faer-core 0.17.1

Basic linear algebra routines
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

pub use faer_entity::{
    one_simd_as_slice, simd_as_slice, simd_as_slice_unit, simd_index_as_slice, slice_as_mut_simd,
    slice_as_simd,
};

fn sum_i32_scalar(slice: &[i32]) -> Option<i32> {
    let mut overflow = false;
    let mut sum = 0i32;
    for &v in slice {
        let o;
        (sum, o) = i32::overflowing_add(sum, v);
        overflow |= o;
    }
    (!overflow).then_some(sum)
}
fn sum_i64_scalar(slice: &[i64]) -> Option<i64> {
    let mut overflow = false;
    let mut sum = 0i64;
    for &v in slice {
        let o;
        (sum, o) = i64::overflowing_add(sum, v);
        overflow |= o;
    }
    (!overflow).then_some(sum)
}

pub fn sum_i32(slice: &[i32]) -> Option<i32> {
    #[cfg(any(target_arch = "x86", target_arch = "x86_64"))]
    if let Some(simd) = pulp::x86::V3::try_new() {
        return x86::sum_i32_v3(simd, slice);
    }
    sum_i32_scalar(slice)
}

pub fn sum_i64(slice: &[i64]) -> Option<i64> {
    #[cfg(any(target_arch = "x86", target_arch = "x86_64"))]
    if let Some(simd) = pulp::x86::V3::try_new() {
        return x86::sum_i64_v3(simd, slice);
    }
    sum_i64_scalar(slice)
}

#[cfg(any(target_arch = "x86", target_arch = "x86_64"))]
mod x86 {
    use super::*;
    use pulp::{x86::*, *};
    pub fn sum_i32_v3(simd: V3, slice: &[i32]) -> Option<i32> {
        struct Impl<'a> {
            simd: V3,
            slice: &'a [i32],
        }

        impl pulp::NullaryFnOnce for Impl<'_> {
            type Output = Option<i32>;

            #[inline(always)]
            fn call(self) -> Self::Output {
                let Self { simd, slice } = self;
                let (head, tail) = V3::i32s_as_simd(slice);

                let zero = simd.splat_i32x8(0);
                let mut sum = zero;
                let mut overflow = simd.splat_m32x8(m32::new(false));

                for &v in head {
                    sum = simd.wrapping_add_i32x8(sum, v);
                    overflow = simd.or_m32x8(overflow, simd.cmp_lt_i32x8(sum, zero));
                }

                if overflow != simd.splat_m32x8(m32::new(false)) {
                    return None;
                }

                i32::checked_add(
                    sum_i32_scalar(tail)?,
                    sum_i32_scalar(bytemuck::cast_slice(&[sum]))?,
                )
            }
        }

        simd.vectorize(Impl { simd, slice })
    }

    pub fn sum_i64_v3(simd: V3, slice: &[i64]) -> Option<i64> {
        struct Impl<'a> {
            simd: V3,
            slice: &'a [i64],
        }

        impl pulp::NullaryFnOnce for Impl<'_> {
            type Output = Option<i64>;

            #[inline(always)]
            fn call(self) -> Self::Output {
                let Self { simd, slice } = self;
                let (head, tail) = V3::i64s_as_simd(slice);

                let zero = simd.splat_i64x4(0);
                let mut sum = zero;
                let mut overflow = simd.splat_m64x4(m64::new(false));

                for &v in head {
                    sum = simd.wrapping_add_i64x4(sum, v);
                    overflow = simd.or_m64x4(overflow, simd.cmp_lt_i64x4(sum, zero));
                }

                if overflow != simd.splat_m64x4(m64::new(false)) {
                    return None;
                }

                i64::checked_add(
                    sum_i64_scalar(tail)?,
                    sum_i64_scalar(bytemuck::cast_slice(&[sum]))?,
                )
            }
        }

        simd.vectorize(Impl { simd, slice })
    }
}

#[cfg(test)]
mod tests {
    use super::*;
    use crate::assert;

    #[test]
    fn test_sum_i32() {
        let array = vec![100_000_000i32; 1000];
        assert!(sum_i32(&array) == None);

        let array = vec![1_000_000i32; 1000];
        assert!(sum_i32(&array) == Some(1_000_000_000i32));
    }

    #[test]
    fn test_sum_i64() {
        let array = vec![i64::MAX / 100; 1000];
        assert!(sum_i64(&array) == None);

        let array = vec![100_000_000i64; 1000];
        assert!(sum_i64(&array) == Some(100_000_000_000i64));

        let array = vec![1_000_000i64; 1000];
        assert!(sum_i64(&array) == Some(1_000_000_000i64));
    }
}