urng 0.4.8

Universal Random Number Generator
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

use crate::{
    rng::{Rng32, Rng32V256, Rng32V512},
    rng32::SplitMix32,
};

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

/// JSF (Jenkins Small Fast) 32-bit RNG implementation.
///
/// # Example
/// ```
/// use urng::rng::Rng32;
/// use urng::rng32::Jsf32;
///
/// let mut rng = Jsf32::new(12345);
/// ```
#[repr(C, align(64))]
pub struct Jsf32 {
    pub(crate) a: u32,
    pub(crate) b: u32,
    pub(crate) c: u32,
    pub(crate) d: u32,
}

impl Jsf32 {
    /// Creates a new `Jsf32` instance with the given seed.
    pub fn new(seed: u32) -> Self {
        let mut seedgen = SplitMix32::new(seed);
        Self {
            a: 0xf1ea5eed,
            b: seedgen.nextu(),
            c: seedgen.nextu(),
            d: seedgen.nextu(),
        }
    }
}

impl Rng32 for Jsf32 {
    #[inline(always)]
    fn nextu(&mut self) -> u32 {
        let e = self.a.wrapping_sub(self.b.rotate_left(27));
        self.a = self.b ^ self.c.rotate_left(17);
        self.b = self.c.wrapping_add(self.d);
        self.c = self.d.wrapping_add(e);
        self.d = e.wrapping_add(self.a);
        self.d
    }
}

#[repr(C, align(64))]
pub struct Jsf32x8 {
    pub(crate) a: __m256i,
    pub(crate) b: __m256i,
    pub(crate) c: __m256i,
    pub(crate) d: __m256i,
}

pub(crate) const JSF32X8: usize = 8;

impl Jsf32x8 {
    /// # Safety
    #[target_feature(enable = "avx2")]
    pub fn new(seed: u32) -> Self {
        let mut seedgen = SplitMix32::new(seed);
        let mut sv = [[0u32; JSF32X8]; 3];
        for vals in sv.iter_mut() {
            for v in vals.iter_mut() {
                *v = seedgen.nextu();
            }
        }
        let a = [0xf1ea5eedu32; JSF32X8];
        unsafe {
            Self {
                a: _mm256_loadu_si256(a.as_ptr() as *const __m256i),
                b: _mm256_loadu_si256(sv[0].as_ptr() as *const __m256i),
                c: _mm256_loadu_si256(sv[1].as_ptr() as *const __m256i),
                d: _mm256_loadu_si256(sv[2].as_ptr() as *const __m256i),
            }
        }
    }
}

impl Rng32V256 for Jsf32x8 {
    #[inline]
    #[target_feature(enable = "avx2")]
    unsafe fn nextuv(&mut self) -> __m256i {
        let e = _mm256_sub_epi32(self.a, unsafe { _mm256_rol_epi32(self.b, 27) });
        self.a = _mm256_xor_si256(self.b, unsafe { _mm256_rol_epi32(self.c, 17) });
        self.b = _mm256_add_epi32(self.c, self.d);
        self.c = _mm256_add_epi32(self.d, e);
        self.d = _mm256_add_epi32(e, self.a);
        self.d
    }

    #[inline(always)]
    fn nextu(&mut self) -> [u32; JSF32X8] {
        unsafe { std::mem::transmute(self.nextuv()) }
    }
}

/// 16-way SIMD implementation of JSF (Jenkins Small Fast) 32-bit RNG.
/// This implementation uses AVX-512 instructions to generate 16 random numbers in parallel.
///
/// # Example
/// ```
/// use urng::rng::{Rng32, Rng32V512};
/// use urng::rng32::Jsf32x16;
///
/// let mut rng = unsafe { Jsf32x16::new(12345) };
/// let _ = unsafe { rng.nextu() };
/// ```
#[repr(C, align(64))]
pub struct Jsf32x16 {
    pub(crate) a: __m512i,
    pub(crate) b: __m512i,
    pub(crate) c: __m512i,
    pub(crate) d: __m512i,
}

pub(crate) const JSF32X16: usize = 16;

impl Jsf32x16 {
    /// # Safety
    #[target_feature(enable = "avx512f")]
    pub fn new(seed: u32) -> Self {
        let mut seedgen = SplitMix32::new(seed);
        let mut sv = [[0u32; JSF32X16]; 3];
        for vals in sv.iter_mut() {
            for v in vals.iter_mut() {
                *v = seedgen.nextu();
            }
        }
        const A: [u32; JSF32X16] = [0xf1ea5eedu32; JSF32X16];
        unsafe {
            Self {
                a: _mm512_loadu_si512(A.as_ptr() as *const __m512i),
                b: _mm512_loadu_si512(sv[0].as_ptr() as *const __m512i),
                c: _mm512_loadu_si512(sv[1].as_ptr() as *const __m512i),
                d: _mm512_loadu_si512(sv[2].as_ptr() as *const __m512i),
            }
        }
    }
}

impl Rng32V512 for Jsf32x16 {
    #[inline]
    #[target_feature(enable = "avx512f")]
    unsafe fn nextuv(&mut self) -> __m512i {
        let e = _mm512_sub_epi32(self.a, _mm512_rol_epi32(self.b, 27));
        self.a = _mm512_xor_si512(self.b, _mm512_rol_epi32(self.c, 17));
        self.b = _mm512_add_epi32(self.c, self.d);
        self.c = _mm512_add_epi32(self.d, e);
        self.d = _mm512_add_epi32(e, self.a);
        self.d
    }
}

#[cfg(test)]
mod tests {
    use super::*;
    use crate::{safe_test, unsafe_test};

    safe_test!(Jsf32);
    unsafe_test!(Jsf32x8);
    unsafe_test!(Jsf32x16);
}