mdarray-linalg 0.2.0

Linear algebra operations for mdarray, with multiple exchangeable backends
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

use criterion::{Criterion, criterion_group, criterion_main};
use std::hint::black_box as bb;

use mdarray::{Const, DArray, Dyn, Slice};

use mdarray_linalg::Naive;
use mdarray_linalg::qr::QR;
use mdarray_linalg_faer::Faer;
use mdarray_linalg_lapack::Lapack;

use nalgebra::{DMatrix, Matrix4};

const N: i32 = 512;

// type Slice4x4Const = Slice<f64, (Const<4>, Const<4>)>;
type Slice4x4Dyn = Slice<f64, (Dyn, Dyn)>;
type SliceNDyn = Slice<f64, (Dyn, Dyn)>;

// ============================================================================
// 4x4 benchmarks nalgebra with static size
// ============================================================================

#[inline(never)]
pub fn qr_4x4_nalgebra_static(
    data: &[f64; 16],
) -> nalgebra::linalg::QR<f64, nalgebra::Const<4>, nalgebra::Const<4>> {
    let a = Matrix4::from_row_slice(data);
    a.qr()
}

// ============================================================================
// 4x4 benchmarks with allocation
// ============================================================================

#[inline(never)]
pub fn qr_4x4_dyn_backend_naive(a: &Slice4x4Dyn) -> (DArray<f64, 2>, DArray<f64, 2>) {
    let mut a_copy = a.to_owned();
    let bd = Naive;
    bd.qr(&mut a_copy)
}

#[inline(never)]
pub fn qr_4x4_dyn_backend_lapack(a: &Slice4x4Dyn) -> (DArray<f64, 2>, DArray<f64, 2>) {
    let mut a_copy = a.to_owned();
    let bd = Lapack::new();
    bd.qr(&mut a_copy)
}

#[inline(never)]
pub fn qr_4x4_dyn_backend_faer(a: &Slice4x4Dyn) -> (DArray<f64, 2>, DArray<f64, 2>) {
    let mut a_copy = a.to_owned();
    let bd = Faer;
    bd.qr(&mut a_copy)
}

#[inline(never)]
pub fn qr_4x4_nalgebra(data: &[f64]) -> nalgebra::linalg::QR<f64, nalgebra::Dyn, nalgebra::Dyn> {
    let a = DMatrix::from_row_slice(4, 4, data);
    a.qr()
}

// ============================================================================
// NxN benchmarks
// ============================================================================

#[inline(never)]
pub fn qr_n_dyn_backend_naive(a: &SliceNDyn) -> (DArray<f64, 2>, DArray<f64, 2>) {
    let mut a_copy = a.to_owned();
    let bd = Naive;
    bd.qr(&mut a_copy)
}

#[inline(never)]
pub fn qr_n_dyn_backend_lapack(a: &SliceNDyn) -> (DArray<f64, 2>, DArray<f64, 2>) {
    let mut a_copy = a.to_owned();
    let bd = Lapack::new();
    bd.qr(&mut a_copy)
}

#[inline(never)]
pub fn qr_n_dyn_backend_faer(a: &SliceNDyn) -> (DArray<f64, 2>, DArray<f64, 2>) {
    let mut a_copy = a.to_owned();
    let bd = Faer;
    bd.qr(&mut a_copy)
}

#[inline(never)]
pub fn qr_n_nalgebra(data: &[f64]) -> nalgebra::linalg::QR<f64, nalgebra::Dyn, nalgebra::Dyn> {
    let a = DMatrix::from_row_slice(N as usize, N as usize, data);
    a.qr()
}

// ============================================================================
// Const dimension benchmarks
// ============================================================================

// #[inline(never)]
// pub fn qr_4x4_const_backend_naive(a: &Slice4x4Const) -> (DArray<f64, 2>, DArray<f64, 2>) {
//     let mut a_copy = a.to_owned();
//     let bd = Naive;
//     bd.qr(&mut a_copy)
// }

// #[inline(never)]
// pub fn qr_4x4_const_backend_lapack(a: &Slice4x4Const) -> (DArray<f64, 2>, DArray<f64, 2>) {
//     let mut a_copy = a.to_owned();
//     let bd = Lapack::new();
//     bd.qr(&mut a_copy)
// }

// #[inline(never)]
// pub fn qr_4x4_const_backend_faer(a: &Slice4x4Const) -> (DArray<f64, 2>, DArray<f64, 2>) {
//     let mut a_copy = a.to_owned();
//     let bd = Faer;
//     bd.qr(&mut a_copy)
// }

fn criterion_benchmark(crit: &mut Criterion) {
    // ========================================================================
    // 4x4 benchmarks
    // ========================================================================

    // 4x4 nalgebra static
    let a_4x4_data_array: [f64; 16] = {
        let mut arr = [0.0; 16];
        for (i, val) in (0..16).map(|x| (x as f64) * 0.5).enumerate() {
            arr[i] = val;
        }
        arr
    };

    crit.bench_function("qr_4x4_nalgebra_static", |bencher| {
        bencher.iter(|| qr_4x4_nalgebra_static(bb(&a_4x4_data_array)))
    });

    // 4x4 dynamic
    let a_4x4_dyn: DArray<_, 1> = (0..16).map(|x| (x as f64) * 0.5).collect::<Vec<_>>().into();
    let a_4x4_dyn = a_4x4_dyn.reshape((4, 4));
    let a_4x4_data: Vec<f64> = (0..16).map(|x| (x as f64) * 0.5).collect();

    // With allocation
    crit.bench_function("qr_4x4_dyn_naive", |bencher| {
        bencher.iter(|| qr_4x4_dyn_backend_naive(bb(&a_4x4_dyn)))
    });

    crit.bench_function("qr_4x4_dyn_lapack", |bencher| {
        bencher.iter(|| qr_4x4_dyn_backend_lapack(bb(&a_4x4_dyn)))
    });

    crit.bench_function("qr_4x4_dyn_faer", |bencher| {
        bencher.iter(|| qr_4x4_dyn_backend_faer(bb(&a_4x4_dyn)))
    });

    crit.bench_function("qr_4x4_nalgebra", |bencher| {
        bencher.iter(|| qr_4x4_nalgebra(bb(&a_4x4_data)))
    });

    // ========================================================================
    // 4x4 Const dimension benchmarks
    // ========================================================================

    // let a_4x4_const: DArray<_, 1> = (0..16).map(|x| (x as f64) * 0.5).collect::<Vec<_>>().into();
    // let a_4x4_const = a_4x4_const.reshape((Const::<4>, Const::<4>));

    // crit.bench_function("qr_4x4_const_naive", |bencher| {
    //     bencher.iter(|| qr_4x4_const_backend_naive(bb(&a_4x4_const)))
    // });

    // crit.bench_function("qr_4x4_const_lapack", |bencher| {
    //     bencher.iter(|| qr_4x4_const_backend_lapack(bb(&a_4x4_const)))
    // });

    // crit.bench_function("qr_4x4_const_faer", |bencher| {
    //     bencher.iter(|| qr_4x4_const_backend_faer(bb(&a_4x4_const)))
    // });

    // ========================================================================
    // NxN benchmarks
    // ========================================================================

    let a_n_dyn: DArray<_, 1> = (0..N * N)
        .map(|x| (x as f64) * 0.5)
        .collect::<Vec<_>>()
        .into();
    let a_n_dyn = a_n_dyn.reshape((N as usize, N as usize));
    let a_n_data: Vec<f64> = (0..N * N).map(|x| (x as f64) * 0.5).collect();

    // With allocation
    crit.bench_function("qr_n_dyn_naive", |bencher| {
        bencher.iter(|| qr_n_dyn_backend_naive(bb(&a_n_dyn)))
    });

    crit.bench_function("qr_n_dyn_lapack", |bencher| {
        bencher.iter(|| qr_n_dyn_backend_lapack(bb(&a_n_dyn)))
    });

    crit.bench_function("qr_n_dyn_faer", |bencher| {
        bencher.iter(|| qr_n_dyn_backend_faer(bb(&a_n_dyn)))
    });

    crit.bench_function("qr_n_nalgebra", |bencher| {
        bencher.iter(|| qr_n_nalgebra(bb(&a_n_data)))
    });
}

criterion_group!(benches, criterion_benchmark);
criterion_main!(benches);