1 // Copyright 2018 Developers of the Rand project.
2 // Copyright 2017 Paul Dicker.
3 // Copyright 2014-2017 Melissa O'Neill and PCG Project contributors
4 //
5 // Licensed under the Apache License, Version 2.0 <LICENSE-APACHE or
6 // https://www.apache.org/licenses/LICENSE-2.0> or the MIT license
7 // <LICENSE-MIT or https://opensource.org/licenses/MIT>, at your
8 // option. This file may not be copied, modified, or distributed
9 // except according to those terms.
10 
11 //! PCG random number generators
12 
13 // This is the default multiplier used by PCG for 64-bit state.
14 const MULTIPLIER: u128 = 0x2360_ED05_1FC6_5DA4_4385_DF64_9FCC_F645;
15 
16 use core::fmt;
17 use rand_core::{RngCore, SeedableRng, Error, le};
18 #[cfg(feature="serde1")] use serde::{Serialize, Deserialize};
19 
20 /// A PCG random number generator (XSL RR 128/64 (LCG) variant).
21 ///
22 /// Permuted Congruential Generator with 128-bit state, internal Linear
23 /// Congruential Generator, and 64-bit output via "xorshift low (bits),
24 /// random rotation" output function.
25 ///
26 /// This is a 128-bit LCG with explicitly chosen stream with the PCG-XSL-RR
27 /// output function. This combination is the standard `pcg64`.
28 ///
29 /// Despite the name, this implementation uses 32 bytes (256 bit) space
30 /// comprising 128 bits of state and 128 bits stream selector. These are both
31 /// set by `SeedableRng`, using a 256-bit seed.
32 #[derive(Clone)]
33 #[cfg_attr(feature="serde1", derive(Serialize,Deserialize))]
34 pub struct Lcg128Xsl64 {
35     state: u128,
36     increment: u128,
37 }
38 
39 /// `Lcg128Xsl64` is also officially known as `pcg64`.
40 pub type Pcg64 = Lcg128Xsl64;
41 
42 impl Lcg128Xsl64 {
43     /// Construct an instance compatible with PCG seed and stream.
44     ///
45     /// Note that PCG specifies default values for both parameters:
46     ///
47     /// - `state = 0xcafef00dd15ea5e5`
48     /// - `stream = 0xa02bdbf7bb3c0a7ac28fa16a64abf96`
new(state: u128, stream: u128) -> Self49     pub fn new(state: u128, stream: u128) -> Self {
50         // The increment must be odd, hence we discard one bit:
51         let increment = (stream << 1) | 1;
52         Lcg128Xsl64::from_state_incr(state, increment)
53     }
54 
55     #[inline]
from_state_incr(state: u128, increment: u128) -> Self56     fn from_state_incr(state: u128, increment: u128) -> Self {
57         let mut pcg = Lcg128Xsl64 { state, increment };
58         // Move away from inital value:
59         pcg.state = pcg.state.wrapping_add(pcg.increment);
60         pcg.step();
61         pcg
62     }
63 
64     #[inline]
step(&mut self)65     fn step(&mut self) {
66         // prepare the LCG for the next round
67         self.state = self.state
68             .wrapping_mul(MULTIPLIER)
69             .wrapping_add(self.increment);
70     }
71 }
72 
73 // Custom Debug implementation that does not expose the internal state
74 impl fmt::Debug for Lcg128Xsl64 {
fmt(&self, f: &mut fmt::Formatter) -> fmt::Result75     fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
76         write!(f, "Lcg128Xsl64 {{}}")
77     }
78 }
79 
80 /// We use a single 255-bit seed to initialise the state and select a stream.
81 /// One `seed` bit (lowest bit of `seed[8]`) is ignored.
82 impl SeedableRng for Lcg128Xsl64 {
83     type Seed = [u8; 32];
84 
from_seed(seed: Self::Seed) -> Self85     fn from_seed(seed: Self::Seed) -> Self {
86         let mut seed_u64 = [0u64; 4];
87         le::read_u64_into(&seed, &mut seed_u64);
88         let state = u128::from(seed_u64[0]) | (u128::from(seed_u64[1]) << 64);
89         let incr = u128::from(seed_u64[2]) | (u128::from(seed_u64[3]) << 64);
90 
91         // The increment must be odd, hence we discard one bit:
92         Lcg128Xsl64::from_state_incr(state, incr | 1)
93     }
94 }
95 
96 impl RngCore for Lcg128Xsl64 {
97     #[inline]
next_u32(&mut self) -> u3298     fn next_u32(&mut self) -> u32 {
99         self.next_u64() as u32
100     }
101 
102     #[inline]
next_u64(&mut self) -> u64103     fn next_u64(&mut self) -> u64 {
104         self.step();
105         output_xsl_rr(self.state)
106     }
107 
108     #[inline]
fill_bytes(&mut self, dest: &mut [u8])109     fn fill_bytes(&mut self, dest: &mut [u8]) {
110         fill_bytes_impl(self, dest)
111     }
112 
113     #[inline]
try_fill_bytes(&mut self, dest: &mut [u8]) -> Result<(), Error>114     fn try_fill_bytes(&mut self, dest: &mut [u8]) -> Result<(), Error> {
115         self.fill_bytes(dest);
116         Ok(())
117     }
118 }
119 
120 
121 /// A PCG random number generator (XSL 128/64 (MCG) variant).
122 ///
123 /// Permuted Congruential Generator with 128-bit state, internal Multiplicative
124 /// Congruential Generator, and 64-bit output via "xorshift low (bits),
125 /// random rotation" output function.
126 ///
127 /// This is a 128-bit MCG with the PCG-XSL-RR output function, also known as
128 /// `pcg64_fast`.
129 /// Note that compared to the standard `pcg64` (128-bit LCG with PCG-XSL-RR
130 /// output function), this RNG is faster, also has a long cycle, and still has
131 /// good performance on statistical tests.
132 #[derive(Clone)]
133 #[cfg_attr(feature="serde1", derive(Serialize,Deserialize))]
134 pub struct Mcg128Xsl64 {
135     state: u128,
136 }
137 
138 /// A friendly name for `Mcg128Xsl64` (also known as `pcg64_fast`).
139 pub type Pcg64Mcg = Mcg128Xsl64;
140 
141 impl Mcg128Xsl64 {
142     /// Construct an instance compatible with PCG seed.
143     ///
144     /// Note that PCG specifies a default value for the parameter:
145     ///
146     /// - `state = 0xcafef00dd15ea5e5`
new(state: u128) -> Self147     pub fn new(state: u128) -> Self {
148         // Force low bit to 1, as in C version (C++ uses `state | 3` instead).
149         Mcg128Xsl64 { state: state | 1 }
150     }
151 }
152 
153 // Custom Debug implementation that does not expose the internal state
154 impl fmt::Debug for Mcg128Xsl64 {
fmt(&self, f: &mut fmt::Formatter) -> fmt::Result155     fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
156         write!(f, "Mcg128Xsl64 {{}}")
157     }
158 }
159 
160 /// We use a single 126-bit seed to initialise the state and select a stream.
161 /// Two `seed` bits (lowest order of last byte) are ignored.
162 impl SeedableRng for Mcg128Xsl64 {
163     type Seed = [u8; 16];
164 
from_seed(seed: Self::Seed) -> Self165     fn from_seed(seed: Self::Seed) -> Self {
166         // Read as if a little-endian u128 value:
167         let mut seed_u64 = [0u64; 2];
168         le::read_u64_into(&seed, &mut seed_u64);
169         let state = u128::from(seed_u64[0])  |
170                     u128::from(seed_u64[1]) << 64;
171         Mcg128Xsl64::new(state)
172     }
173 }
174 
175 impl RngCore for Mcg128Xsl64 {
176     #[inline]
next_u32(&mut self) -> u32177     fn next_u32(&mut self) -> u32 {
178         self.next_u64() as u32
179     }
180 
181     #[inline]
next_u64(&mut self) -> u64182     fn next_u64(&mut self) -> u64 {
183         self.state = self.state.wrapping_mul(MULTIPLIER);
184         output_xsl_rr(self.state)
185     }
186 
187     #[inline]
fill_bytes(&mut self, dest: &mut [u8])188     fn fill_bytes(&mut self, dest: &mut [u8]) {
189         fill_bytes_impl(self, dest)
190     }
191 
192     #[inline]
try_fill_bytes(&mut self, dest: &mut [u8]) -> Result<(), Error>193     fn try_fill_bytes(&mut self, dest: &mut [u8]) -> Result<(), Error> {
194         self.fill_bytes(dest);
195         Ok(())
196     }
197 }
198 
199 #[inline(always)]
output_xsl_rr(state: u128) -> u64200 fn output_xsl_rr(state: u128) -> u64 {
201     // Output function XSL RR ("xorshift low (bits), random rotation")
202     // Constants are for 128-bit state, 64-bit output
203     const XSHIFT: u32 = 64;     // (128 - 64 + 64) / 2
204     const ROTATE: u32 = 122;    // 128 - 6
205 
206     let rot = (state >> ROTATE) as u32;
207     let xsl = ((state >> XSHIFT) as u64) ^ (state as u64);
208     xsl.rotate_right(rot)
209 }
210 
211 #[inline(always)]
fill_bytes_impl<R: RngCore + ?Sized>(rng: &mut R, dest: &mut [u8])212 fn fill_bytes_impl<R: RngCore + ?Sized>(rng: &mut R, dest: &mut [u8]) {
213     let mut left = dest;
214     while left.len() >= 8 {
215         let (l, r) = {left}.split_at_mut(8);
216         left = r;
217         let chunk: [u8; 8] = rng.next_u64().to_le_bytes();
218         l.copy_from_slice(&chunk);
219     }
220     let n = left.len();
221     if n > 0 {
222         let chunk: [u8; 8] = rng.next_u64().to_le_bytes();
223         left.copy_from_slice(&chunk[..n]);
224     }
225 }
226