Xorshift1024star

Xorshift* generators extend the basic xorshift generation mechanism by scrambling its output through multiplication by a constant factor, without touching the underlying state. The result is an extremely fast family of generators with very high-quality statistical properties.

The xorshift1024* generator offers a long (2 ^^ 1024 - 1) period, making it suitable for many massively parallel applications, while its speed and quality makes it useful as a good general purpose random number generator.

If 1024 bits of state are too much, then it is suggested to use the Xoroshiro128+ generator instead.

Disabled Default Constructor

A disabled default is present on this object. To use it, use one of the other constructors or a factory function.

Constructors

this
this(ulong s)
this(ulong[16] s)
this(R range)

Constructor (RNG instances can only be initialized with a specified seed).

Postblit

Copying this object is disabled.

A postblit is present on this object, but not explicitly documented in the source.

Members

Functions

dup
typeof(this) dup()

Provides a copy of this RNG instance with identical internal state.

front
ulong front()

Range primitives

jump
void jump()

Jump function, equivalent to 2 ^^ 512 calls to popFront(); can be used to generate 2 ^^ 512 non-overlapping subsequences for parallel computation.

popFront
void popFront()

Range primitives

seed
void seed(ulong s)
void seed(ulong[16] s)
void seed(R range)

(Re)seeds the generator.

Variables

empty
enum bool empty;

Range primitives

isUniformRandom
enum bool isUniformRandom;

Marks this range as a uniform random number generator

max
enum ulong max;

Largest generated value

min
enum ulong min;

Smallest generated value (0)

Examples

1 import std.array : array;
2 import std.random : isUniformRNG, randomCover, uniform;
3 import std.range : iota, take;
4 import dxorshift.xorshift1024star;
5 
6 // xorshift1024* generators must be initialized
7 // with a specified seed
8 auto gen = Xorshift1024star(123456);
9 
10 // verify it is indeed a uniform RNG as defined
11 // in the standard library, whether accessed
12 // directly or via a pointer
13 static assert(isUniformRNG!(typeof(gen)));
14 static assert(isUniformRNG!(typeof(&gen)));
15 
16 // since the postblit is disabled, we must
17 // pass a pointer to any functionality that
18 // would otherwise copy the RNG by value
19 assert((&gen).take(2).array == [1060672336872339994uL,
20                                 1269657541839679748uL]);
21 
22 // this means, of course, that we must guarantee
23 // the lifetime of the pointer is valid for the
24 // lifetime of any functionality that uses it
25 auto sample = iota(100).randomCover(&gen).array;
26 
27 // however, we can pass the RNG as-is to any
28 // functionality that takes it by ref and does
29 // not try to copy it by value
30 auto val = uniform!"[]"(0.0, 1.0, gen);
31 
32 // in circumstances where we really want to
33 // copy the RNG state, we can use `dup`
34 auto gen2 = gen.dup;
35 assert((&gen).take(6).array == (&gen2).take(6).array);

Meta

Credits

This code is ported from the public-domain reference implementation by Sebastiano Vigna, available online at http://xorshift.di.unimi.it/xorshift1024star.c

See also the research paper introducing the xorshift* family of generators: http://vigna.di.unimi.it/ftp/papers/xorshift.pdf