = Self::Real::zero();
                let mut info = 0;
                unsafe {
                    $gtcon(
                        NormType::One as u8,
                        n,
                        &lu.a.dl,
                        &lu.a.d,
                        &lu.a.du,
                        &lu.du2,
                        ipiv,
                        lu.a_opnorm_one,
                        &mut rcond,
                        &mut work,
                        $(&mut $iwork,)*
                        &mut info,
                    );
                }
                info.as_lapack_result()?;
                Ok(rcond)
            }

            fn solve_tridiagonal(
                lu: &LUFactorizedTridiagonal<Self>,
                b_layout: MatrixLayout,
                t: Transpose,
                b: &mut [Self],
            ) -> Result<()> {
                let (n, _) = lu.a.l.size();
                let ipiv = &lu.ipiv;
                // Transpose if b is C-continuous
                let mut b_t = None;
                let b_layout = match b_layout {
                    MatrixLayout::C { .. } => {
                        b_t = Some(unsafe { vec_uninit( b.len()) });
                        transpose(b_layout, b, b_t.as_mut().unwrap())
                    }
                    MatrixLayout::F { .. } => b_layout,
                };
                let (ldb, nrhs) = b_layout.size();
                let mut info = 0;
                unsafe {
                    $gttrs(
                        t as u8,
                        n,
                        nrhs,
                        &lu.a.dl,
                        &lu.a.d,
                        &lu.a.du,
                        &lu.du2,
                        ipiv,
                        b_t.as_mut().map(|v| v.as_mut_slice()).unwrap_or(b),
                        ldb,
                        &mut info,
                    );
                }
                info.as_lapack_result()?;
                if let Some(b_t) = b_t {
                    transpose(b_layout, &b_t, b);
                }
                Ok(())
            }
        }
    };
} // impl_tridiagonal!

impl_tridiagonal!(@real, f64, lapack::dgttrf, lapack::dgtcon, lapack::dgttrs);
impl_tridiagonal!(@real, f32, lapack::sgttrf, lapack::sgtcon, lapack::sgttrs);
impl_tridiagonal!(@complex, c64, lapack::zgttrf, lapack::zgtcon, lapack::zgttrs);
impl_tridiagonal!(@complex, c32, lapack::cgttrf, lapack::cgtcon, lapack::cgttrs);