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) = $new(rng.random()) { |
89 | break nz; |
90 | } |
91 | } |
92 | } |
93 | } |
94 | }; |
95 | } |
96 | |
97 | impl_nzint!(NonZeroU8, NonZeroU8::new); |
98 | impl_nzint!(NonZeroU16, NonZeroU16::new); |
99 | impl_nzint!(NonZeroU32, NonZeroU32::new); |
100 | impl_nzint!(NonZeroU64, NonZeroU64::new); |
101 | impl_nzint!(NonZeroU128, NonZeroU128::new); |
102 | |
103 | impl_nzint!(NonZeroI8, NonZeroI8::new); |
104 | impl_nzint!(NonZeroI16, NonZeroI16::new); |
105 | impl_nzint!(NonZeroI32, NonZeroI32::new); |
106 | impl_nzint!(NonZeroI64, NonZeroI64::new); |
107 | impl_nzint!(NonZeroI128, NonZeroI128::new); |
108 | |
109 | #[cfg (any(target_arch = "x86" , target_arch = "x86_64" ))] |
110 | macro_rules! x86_intrinsic_impl { |
111 | ($meta:meta, $($intrinsic:ident),+) => {$( |
112 | #[cfg($meta)] |
113 | impl Distribution<$intrinsic> for StandardUniform { |
114 | #[inline] |
115 | fn sample<R: Rng + ?Sized>(&self, rng: &mut R) -> $intrinsic { |
116 | // On proper hardware, this should compile to SIMD instructions |
117 | // Verified on x86 Haswell with __m128i, __m256i |
118 | let mut buf = [0_u8; core::mem::size_of::<$intrinsic>()]; |
119 | rng.fill_bytes(&mut buf); |
120 | // x86 is little endian so no need for conversion |
121 | zerocopy::transmute!(buf) |
122 | } |
123 | } |
124 | )+}; |
125 | } |
126 | |
127 | #[cfg (feature = "simd_support" )] |
128 | macro_rules! simd_impl { |
129 | ($($ty:ty),+) => {$( |
130 | /// Requires nightly Rust and the [`simd_support`] feature |
131 | /// |
132 | /// [`simd_support`]: https://github.com/rust-random/rand#crate-features |
133 | #[cfg(feature = "simd_support" )] |
134 | impl<const LANES: usize> Distribution<Simd<$ty, LANES>> for StandardUniform |
135 | where |
136 | LaneCount<LANES>: SupportedLaneCount, |
137 | { |
138 | #[inline] |
139 | fn sample<R: Rng + ?Sized>(&self, rng: &mut R) -> Simd<$ty, LANES> { |
140 | let mut vec = Simd::default(); |
141 | rng.fill(vec.as_mut_array().as_mut_slice()); |
142 | vec |
143 | } |
144 | } |
145 | )+}; |
146 | } |
147 | |
148 | #[cfg (feature = "simd_support" )] |
149 | simd_impl!(u8, i8, u16, i16, u32, i32, u64, i64); |
150 | |
151 | #[cfg (any(target_arch = "x86" , target_arch = "x86_64" ))] |
152 | x86_intrinsic_impl!( |
153 | any(target_arch = "x86" , target_arch = "x86_64" ), |
154 | __m128i, |
155 | __m256i |
156 | ); |
157 | #[cfg (all( |
158 | any(target_arch = "x86" , target_arch = "x86_64" ), |
159 | feature = "simd_support" |
160 | ))] |
161 | x86_intrinsic_impl!( |
162 | all( |
163 | any(target_arch = "x86" , target_arch = "x86_64" ), |
164 | feature = "simd_support" |
165 | ), |
166 | __m512i |
167 | ); |
168 | |
169 | #[cfg (test)] |
170 | mod tests { |
171 | use super::*; |
172 | |
173 | #[test] |
174 | fn test_integers() { |
175 | let mut rng = crate::test::rng(806); |
176 | |
177 | rng.sample::<i8, _>(StandardUniform); |
178 | rng.sample::<i16, _>(StandardUniform); |
179 | rng.sample::<i32, _>(StandardUniform); |
180 | rng.sample::<i64, _>(StandardUniform); |
181 | rng.sample::<i128, _>(StandardUniform); |
182 | |
183 | rng.sample::<u8, _>(StandardUniform); |
184 | rng.sample::<u16, _>(StandardUniform); |
185 | rng.sample::<u32, _>(StandardUniform); |
186 | rng.sample::<u64, _>(StandardUniform); |
187 | rng.sample::<u128, _>(StandardUniform); |
188 | } |
189 | |
190 | #[cfg (any(target_arch = "x86" , target_arch = "x86_64" ))] |
191 | #[test] |
192 | fn x86_integers() { |
193 | let mut rng = crate::test::rng(807); |
194 | |
195 | rng.sample::<__m128i, _>(StandardUniform); |
196 | rng.sample::<__m256i, _>(StandardUniform); |
197 | #[cfg (feature = "simd_support" )] |
198 | rng.sample::<__m512i, _>(StandardUniform); |
199 | } |
200 | |
201 | #[test] |
202 | fn value_stability() { |
203 | fn test_samples<T: Copy + core::fmt::Debug + PartialEq>(zero: T, expected: &[T]) |
204 | where |
205 | StandardUniform: Distribution<T>, |
206 | { |
207 | let mut rng = crate::test::rng(807); |
208 | let mut buf = [zero; 3]; |
209 | for x in &mut buf { |
210 | *x = rng.sample(StandardUniform); |
211 | } |
212 | assert_eq!(&buf, expected); |
213 | } |
214 | |
215 | test_samples(0u8, &[9, 247, 111]); |
216 | test_samples(0u16, &[32265, 42999, 38255]); |
217 | test_samples(0u32, &[2220326409, 2575017975, 2018088303]); |
218 | test_samples( |
219 | 0u64, |
220 | &[ |
221 | 11059617991457472009, |
222 | 16096616328739788143, |
223 | 1487364411147516184, |
224 | ], |
225 | ); |
226 | test_samples( |
227 | 0u128, |
228 | &[ |
229 | 296930161868957086625409848350820761097, |
230 | 145644820879247630242265036535529306392, |
231 | 111087889832015897993126088499035356354, |
232 | ], |
233 | ); |
234 | |
235 | test_samples(0i8, &[9, -9, 111]); |
236 | // Skip further i* types: they are simple reinterpretation of u* samples |
237 | |
238 | #[cfg (feature = "simd_support" )] |
239 | { |
240 | // We only test a sub-set of types here and make assumptions about the rest. |
241 | |
242 | test_samples( |
243 | u8x4::default(), |
244 | &[ |
245 | u8x4::from([9, 126, 87, 132]), |
246 | u8x4::from([247, 167, 123, 153]), |
247 | u8x4::from([111, 149, 73, 120]), |
248 | ], |
249 | ); |
250 | test_samples( |
251 | u8x8::default(), |
252 | &[ |
253 | u8x8::from([9, 126, 87, 132, 247, 167, 123, 153]), |
254 | u8x8::from([111, 149, 73, 120, 68, 171, 98, 223]), |
255 | u8x8::from([24, 121, 1, 50, 13, 46, 164, 20]), |
256 | ], |
257 | ); |
258 | |
259 | test_samples( |
260 | i64x8::default(), |
261 | &[ |
262 | i64x8::from([ |
263 | -7387126082252079607, |
264 | -2350127744969763473, |
265 | 1487364411147516184, |
266 | 7895421560427121838, |
267 | 602190064936008898, |
268 | 6022086574635100741, |
269 | -5080089175222015595, |
270 | -4066367846667249123, |
271 | ]), |
272 | i64x8::from([ |
273 | 9180885022207963908, |
274 | 3095981199532211089, |
275 | 6586075293021332726, |
276 | 419343203796414657, |
277 | 3186951873057035255, |
278 | 5287129228749947252, |
279 | 444726432079249540, |
280 | -1587028029513790706, |
281 | ]), |
282 | i64x8::from([ |
283 | 6075236523189346388, |
284 | 1351763722368165432, |
285 | -6192309979959753740, |
286 | -7697775502176768592, |
287 | -4482022114172078123, |
288 | 7522501477800909500, |
289 | -1837258847956201231, |
290 | -586926753024886735, |
291 | ]), |
292 | ], |
293 | ); |
294 | } |
295 | } |
296 | } |
297 | |