|
1 /* |
|
2 * Copyright (c) 2013, 2019, Oracle and/or its affiliates. All rights reserved. |
|
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. |
|
4 * |
|
5 * This code is free software; you can redistribute it and/or modify it |
|
6 * under the terms of the GNU General Public License version 2 only, as |
|
7 * published by the Free Software Foundation. Oracle designates this |
|
8 * particular file as subject to the "Classpath" exception as provided |
|
9 * by Oracle in the LICENSE file that accompanied this code. |
|
10 * |
|
11 * This code is distributed in the hope that it will be useful, but WITHOUT |
|
12 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or |
|
13 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License |
|
14 * version 2 for more details (a copy is included in the LICENSE file that |
|
15 * accompanied this code). |
|
16 * |
|
17 * You should have received a copy of the GNU General Public License version |
|
18 * 2 along with this work; if not, write to the Free Software Foundation, |
|
19 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. |
|
20 * |
|
21 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA |
|
22 * or visit www.oracle.com if you need additional information or have any |
|
23 * questions. |
|
24 */ |
|
25 package java.util; |
|
26 |
|
27 import java.math.BigInteger; |
|
28 import java.util.concurrent.atomic.AtomicLong; |
|
29 |
|
30 /** |
|
31 * A generator of uniform pseudorandom values applicable for use in |
|
32 * (among other contexts) isolated parallel computations that may |
|
33 * generate subtasks. Class {@code L64X1024Random} implements |
|
34 * interfaces {@link java.util.Rng} and {@link java.util.SplittableRng}, |
|
35 * and therefore supports methods for producing pseudorandomly chosen |
|
36 * numbers of type {@code int}, {@code long}, {@code float}, and {@code double} |
|
37 * as well as creating new split-off {@code L64X1024Random} objects, |
|
38 * with similar usages as for class {@link java.util.SplittableRandom}. |
|
39 * |
|
40 * <p>Series of generated values pass the TestU01 BigCrush and PractRand test suites |
|
41 * that measure independence and uniformity properties of random number generators. |
|
42 * (Most recently validated with |
|
43 * <a href="http://simul.iro.umontreal.ca/testu01/tu01.html">version 1.2.3 of TestU01</a> |
|
44 * and <a href="http://pracrand.sourceforge.net">version 0.90 of PractRand</a>. |
|
45 * Note that TestU01 BigCrush was used to test not only values produced by the {@code nextLong()} |
|
46 * method but also the result of bit-reversing each value produced by {@code nextLong()}.) |
|
47 * These tests validate only the methods for certain |
|
48 * types and ranges, but similar properties are expected to hold, at |
|
49 * least approximately, for others as well. |
|
50 * |
|
51 * <p>{@code L64X1024Random} is a specific member of the LXM family of algorithms |
|
52 * for pseudorandom number generators. Every LXM generator consists of two |
|
53 * subgenerators; one is an LCG (Linear Congruential Generator) and the other is |
|
54 * an Xorshift generator. Each output of an LXM generator is the sum of one |
|
55 * output from each subgenerator, possibly processed by a final mixing function |
|
56 * (but {@code L64X1024Random} does not use a mixing function). |
|
57 * |
|
58 * <p>The LCG subgenerator for {@code L64X1024Random} has an update step of the |
|
59 * form {@code s = m * s + a}, where {@code s}, {@code m}, and {@code a} are all |
|
60 * of type {@code long}; {@code s} is the mutable state, the multiplier {@code m} |
|
61 * is fixed (the same for all instances of {@code L64X1024Random}}) and the addend |
|
62 * {@code a} is a parameter (a final field of the instance). The parameter |
|
63 * {@code a} is required to be odd (this allows the LCG to have the maximal |
|
64 * period, namely 2<sup>64</sup>); therefore there are 2<sup>63</sup> distinct choices |
|
65 * of parameter. |
|
66 * |
|
67 * <p>The Xorshift subgenerator for {@code L64X1024Random} is the {@code xoroshiro1024} |
|
68 * algorithm (parameters 25, 27, and 36), without any final scrambler such as "+" or "**". |
|
69 * Its state consists of an array {@code x} of sixteen {@code long} values, |
|
70 * which can take on any values provided that they are not all zero. |
|
71 * The period of this subgenerator is 2<sup>1024</sup>-1. |
|
72 * |
|
73 * <p> Because the periods 2<sup>64</sup> and 2<sup>1024</sup>-1 of the two subgenerators |
|
74 * are relatively prime, the <em>period</em> of any single {@code L64X1024Random} object |
|
75 * (the length of the series of generated 64-bit values before it repeats) is the product |
|
76 * of the periods of the subgenerators, that is, 2<sup>64</sup>(2<sup>1024</sup>-1), |
|
77 * which is just slightly smaller than 2<sup>1088</sup>. Moreover, if two distinct |
|
78 * {@code L64X1024Random} objects have different {@code a} parameters, then their |
|
79 * cycles of produced values will be different. |
|
80 * |
|
81 * <p>The 64-bit values produced by the {@code nextLong()} method are exactly equidistributed. |
|
82 * For any specific instance of {@code L64X1024Random}, over the course of its cycle each |
|
83 * of the 2<sup>64</sup> possible {@code long} values will be produced 2<sup>1024</sup>-1 times. |
|
84 * The values produced by the {@code nextInt()}, {@code nextFloat()}, and {@code nextDouble()} |
|
85 * methods are likewise exactly equidistributed. |
|
86 * |
|
87 * <p>In fact, the 64-bit values produced by the {@code nextLong()} method are 16-equidistributed. |
|
88 * To be precise: for any specific instance of {@code L64X1024Random}, consider |
|
89 * the (overlapping) length-16 subsequences of the cycle of 64-bit values produced by |
|
90 * {@code nextLong()} (assuming no other methods are called that would affect the state). |
|
91 * There are 2<sup>64</sup>(2<sup>1024</sup>-1) such subsequences, and each subsequence, |
|
92 * which consists of 16 64-bit values, can have one of 2<sup>1024</sup> values. Of those |
|
93 * 2<sup>1024</sup> subsequence values, nearly all of them (2<sup>1024</sup>-2<sup>64</sup>) |
|
94 * occur 2<sup>64</sup> times over the course of the entire cycle, and the other |
|
95 * 2<sup>64</sup> subsequence values occur only 2<sup>64</sup>-1 times. So the ratio |
|
96 * of the probability of getting one of the less common subsequence values and the |
|
97 * probability of getting one of the more common subsequence values is 1-2<sup>-64</sup>. |
|
98 * (Note that the set of 2<sup>64</sup> less-common subsequence values will differ from |
|
99 * one instance of {@code L64X1024Random} to another, as a function of the additive |
|
100 * parameter of the LCG.) The values produced by the {@code nextInt()}, {@code nextFloat()}, |
|
101 * and {@code nextDouble()} methods are likewise 16-equidistributed. |
|
102 * |
|
103 * <p>Method {@link #split} constructs and returns a new {@code L64X1024Random} |
|
104 * instance that shares no mutable state with the current instance. However, with |
|
105 * very high probability, the values collectively generated by the two objects |
|
106 * have the same statistical properties as if the same quantity of values were |
|
107 * generated by a single thread using a single {@code L64X1024Random} object. |
|
108 * This is because, with high probability, distinct {@code L64X1024Random} objects |
|
109 * have distinct {@code a} parameters and therefore use distinct members of the |
|
110 * algorithmic family; and even if their {@code a} parameters are the same, with |
|
111 * very high probability they will traverse different parts of their common state |
|
112 * cycle. |
|
113 * |
|
114 * <p>As with {@link java.util.SplittableRandom}, instances of |
|
115 * {@code L64X1024Random} are <em>not</em> thread-safe. |
|
116 * They are designed to be split, not shared, across threads. For |
|
117 * example, a {@link java.util.concurrent.ForkJoinTask} fork/join-style |
|
118 * computation using random numbers might include a construction |
|
119 * of the form {@code new Subtask(someL64X1024Random.split()).fork()}. |
|
120 * |
|
121 * <p>This class provides additional methods for generating random |
|
122 * streams, that employ the above techniques when used in |
|
123 * {@code stream.parallel()} mode. |
|
124 * |
|
125 * <p>Instances of {@code L64X1024Random} are not cryptographically |
|
126 * secure. Consider instead using {@link java.security.SecureRandom} |
|
127 * in security-sensitive applications. Additionally, |
|
128 * default-constructed instances do not use a cryptographically random |
|
129 * seed unless the {@linkplain System#getProperty system property} |
|
130 * {@code java.util.secureRandomSeed} is set to {@code true}. |
|
131 * |
|
132 * @author Guy Steele |
|
133 * @since 1.9 |
|
134 */ |
|
135 public final class L64X1024Random extends AbstractSplittableRng { |
|
136 |
|
137 /* |
|
138 * Implementation Overview. |
|
139 * |
|
140 * The split() operation uses the current generator to choose 18 new 64-bit |
|
141 * long values that are then used to initialize the parameter `a`, the |
|
142 * state variable `s`, and the array `x` for a newly constructed generator. |
|
143 * |
|
144 * With extremely high probability, no two generators so chosen |
|
145 * will have the same `a` parameter, and testing has indicated |
|
146 * that the values generated by two instances of {@code L64X1024Random} |
|
147 * will be (approximately) independent if have different values for `a`. |
|
148 * |
|
149 * The default (no-argument) constructor, in essence, uses |
|
150 * "defaultGen" to generate 18 new 64-bit values for the same |
|
151 * purpose. Multiple generators created in this way will certainly |
|
152 * differ in their `a` parameters. The defaultGen state must be accessed |
|
153 * in a thread-safe manner, so we use an AtomicLong to represent |
|
154 * this state. To bootstrap the defaultGen, we start off using a |
|
155 * seed based on current time unless the |
|
156 * java.util.secureRandomSeed property is set. This serves as a |
|
157 * slimmed-down (and insecure) variant of SecureRandom that also |
|
158 * avoids stalls that may occur when using /dev/random. |
|
159 * |
|
160 * File organization: First static fields, then instance |
|
161 * fields, then constructors, then instance methods. |
|
162 */ |
|
163 |
|
164 /* ---------------- static fields ---------------- */ |
|
165 |
|
166 /* |
|
167 * The length of the array x. |
|
168 */ |
|
169 |
|
170 private static final int N = 16; |
|
171 |
|
172 /** |
|
173 * The seed generator for default constructors. |
|
174 */ |
|
175 private static final AtomicLong defaultGen = new AtomicLong(RngSupport.initialSeed()); |
|
176 |
|
177 /* |
|
178 * The period of this generator, which is (2**1024 - 1) * 2**64. |
|
179 */ |
|
180 private static final BigInteger thePeriod = |
|
181 BigInteger.ONE.shiftLeft(N*64).subtract(BigInteger.ONE).shiftLeft(64); |
|
182 |
|
183 /* |
|
184 * Multiplier used in the LCG portion of the algorithm, taken from |
|
185 * Pierre L'Ecuyer, Tables of linear congruential generators of |
|
186 * different sizes and good lattice structure, <em>Mathematics of |
|
187 * Computation</em> 68, 225 (January 1999), pages 249-260, |
|
188 * Table 4 (first multiplier for size 2<sup>64</sup>). |
|
189 */ |
|
190 |
|
191 private static final long m = 2862933555777941757L; |
|
192 |
|
193 /* ---------------- instance fields ---------------- */ |
|
194 |
|
195 /** |
|
196 * The parameter that is used as an additive constant for the LCG. |
|
197 * Must be odd. |
|
198 */ |
|
199 private final long a; |
|
200 |
|
201 /** |
|
202 * The per-instance state: s for the LCG; the array x for the xorshift; |
|
203 * p is the rotating pointer into the array x. |
|
204 * At least one of the 16 elements of the array x must be nonzero. |
|
205 */ |
|
206 private long s; |
|
207 private final long[] x; |
|
208 private int p = N - 1; |
|
209 |
|
210 /* ---------------- constructors ---------------- */ |
|
211 |
|
212 /** |
|
213 * Basic constructor that initializes all fields from parameters. |
|
214 * It then adjusts the field values if necessary to ensure that |
|
215 * all constraints on the values of fields are met. |
|
216 * |
|
217 * @param a additive parameter for the LCG |
|
218 * @param s initial state for the LCG |
|
219 * @param x0 first word of the initial state for the xorshift generator |
|
220 * @param x1 second word of the initial state for the xorshift generator |
|
221 * @param x2 third word of the initial state for the xorshift generator |
|
222 * @param x3 fourth word of the initial state for the xorshift generator |
|
223 * @param x4 fifth word of the initial state for the xorshift generator |
|
224 * @param x5 sixth word of the initial state for the xorshift generator |
|
225 * @param x6 seventh word of the initial state for the xorshift generator |
|
226 * @param x7 eight word of the initial state for the xorshift generator |
|
227 * @param x8 ninth word of the initial state for the xorshift generator |
|
228 * @param x9 tenth word of the initial state for the xorshift generator |
|
229 * @param x10 eleventh word of the initial state for the xorshift generator |
|
230 * @param x11 twelfth word of the initial state for the xorshift generator |
|
231 * @param x12 thirteenth word of the initial state for the xorshift generator |
|
232 * @param x13 fourteenth word of the initial state for the xorshift generator |
|
233 * @param x14 fifteenth word of the initial state for the xorshift generator |
|
234 * @param x15 sixteenth word of the initial state for the xorshift generator |
|
235 */ |
|
236 public L64X1024Random(long a, long s, |
|
237 long x0, long x1, long x2, long x3, |
|
238 long x4, long x5, long x6, long x7, |
|
239 long x8, long x9, long x10, long x11, |
|
240 long x12, long x13, long x14, long x15) { |
|
241 // Force a to be odd. |
|
242 this.a = a | 1; |
|
243 this.s = s; |
|
244 this.x = new long[N]; |
|
245 this.x[0] = x0; |
|
246 this.x[1] = x1; |
|
247 this.x[2] = x2; |
|
248 this.x[3] = x3; |
|
249 this.x[4] = x4; |
|
250 this.x[5] = x5; |
|
251 this.x[6] = x6; |
|
252 this.x[7] = x7; |
|
253 this.x[8] = x8; |
|
254 this.x[9] = x9; |
|
255 this.x[10] = x10; |
|
256 this.x[11] = x11; |
|
257 this.x[12] = x12; |
|
258 this.x[13] = x13; |
|
259 this.x[14] = x14; |
|
260 this.x[15] = x15; |
|
261 // If x0, x1, ..., x15 are all zero (very unlikely), we must choose nonzero values. |
|
262 if ((x0 | x1 | x2 | x3 | x4 | x5 | x6 | x7 | x8 | x9 | x10 | x11 | x12 | x13 | x14 | x15) == 0) { |
|
263 for (int j = 0; j < N; j++) { |
|
264 this.x[j] = RngSupport.mixStafford13(s += RngSupport.GOLDEN_RATIO_64); |
|
265 } |
|
266 } |
|
267 } |
|
268 |
|
269 /** |
|
270 * Creates a new instance of {@code L64X1024Random} using the |
|
271 * specified {@code long} value as the initial seed. Instances of |
|
272 * {@code L64X1024Random} created with the same seed in the same |
|
273 * program execution generate identical sequences of values. |
|
274 * |
|
275 * @param seed the initial seed |
|
276 */ |
|
277 public L64X1024Random(long seed) { |
|
278 // Using a value with irregularly spaced 1-bits to xor the seed |
|
279 // argument tends to improve "pedestrian" seeds such as 0 or |
|
280 // other small integers. We may as well use SILVER_RATIO_64. |
|
281 // |
|
282 // The seed is hashed by mixMurmur64 to produce the `a` parameter. |
|
283 // The seed is hashed by mixStafford13 to produce the initial `x[0]`, |
|
284 // which will then be used to produce the first generated value. |
|
285 // The other x values are filled in as if by a SplitMix PRNG with |
|
286 // GOLDEN_RATIO_64 as the gamma value and Stafford13 as the mixer. |
|
287 this(RngSupport.mixMurmur64(seed ^= RngSupport.SILVER_RATIO_64), |
|
288 1, |
|
289 RngSupport.mixStafford13(seed), |
|
290 RngSupport.mixStafford13(seed += RngSupport.GOLDEN_RATIO_64), |
|
291 RngSupport.mixStafford13(seed += RngSupport.GOLDEN_RATIO_64), |
|
292 RngSupport.mixStafford13(seed += RngSupport.GOLDEN_RATIO_64), |
|
293 RngSupport.mixStafford13(seed += RngSupport.GOLDEN_RATIO_64), |
|
294 RngSupport.mixStafford13(seed += RngSupport.GOLDEN_RATIO_64), |
|
295 RngSupport.mixStafford13(seed += RngSupport.GOLDEN_RATIO_64), |
|
296 RngSupport.mixStafford13(seed += RngSupport.GOLDEN_RATIO_64), |
|
297 RngSupport.mixStafford13(seed += RngSupport.GOLDEN_RATIO_64), |
|
298 RngSupport.mixStafford13(seed += RngSupport.GOLDEN_RATIO_64), |
|
299 RngSupport.mixStafford13(seed += RngSupport.GOLDEN_RATIO_64), |
|
300 RngSupport.mixStafford13(seed += RngSupport.GOLDEN_RATIO_64), |
|
301 RngSupport.mixStafford13(seed += RngSupport.GOLDEN_RATIO_64), |
|
302 RngSupport.mixStafford13(seed += RngSupport.GOLDEN_RATIO_64), |
|
303 RngSupport.mixStafford13(seed += RngSupport.GOLDEN_RATIO_64), |
|
304 RngSupport.mixStafford13(seed + RngSupport.GOLDEN_RATIO_64)); |
|
305 } |
|
306 |
|
307 /** |
|
308 * Creates a new instance of {@code L64X1024Random} that is likely to |
|
309 * generate sequences of values that are statistically independent |
|
310 * of those of any other instances in the current program execution, |
|
311 * but may, and typically does, vary across program invocations. |
|
312 */ |
|
313 public L64X1024Random() { |
|
314 // Using GOLDEN_RATIO_64 here gives us a good Weyl sequence of values. |
|
315 this(defaultGen.getAndAdd(RngSupport.GOLDEN_RATIO_64)); |
|
316 } |
|
317 |
|
318 /** |
|
319 * Creates a new instance of {@code L64X1024Random} using the specified array of |
|
320 * initial seed bytes. Instances of {@code L64X1024Random} created with the same |
|
321 * seed array in the same program execution generate identical sequences of values. |
|
322 * |
|
323 * @param seed the initial seed |
|
324 */ |
|
325 public L64X1024Random(byte[] seed) { |
|
326 // Convert the seed to 18 long values, of which the last 16 are not all zero. |
|
327 long[] data = RngSupport.convertSeedBytesToLongs(seed, 18, 16); |
|
328 long a = data[0], s = data[1]; |
|
329 // Force a to be odd. |
|
330 this.a = a | 1; |
|
331 this.s = s; |
|
332 this.x = new long[N]; |
|
333 for (int j = 0; j < N; j++) { |
|
334 this.x[j] = data[2+j]; |
|
335 } |
|
336 } |
|
337 |
|
338 /* ---------------- public methods ---------------- */ |
|
339 |
|
340 /** |
|
341 * Constructs and returns a new instance of {@code L64X1024Random} |
|
342 * that shares no mutable state with this instance. |
|
343 * However, with very high probability, the set of values collectively |
|
344 * generated by the two objects has the same statistical properties as if |
|
345 * same the quantity of values were generated by a single thread using |
|
346 * a single {@code L64X1024Random} object. Either or both of the two |
|
347 * objects may be further split using the {@code split} method, |
|
348 * and the same expected statistical properties apply to the |
|
349 * entire set of generators constructed by such recursive splitting. |
|
350 * |
|
351 * @param source a {@code SplittableRng} instance to be used instead |
|
352 * of this one as a source of pseudorandom bits used to |
|
353 * initialize the state of the new ones. |
|
354 * @return a new instance of {@code L64X1024Random} |
|
355 */ |
|
356 public L64X1024Random split(SplittableRng source) { |
|
357 // Literally pick a new instance "at random". |
|
358 return new L64X1024Random(source.nextLong(), source.nextLong(), |
|
359 source.nextLong(), source.nextLong(), |
|
360 source.nextLong(), source.nextLong(), |
|
361 source.nextLong(), source.nextLong(), |
|
362 source.nextLong(), source.nextLong(), |
|
363 source.nextLong(), source.nextLong(), |
|
364 source.nextLong(), source.nextLong(), |
|
365 source.nextLong(), source.nextLong(), |
|
366 source.nextLong(), source.nextLong()); |
|
367 } |
|
368 |
|
369 /** |
|
370 * Returns a pseudorandom {@code long} value. |
|
371 * |
|
372 * @return a pseudorandom {@code long} value |
|
373 */ |
|
374 |
|
375 public long nextLong() { |
|
376 // First part of xoroshiro1024: fetch array data |
|
377 final int q = p; |
|
378 final long s0 = x[p = (p + 1) & (N - 1)]; |
|
379 long s15 = x[q]; |
|
380 |
|
381 final long z = s + s0; |
|
382 s = m * s + a; // LCG |
|
383 |
|
384 // Second part of xoroshiro1024: update array data |
|
385 s15 ^= s0; |
|
386 x[q] = Long.rotateLeft(s0, 25) ^ s15 ^ (s15 << 27); |
|
387 x[p] = Long.rotateLeft(s15, 36); |
|
388 |
|
389 return z; |
|
390 } |
|
391 |
|
392 public BigInteger period() { return thePeriod; } |
|
393 } |