bola 0.1.1

A lightweight native rust linear algebra library.
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

#[cfg(target_arch = "x86_64")]
use std::{arch::x86_64::*, mem};

#[cfg(target_arch = "aarch64")]
use std::arch::aarch64::*;

/// Computes the dot product of two vectors.
///
/// # Arguments
/// * `n`    - The number of elements in both vectors `x` and `y`.
/// * `x`    - A raw constant pointer to the first element of vector `x`.
/// * `incx` - The stride between consecutive elements of `x`.
/// * `y`    - A raw constant pointer to the first element of vector `y`.
/// * `incy` - The stride between consecutive elements of `y`.
///
/// # Returns
/// The dot product as an `f64` value.
#[cfg(target_arch = "x86_64")]
#[inline]
#[allow(unsafe_op_in_unsafe_fn, clippy::missing_safety_doc)]
#[target_feature(enable = "avx2,fma")]
pub unsafe fn dot_product(n: usize, x: *const f64, incx: usize, y: *const f64, incy: usize) -> f64 {
    if n == 0 {
        return 0.0;
    }

    if incx == 1 && incy == 1 {
        let n_chunks = n / 4;
        let mut sum_vec = _mm256_setzero_pd();
        for i in 0..n_chunks {
            let x_vec = _mm256_loadu_pd(x.add(i * 4));
            let y_vec = _mm256_loadu_pd(y.add(i * 4));
            sum_vec = _mm256_fmadd_pd(x_vec, y_vec, sum_vec);
        }

        // Horizontal sum of the vector register
        let mut temp_sum = 0.0;
        let temp_arr: [f64; 4] = mem::transmute(sum_vec);
        temp_sum += temp_arr.iter().sum::<f64>();

        // Handle remainder
        for i in (n_chunks * 4)..n {
            temp_sum += *x.add(i) * *y.add(i);
        }
        temp_sum
    } else {
        // Scalar fallback for non-unit strides
        let mut temp_sum = 0.0;
        for i in 0..n {
            temp_sum += *x.add(i * incx) * *y.add(i * incy);
        }
        temp_sum
    }
}

/// Computes the dot product of two vectors.
/// ... see above function
#[cfg(target_arch = "aarch64")]
#[inline]
#[allow(unsafe_op_in_unsafe_fn, clippy::missing_safety_doc)]
#[target_feature(enable = "neon")]
pub unsafe fn dot_product(n: usize, x: *const f64, incx: usize, y: *const f64, incy: usize) -> f64 {
    if n == 0 {
        return 0.0;
    }

    if incx == 1 && incy == 1 {
        let n_chunks = n / 2;
        let mut sum_vec = vdupq_n_f64(0.0);
        for i in 0..n_chunks {
            let x_vec = vld1q_f64(x.add(i * 2));
            let y_vec = vld1q_f64(y.add(i * 2));
            sum_vec = vfmaq_f64(sum_vec, x_vec, y_vec);
        }

        // Horizontal sum of the vector register
        let mut temp_sum = vaddvq_f64(sum_vec);

        // Handle remainder
        for i in (n_chunks * 2)..n {
            temp_sum += *x.add(i) * *y.add(i);
        }
        temp_sum
    } else {
        // Scalar fallback for non-unit strides
        let mut temp_sum = 0.0;
        for i in 0..n {
            temp_sum += *x.add(i * incx) * *y.add(i * incy);
        }
        temp_sum
    }
}

/// Performs a SIMD-accelerated AXPY operation.
///
/// # Arguments
/// * `n`     - The number of elements in vectors `x` and `y`.
/// * `alpha` - The scalar multiplier `alpha`.
/// * `x`     - A raw constant pointer to the first element of vector `x`.
/// * `incx`  - The stride between consecutive elements of `x`.
/// * `y`     - A raw mutable pointer to the first element of vector `y`.
/// * `incy`  - The stride between consecutive elements of `y`.
#[cfg(target_arch = "x86_64")]
#[inline]
#[allow(unsafe_op_in_unsafe_fn, clippy::missing_safety_doc)]
#[target_feature(enable = "avx2,fma")]
pub unsafe fn daxpy_update(n: usize, alpha: f64, x: *const f64, incx: usize, y: *mut f64, incy: usize) {
    if n == 0 || alpha == 0.0 {
        return;
    }

    if incx == 1 && incy == 1 {
        let n_chunks = n / 4;
        let alpha_vec = _mm256_set1_pd(alpha);
        for i in 0..n_chunks {
            let x_ptr = x.add(i * 4);
            let y_ptr = y.add(i * 4);
            _mm256_storeu_pd(
                y_ptr,
                _mm256_fmadd_pd(alpha_vec, _mm256_loadu_pd(x_ptr), _mm256_loadu_pd(y_ptr)),
            );
        }
        for i in (n_chunks * 4)..n {
            *y.add(i) += alpha * (*x.add(i));
        }
    } else {
        // Scalar fallback for non-unit increments
        for i in 0..n {
            *y.add(i * incy) += alpha * (*x.add(i * incx));
        }
    }
}

/// Performs a SIMD-accelerated AXPY operation.
/// ... see above function
#[cfg(target_arch = "aarch64")]
#[inline]
#[allow(unsafe_op_in_unsafe_fn, clippy::missing_safety_doc)]
#[target_feature(enable = "neon")]
pub unsafe fn daxpy_update(n: usize, alpha: f64, x: *const f64, incx: usize, y: *mut f64, incy: usize) {
    if n == 0 || alpha == 0.0 {
        return;
    }

    if incx == 1 && incy == 1 {
        let n_chunks = n / 2;
        let alpha_vec = vdupq_n_f64(alpha);
        for i in 0..n_chunks {
            let x_ptr = x.add(i * 2);
            let y_ptr = y.add(i * 2);
            let x_vec = vld1q_f64(x_ptr);
            let y_vec = vld1q_f64(y_ptr);
            vst1q_f64(y_ptr, vfmaq_f64(y_vec, alpha_vec, x_vec));
        }
        for i in (n_chunks * 2)..n {
            *y.add(i) += alpha * (*x.add(i));
        }
    } else {
        // Scalar fallback for non-unit increments
        for i in 0..n {
            *y.add(i * incy) += alpha * (*x.add(i * incx));
        }
    }
}