1 | // Copyright 2018 Developers of the Rand project. |
2 | // |
3 | // Licensed under the Apache License, Version 2.0 <LICENSE-APACHE or |
4 | // https://www.apache.org/licenses/LICENSE-2.0> or the MIT license |
5 | // <LICENSE-MIT or https://opensource.org/licenses/MIT>, at your |
6 | // option. This file may not be copied, modified, or distributed |
7 | // except according to those terms. |
8 | |
9 | //! The implementations of the `StandardUniform` distribution for integer types. |
10 | |
11 | use crate::distr::{Distribution, StandardUniform}; |
12 | use crate::Rng; |
13 | #[cfg (all(target_arch = "x86" , feature = "simd_support" ))] |
14 | use core::arch::x86::__m512i; |
15 | #[cfg (target_arch = "x86" )] |
16 | use core::arch::x86::{__m128i, __m256i}; |
17 | #[cfg (all(target_arch = "x86_64" , feature = "simd_support" ))] |
18 | use core::arch::x86_64::__m512i; |
19 | #[cfg (target_arch = "x86_64" )] |
20 | use core::arch::x86_64::{__m128i, __m256i}; |
21 | use core::num::{ |
22 | NonZeroI128, NonZeroI16, NonZeroI32, NonZeroI64, NonZeroI8, NonZeroU128, NonZeroU16, |
23 | NonZeroU32, NonZeroU64, NonZeroU8, |
24 | }; |
25 | #[cfg (feature = "simd_support" )] |
26 | use core::simd::*; |
27 | |
28 | impl Distribution<u8> for StandardUniform { |
29 | #[inline ] |
30 | fn sample<R: Rng + ?Sized>(&self, rng: &mut R) -> u8 { |
31 | rng.next_u32() as u8 |
32 | } |
33 | } |
34 | |
35 | impl Distribution<u16> for StandardUniform { |
36 | #[inline ] |
37 | fn sample<R: Rng + ?Sized>(&self, rng: &mut R) -> u16 { |
38 | rng.next_u32() as u16 |
39 | } |
40 | } |
41 | |
42 | impl Distribution<u32> for StandardUniform { |
43 | #[inline ] |
44 | fn sample<R: Rng + ?Sized>(&self, rng: &mut R) -> u32 { |
45 | rng.next_u32() |
46 | } |
47 | } |
48 | |
49 | impl Distribution<u64> for StandardUniform { |
50 | #[inline ] |
51 | fn sample<R: Rng + ?Sized>(&self, rng: &mut R) -> u64 { |
52 | rng.next_u64() |
53 | } |
54 | } |
55 | |
56 | impl Distribution<u128> for StandardUniform { |
57 | #[inline ] |
58 | fn sample<R: Rng + ?Sized>(&self, rng: &mut R) -> u128 { |
59 | // Use LE; we explicitly generate one value before the next. |
60 | let x = u128::from(rng.next_u64()); |
61 | let y = u128::from(rng.next_u64()); |
62 | (y << 64) | x |
63 | } |
64 | } |
65 | |
66 | macro_rules! impl_int_from_uint { |
67 | ($ty:ty, $uty:ty) => { |
68 | impl Distribution<$ty> for StandardUniform { |
69 | #[inline] |
70 | fn sample<R: Rng + ?Sized>(&self, rng: &mut R) -> $ty { |
71 | rng.random::<$uty>() as $ty |
72 | } |
73 | } |
74 | }; |
75 | } |
76 | |
77 | impl_int_from_uint! { i8, u8 } |
78 | impl_int_from_uint! { i16, u16 } |
79 | impl_int_from_uint! { i32, u32 } |
80 | impl_int_from_uint! { i64, u64 } |
81 | impl_int_from_uint! { i128, u128 } |
82 | |
83 | macro_rules! impl_nzint { |
84 | ($ty:ty, $new:path) => { |
85 | impl Distribution<$ty> for StandardUniform { |
86 | fn sample<R: Rng + ?Sized>(&self, rng: &mut R) -> $ty { |
87 | loop { |
88 | if let Some(nz) = $new(rng.random()) { |
89 | break nz; |
90 | } |
91 | } |
92 | } |
93 | } |
94 | }; |
95 | } |
96 | |
97 | impl_nzint!(NonZeroU8, NonZeroU8::new); |
98 | impl_nzint!(NonZeroU16, NonZeroU16::new); |
99 | impl_nzint!(NonZeroU32, NonZeroU32::new); |
100 | impl_nzint!(NonZeroU64, NonZeroU64::new); |
101 | impl_nzint!(NonZeroU128, NonZeroU128::new); |
102 | |
103 | impl_nzint!(NonZeroI8, NonZeroI8::new); |
104 | impl_nzint!(NonZeroI16, NonZeroI16::new); |
105 | impl_nzint!(NonZeroI32, NonZeroI32::new); |
106 | impl_nzint!(NonZeroI64, NonZeroI64::new); |
107 | impl_nzint!(NonZeroI128, NonZeroI128::new); |
108 | |
109 | #[cfg (any(target_arch = "x86" , target_arch = "x86_64" ))] |
110 | impl Distribution<__m128i> for StandardUniform { |
111 | #[inline ] |
112 | fn sample<R: Rng + ?Sized>(&self, rng: &mut R) -> __m128i { |
113 | // NOTE: It's tempting to use the u128 impl here, but confusingly this |
114 | // results in different code (return via rdx, r10 instead of rax, rdx |
115 | // with u128 impl) and is much slower (+130 time). This version calls |
116 | // impls::fill_bytes_via_next but performs well. |
117 | |
118 | let mut buf: [u8; _] = [0_u8; core::mem::size_of::<__m128i>()]; |
119 | rng.fill_bytes(&mut buf); |
120 | // x86 is little endian so no need for conversion |
121 | |
122 | // SAFETY: All byte sequences of `buf` represent values of the output type. |
123 | unsafe { core::mem::transmute(buf) } |
124 | } |
125 | } |
126 | |
127 | #[cfg (any(target_arch = "x86" , target_arch = "x86_64" ))] |
128 | impl Distribution<__m256i> for StandardUniform { |
129 | #[inline ] |
130 | fn sample<R: Rng + ?Sized>(&self, rng: &mut R) -> __m256i { |
131 | let mut buf: [u8; _] = [0_u8; core::mem::size_of::<__m256i>()]; |
132 | rng.fill_bytes(&mut buf); |
133 | // x86 is little endian so no need for conversion |
134 | |
135 | // SAFETY: All byte sequences of `buf` represent values of the output type. |
136 | unsafe { core::mem::transmute(buf) } |
137 | } |
138 | } |
139 | |
140 | #[cfg (all( |
141 | any(target_arch = "x86" , target_arch = "x86_64" ), |
142 | feature = "simd_support" |
143 | ))] |
144 | impl Distribution<__m512i> for StandardUniform { |
145 | #[inline ] |
146 | fn sample<R: Rng + ?Sized>(&self, rng: &mut R) -> __m512i { |
147 | let mut buf = [0_u8; core::mem::size_of::<__m512i>()]; |
148 | rng.fill_bytes(&mut buf); |
149 | // x86 is little endian so no need for conversion |
150 | |
151 | // SAFETY: All byte sequences of `buf` represent values of the output type. |
152 | unsafe { core::mem::transmute(buf) } |
153 | } |
154 | } |
155 | |
156 | #[cfg (feature = "simd_support" )] |
157 | macro_rules! simd_impl { |
158 | ($($ty:ty),+) => {$( |
159 | /// Requires nightly Rust and the [`simd_support`] feature |
160 | /// |
161 | /// [`simd_support`]: https://github.com/rust-random/rand#crate-features |
162 | #[cfg(feature = "simd_support" )] |
163 | impl<const LANES: usize> Distribution<Simd<$ty, LANES>> for StandardUniform |
164 | where |
165 | LaneCount<LANES>: SupportedLaneCount, |
166 | { |
167 | #[inline] |
168 | fn sample<R: Rng + ?Sized>(&self, rng: &mut R) -> Simd<$ty, LANES> { |
169 | let mut vec = Simd::default(); |
170 | rng.fill(vec.as_mut_array().as_mut_slice()); |
171 | vec |
172 | } |
173 | } |
174 | )+}; |
175 | } |
176 | |
177 | #[cfg (feature = "simd_support" )] |
178 | simd_impl!(u8, i8, u16, i16, u32, i32, u64, i64); |
179 | |
180 | #[cfg (test)] |
181 | mod tests { |
182 | use super::*; |
183 | |
184 | #[test] |
185 | fn test_integers() { |
186 | let mut rng = crate::test::rng(806); |
187 | |
188 | rng.sample::<i8, _>(StandardUniform); |
189 | rng.sample::<i16, _>(StandardUniform); |
190 | rng.sample::<i32, _>(StandardUniform); |
191 | rng.sample::<i64, _>(StandardUniform); |
192 | rng.sample::<i128, _>(StandardUniform); |
193 | |
194 | rng.sample::<u8, _>(StandardUniform); |
195 | rng.sample::<u16, _>(StandardUniform); |
196 | rng.sample::<u32, _>(StandardUniform); |
197 | rng.sample::<u64, _>(StandardUniform); |
198 | rng.sample::<u128, _>(StandardUniform); |
199 | } |
200 | |
201 | #[cfg (any(target_arch = "x86" , target_arch = "x86_64" ))] |
202 | #[test] |
203 | fn x86_integers() { |
204 | let mut rng = crate::test::rng(807); |
205 | |
206 | rng.sample::<__m128i, _>(StandardUniform); |
207 | rng.sample::<__m256i, _>(StandardUniform); |
208 | #[cfg (feature = "simd_support" )] |
209 | rng.sample::<__m512i, _>(StandardUniform); |
210 | } |
211 | |
212 | #[test] |
213 | fn value_stability() { |
214 | fn test_samples<T: Copy + core::fmt::Debug + PartialEq>(zero: T, expected: &[T]) |
215 | where |
216 | StandardUniform: Distribution<T>, |
217 | { |
218 | let mut rng = crate::test::rng(807); |
219 | let mut buf = [zero; 3]; |
220 | for x in &mut buf { |
221 | *x = rng.sample(StandardUniform); |
222 | } |
223 | assert_eq!(&buf, expected); |
224 | } |
225 | |
226 | test_samples(0u8, &[9, 247, 111]); |
227 | test_samples(0u16, &[32265, 42999, 38255]); |
228 | test_samples(0u32, &[2220326409, 2575017975, 2018088303]); |
229 | test_samples( |
230 | 0u64, |
231 | &[ |
232 | 11059617991457472009, |
233 | 16096616328739788143, |
234 | 1487364411147516184, |
235 | ], |
236 | ); |
237 | test_samples( |
238 | 0u128, |
239 | &[ |
240 | 296930161868957086625409848350820761097, |
241 | 145644820879247630242265036535529306392, |
242 | 111087889832015897993126088499035356354, |
243 | ], |
244 | ); |
245 | |
246 | test_samples(0i8, &[9, -9, 111]); |
247 | // Skip further i* types: they are simple reinterpretation of u* samples |
248 | |
249 | #[cfg (feature = "simd_support" )] |
250 | { |
251 | // We only test a sub-set of types here and make assumptions about the rest. |
252 | |
253 | test_samples( |
254 | u8x4::default(), |
255 | &[ |
256 | u8x4::from([9, 126, 87, 132]), |
257 | u8x4::from([247, 167, 123, 153]), |
258 | u8x4::from([111, 149, 73, 120]), |
259 | ], |
260 | ); |
261 | test_samples( |
262 | u8x8::default(), |
263 | &[ |
264 | u8x8::from([9, 126, 87, 132, 247, 167, 123, 153]), |
265 | u8x8::from([111, 149, 73, 120, 68, 171, 98, 223]), |
266 | u8x8::from([24, 121, 1, 50, 13, 46, 164, 20]), |
267 | ], |
268 | ); |
269 | |
270 | test_samples( |
271 | i64x8::default(), |
272 | &[ |
273 | i64x8::from([ |
274 | -7387126082252079607, |
275 | -2350127744969763473, |
276 | 1487364411147516184, |
277 | 7895421560427121838, |
278 | 602190064936008898, |
279 | 6022086574635100741, |
280 | -5080089175222015595, |
281 | -4066367846667249123, |
282 | ]), |
283 | i64x8::from([ |
284 | 9180885022207963908, |
285 | 3095981199532211089, |
286 | 6586075293021332726, |
287 | 419343203796414657, |
288 | 3186951873057035255, |
289 | 5287129228749947252, |
290 | 444726432079249540, |
291 | -1587028029513790706, |
292 | ]), |
293 | i64x8::from([ |
294 | 6075236523189346388, |
295 | 1351763722368165432, |
296 | -6192309979959753740, |
297 | -7697775502176768592, |
298 | -4482022114172078123, |
299 | 7522501477800909500, |
300 | -1837258847956201231, |
301 | -586926753024886735, |
302 | ]), |
303 | ], |
304 | ); |
305 | } |
306 | } |
307 | } |
308 | |