Series of generated values pass the TestU01 BigCrush and PractRand test suites + * that measure independence and uniformity properties of random number generators. + * (Most recently validated with + * version 1.2.3 of TestU01 + * and version 0.90 of PractRand. + * Note that TestU01 BigCrush was used to test not only values produced by the {@code nextLong()} + * method but also the result of bit-reversing each value produced by {@code nextLong()}.) + * These tests validate only the methods for certain + * types and ranges, but similar properties are expected to hold, at + * least approximately, for others as well. + * + *
{@code L128X256MixRandom} is a specific member of the LXM family of algorithms + * for pseudorandom number generators. Every LXM generator consists of two + * subgenerators; one is an LCG (Linear Congruential Generator) and the other is + * an Xorshift generator. Each output of an LXM generator is the sum of one + * output from each subgenerator, possibly processed by a final mixing function + * (and {@code L128X256MixRandom} does use a mixing function). + * + *
The LCG subgenerator for {@code L128X256MixRandom} has an update step of the + * form {@code s = m * s + a}, where {@code s}, {@code m}, and {@code a} are all + * 128-bit integers; {@code s} is the mutable state, the multiplier {@code m} + * is fixed (the same for all instances of {@code L128X256MixRandom}}) and the addend + * {@code a} is a parameter (a final field of the instance). The parameter + * {@code a} is required to be odd (this allows the LCG to have the maximal + * period, namely 2128); therefore there are 2127 distinct choices + * of parameter. + * + *
The Xorshift subgenerator for {@code L128X256MixRandom} is the {@code xoshiro256} algorithm, + * version 1.0 (parameters 17, 45), without any final scrambler such as "+" or "**". + * Its state consists of four {@code long} fields {@code x0}, {@code x1}, {@code x2}, + * and {@code x3}, which can take on any values provided that they are not all zero. + * The period of this subgenerator is 2256-1. + * + *
The mixing function for {@code L128X256MixRandom} is the 64-bit MurmurHash3 finalizer. + * + *
Because the periods 2128 and 2256-1 of the two subgenerators + * are relatively prime, the period of any single {@code L128X256MixRandom} object + * (the length of the series of generated 64-bit values before it repeats) is the product + * of the periods of the subgenerators, that is, 2128(2256-1), + * which is just slightly smaller than 2384. Moreover, if two distinct + * {@code L128X256MixRandom} objects have different {@code a} parameters, then their + * cycles of produced values will be different. + * + *
The 64-bit values produced by the {@code nextLong()} method are exactly equidistributed. + * For any specific instance of {@code L128X256MixRandom}, over the course of its cycle each + * of the 264 possible {@code long} values will be produced 2256-1 times. + * The values produced by the {@code nextInt()}, {@code nextFloat()}, and {@code nextDouble()} + * methods are likewise exactly equidistributed. + * + *
In fact, the 64-bit values produced by the {@code nextLong()} method are exactly + * 2-equidistributed. For any specific instance of {@code L128X256MixRandom}, consider + * the (overlapping) length-2 subsequences of the cycle of 64-bit values produced by + * {@code nextLong()} (assuming no other methods are called that would affect the state). + * There are 2128(2256-1) such subsequences, and each subsequence, + * which consists of 2 64-bit values, can have one of 2128 values, and each + * such value occurs 2256-1 times. The values produced by the {@code nextInt()}, + * {@code nextFloat()}, and {@code nextDouble()} methods are likewise exactly 2-equidistributed. + * + *
Moreover, the 64-bit values produced by the {@code nextLong()} method are 4-equidistributed. + * To be precise: for any specific instance of {@code L128X256MixRandom}, consider + * the (overlapping) length-4 subsequences of the cycle of 64-bit values produced by + * {@code nextLong()} (assuming no other methods are called that would affect the state). + * There are 128(2256-1) such subsequences, and each subsequence, + * which consists of 4 64-bit values, can have one of 2256 values. Of those + * 2256 subsequence values, nearly all of them (2256-2128) + * occur 2128 times over the course of the entire cycle, and the other + * 2128 subsequence values occur only 2128-1 times. So the ratio + * of the probability of getting one of the less common subsequence values and the + * probability of getting one of the more common subsequence values is 1-2-128. + * (Note that the set of 2128 less-common subsequence values will differ from + * one instance of {@code L128X256MixRandom} to another, as a function of the additive + * parameter of the LCG.) The values produced by the {@code nextInt()}, {@code nextFloat()}, + * and {@code nextDouble()} methods are likewise 4-equidistributed. + * + *
Method {@link #split} constructs and returns a new {@code L128X256MixRandom} + * instance that shares no mutable state with the current instance. However, with + * very high probability, the values collectively generated by the two objects + * have the same statistical properties as if the same quantity of values were + * generated by a single thread using a single {@code L128X256MixRandom} object. + * This is because, with high probability, distinct {@code L128X256MixRandom} objects + * have distinct {@code a} parameters and therefore use distinct members of the + * algorithmic family; and even if their {@code a} parameters are the same, with + * very high probability they will traverse different parts of their common state + * cycle. + * + *
As with {@link java.util.SplittableRandom}, instances of + * {@code L128X256MixRandom} are not thread-safe. + * They are designed to be split, not shared, across threads. For + * example, a {@link java.util.concurrent.ForkJoinTask} fork/join-style + * computation using random numbers might include a construction + * of the form {@code new Subtask(someL128X256MixRandom.split()).fork()}. + * + *
This class provides additional methods for generating random + * streams, that employ the above techniques when used in + * {@code stream.parallel()} mode. + * + *
Instances of {@code L128X256MixRandom} are not cryptographically + * secure. Consider instead using {@link java.security.SecureRandom} + * in security-sensitive applications. Additionally, + * default-constructed instances do not use a cryptographically random + * seed unless the {@linkplain System#getProperty system property} + * {@code java.util.secureRandomSeed} is set to {@code true}. + * + * @author Guy Steele + * @since 1.9 + */ +public final class L128X256MixRandom extends AbstractSplittableRng { + + /* + * Implementation Overview. + * + * The 128-bit parameter `a` is represented as two long fields `ah` and `al`. + * The 128-bit state variable `s` is represented as two long fields `sh` and `sl`. + * + * The split operation uses the current generator to choose eight + * new 64-bit long values that are then used to initialize the + * parameters `ah` and `al` and the state variables `sh`, `sl`, + * `x0`, `x1`, `x2`, and `x3` for a newly constructed generator. + * + * With extremely high probability, no two generators so chosen + * will have the same `a` parameter, and testing has indicated + * that the values generated by two instances of {@code L128X256MixRandom} + * will be (approximately) independent if have different values for `a`. + * + * The default (no-argument) constructor, in essence, uses + * "defaultGen" to generate eight new 64-bit values for the same + * purpose. Multiple generators created in this way will certainly + * differ in their `a` parameters. The defaultGen state must be accessed + * in a thread-safe manner, so we use an AtomicLong to represent + * this state. To bootstrap the defaultGen, we start off using a + * seed based on current time unless the + * java.util.secureRandomSeed property is set. This serves as a + * slimmed-down (and insecure) variant of SecureRandom that also + * avoids stalls that may occur when using /dev/random. + * + * File organization: First static fields, then instance + * fields, then constructors, then instance methods. + */ + + /* ---------------- static fields ---------------- */ + + /** + * The seed generator for default constructors. + */ + private static final AtomicLong defaultGen = new AtomicLong(RngSupport.initialSeed()); + + /* + * The period of this generator, which is (2**256 - 1) * 2**128. + */ + private static final BigInteger thePeriod = + BigInteger.ONE.shiftLeft(256).subtract(BigInteger.ONE).shiftLeft(128); + + /* + * The multiplier used in the LCG portion of the algorithm is 2**64 + m; + * where m is taken from + * Pierre L'Ecuyer, Tables of linear congruential generators of + * different sizes and good lattice structure, Mathematics of + * Computation 68, 225 (January 1999), pages 249-260, + * Table 4 (first multiplier for size 264). + * + * This is almost certainly not the best possible 128-bit multiplier + * for an LCG, but it is sufficient for our purposes here; because + * is is larger than 2**64, the 64-bit values produced by nextLong() + * are exactly 2-equidistributed, and the fact that it is of the + * form (2**64 + m) simplifies the code, given that we have only + * 64-bit arithmetic to work with. + */ + + private static final long m = 2862933555777941757L; + + /* ---------------- instance fields ---------------- */ + + /** + * The parameter that is used as an additive constant for the LCG. + * Must be odd. + */ + private final long ah, al; + + /** + * The per-instance state: sh and sl for the LCG; x0, x1, x2, and x3 for the xorshift. + * At least one of the four fields x0, x1, x2, and x3 must be nonzero. + */ + private long sh, sl, x0, x1, x2, x3; + + /* ---------------- constructors ---------------- */ + + /** + * Basic constructor that initializes all fields from parameters. + * It then adjusts the field values if necessary to ensure that + * all constraints on the values of fields are met. + * + * @param ah high half of the additive parameter for the LCG + * @param al low half of the additive parameter for the LCG + * @param sh high half of the initial state for the LCG + * @param sl low half of the initial state for the LCG + * @param x0 first word of the initial state for the xorshift generator + * @param x1 second word of the initial state for the xorshift generator + * @param x2 third word of the initial state for the xorshift generator + * @param x3 fourth word of the initial state for the xorshift generator + */ + public L128X256MixRandom(long ah, long al, long sh, long sl, long x0, long x1, long x2, long x3) { + // Force a to be odd. + this.ah = ah; + this.al = al | 1; + this.sh = sh; + this.sl = sl; + this.x0 = x0; + this.x1 = x1; + this.x2 = x2; + this.x3 = x3; + // If x0, x1, x2, and x3 are all zero, we must choose nonzero values. + if ((x0 | x1 | x2 | x3) == 0) { + // At least three of the four values generated here will be nonzero. + this.x0 = RngSupport.mixStafford13(sh += RngSupport.GOLDEN_RATIO_64); + this.x1 = RngSupport.mixStafford13(sh += RngSupport.GOLDEN_RATIO_64); + this.x2 = RngSupport.mixStafford13(sh += RngSupport.GOLDEN_RATIO_64); + this.x3 = RngSupport.mixStafford13(sh + RngSupport.GOLDEN_RATIO_64); + } + } + + /** + * Creates a new instance of {@code L128X256MixRandom} using the + * specified {@code long} value as the initial seed. Instances of + * {@code L128X256MixRandom} created with the same seed in the same + * program generate identical sequences of values. + * + * @param seed the initial seed + */ + public L128X256MixRandom(long seed) { + // Using a value with irregularly spaced 1-bits to xor the seed + // argument tends to improve "pedestrian" seeds such as 0 or + // other small integers. We may as well use SILVER_RATIO_64. + // + // The seed is hashed by mixMurmur64 to produce the `a` parameter. + // The seed is hashed by mixStafford13 to produce the initial `x0`, + // which will then be used to produce the first generated value. + // The other x values are filled in as if by a SplitMix PRNG with + // GOLDEN_RATIO_64 as the gamma value and Stafford13 as the mixer. + this(RngSupport.mixMurmur64(seed ^= RngSupport.SILVER_RATIO_64), + RngSupport.mixMurmur64(seed += RngSupport.GOLDEN_RATIO_64), + 0, + 1, + RngSupport.mixStafford13(seed), + RngSupport.mixStafford13(seed += RngSupport.GOLDEN_RATIO_64), + RngSupport.mixStafford13(seed += RngSupport.GOLDEN_RATIO_64), + RngSupport.mixStafford13(seed + RngSupport.GOLDEN_RATIO_64)); + } + + /** + * Creates a new instance of {@code L128X256MixRandom} that is likely to + * generate sequences of values that are statistically independent + * of those of any other instances in the current program execution, + * but may, and typically does, vary across program invocations. + */ + public L128X256MixRandom() { + // Using GOLDEN_RATIO_64 here gives us a good Weyl sequence of values. + this(defaultGen.getAndAdd(RngSupport.GOLDEN_RATIO_64)); + } + + /** + * Creates a new instance of {@code L128X256MixRandom} using the specified array of + * initial seed bytes. Instances of {@code L128X256MixRandom} created with the same + * seed array in the same program execution generate identical sequences of values. + * + * @param seed the initial seed + */ + public L128X256MixRandom(byte[] seed) { + // Convert the seed to 6 long values, of which the last 4 are not all zero. + long[] data = RngSupport.convertSeedBytesToLongs(seed, 6, 4); + long ah = data[0], al = data[1], sh = data[2], sl = data[3], x0 = data[4], x1 = data[5], x2 = data[6], x3 = data[7]; + // Force a to be odd. + this.ah = ah; + this.al = al | 1; + this.sh = sh; + this.sl = sl; + this.x0 = x0; + this.x1 = x1; + this.x2 = x2; + this.x3 = x3; + } + + /* ---------------- public methods ---------------- */ + + /** + * Constructs and returns a new instance of {@code L128X256MixRandom} + * that shares no mutable state with this instance. + * However, with very high probability, the set of values collectively + * generated by the two objects has the same statistical properties as if + * same the quantity of values were generated by a single thread using + * a single {@code L128X256MixRandom} object. Either or both of the two + * objects may be further split using the {@code split} method, + * and the same expected statistical properties apply to the + * entire set of generators constructed by such recursive splitting. + * + * @param source a {@code SplittableRng} instance to be used instead + * of this one as a source of pseudorandom bits used to + * initialize the state of the new ones. + * @return a new instance of {@code L128X256MixRandom} + */ + public L128X256MixRandom split(SplittableRng source) { + // Literally pick a new instance "at random". + return new L128X256MixRandom(source.nextLong(), source.nextLong(), + source.nextLong(), source.nextLong(), + source.nextLong(), source.nextLong(), + source.nextLong(), source.nextLong()); + } + + /** + * Returns a pseudorandom {@code long} value. + * + * @return a pseudorandom {@code long} value + */ + + public long nextLong() { + final long z = sh + x0; + // The LCG: in effect, s = ((1LL << 64) + m) * s + a, if only we had 128-bit arithmetic. + final long u = m * sl; + sh = (m * sh) + Math.multiplyHigh(m, sl) + sl + ah; + sl = u + al; + if (Long.compareUnsigned(sl, u) < 0) ++sh; // Handle the carry propagation from low half to high half. + long q0 = x0, q1 = x1, q2 = x2, q3 = x3; + { long t = q1 << 17; q2 ^= q0; q3 ^= q1; q1 ^= q2; q0 ^= q3; q2 ^= t; q3 = Long.rotateLeft(q3, 45); } // xoshiro256 1.0 + x0 = q0; x1 = q1; x2 = q2; x3 = q3; + return RngSupport.mixLea64(z); // mixing function + } + + public BigInteger period() { return thePeriod; } +}