A 64-bit Xorshift PRNG combined a Weyl Generator. Passes all BigCrush tests.

xorshift64.h

/*******************************************************************************
Copyright (C) 2009-2017 Reed A. Cartwright, PhD <[email protected]>
All rights reserved.

Redistribution and use in source and binary forms, with or without
modification, are permitted provided that the following conditions are met:

1. Redistributions of source code must retain the above copyright notice, this
   list of conditions and the following disclaimer.
2. Redistributions in binary form must reproduce the above copyright notice,
   this list of conditions and the following disclaimer in the documentation
   and/or other materials provided with the distribution.

THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND
ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR
ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
(INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

*******************************************************************************

A 64-bit Xorshift PRNG combined a Weyl Generator.
From ideas of Marsgalia, Brent, and others.

Passes all the BigCrush tests.

References:
http://www.jstatsoft.org/v08/i14/paper
http://wwwmaths.anu.edu.au/~brent/random.html
http://www.iro.umontreal.ca/~panneton/These.pdf

*******************************************************************************/

#pragma once
#ifndef XORSHIFT64_H
#define XORSHIFT64_H

#ifndef __STDC_CONSTANT_MACROS
#   define __STDC_CONSTANT_MACROS 1
#endif
#ifndef __STDC_LIMIT_MACROS
#   define __STDC_LIMIT_MACROS 1
#endif

#if defined(__unix__) || (defined(__APPLE__) && defined(__MACH__))
#   include <sys/param.h>
#endif

#if defined(__APPLE__) && defined(__MACH__)
#   include <machine/endian.h>
#elif defined(BSD)
#   include <sys/endian.h>
#else
#   include <endian.h>
#endif

#if defined(_WIN64) || defined(_WIN32)
#   include <process.h>
    inline int getpid() { return _getpid(); }
#else
#   include <unistd.h>
#endif

#include <cfloat>
#include <cstdint>
#include <algorithm>

#if __cpluscplus >= 201103L
#   include <chrono>
#   include <random>
#else
#   include <ctime>    
#endif

class xorshift64 {
public:
    explicit xorshift64(uint64_t seed1 = 0, uint64_t seed2 = 0) {    
        seed(seed1,seed2);
    }

    explicit xorshift64(std::pair<uint64_t,uint64_t> p) {
        seed(p.first,p.second);
    }

    void seed(uint64_t seed1 = 0, uint64_t seed2 = 0) {
        u = (seed1 == 0) ? UINT64_C(15191868757011070976) : seed1;
        w = (seed2 == 0) ? UINT64_C(0x61C8864680B583EB) : seed2;
        // Burn in the seed
        for(int i=0;i<256;++i)
            get_raw();
    }

    void seed(std::pair<uint64_t,uint64_t> p) {
        seed(p.first,p.second);
    }
    
    std::pair<uint64_t,uint64_t> get_state() const {
        return std::make_pair(u,w);
    }
    
    // Xorshift + Weyl Generator + some extra magic for low bits
    uint64_t get_raw() {
        u ^= (u << 5); u ^= (u >> 15); u ^= (u << 27);
        w += UINT64_C(0x61C8864680B583EB);
        return u+(w^(w>>27));
    }
    
    uint64_t get_uint64() {
        return get_raw();
    }
    
    uint32_t get_uint32() {
        return static_cast<uint32_t>(get_raw() >> 32);
    }

    std::pair<uint32_t,uint32_t> get_uint32_pair() {
        uint64_t u = get_raw();
        return std::make_pair(
            static_cast<uint32_t>(u >> 32),
            static_cast<uint32_t>(u)
        );
    }

    // Uniform [0,n) with 64-bits of precision
    // If n is a power of two, this only uses not-well-distributed low-bits
    uint64_t get_uint64(uint64_t n) {
        return get_uint64() % n;
    }
    
    // Uniform [0,n) exactly
    // This uses low bits which are not very random
    uint64_t get_uint64x(uint64_t n) {
        // Find a mask
        uint64_t v = n;
        v--;
        v |= v >> 1;
        v |= v >> 2;
        v |= v >> 4;
        v |= v >> 8;
        v |= v >> 16;
        v |= v >> 32;
        uint64_t u;
        do {
            u = get_uint64() & v;
        } while( u >= n);
        return u;
    }

    // Uniform [0,n) exactly
    // This uses high bits, but might not be as fast as get_uint64x
    uint64_t get_uint64h(uint64_t n) {
        // Find a shift
        uint64_t v = n-1;
        int r = 64;
        while(v >>= 1) {
            r--;
        }
        uint64_t u;
        do {
            u = get_uint64() >> r;
        } while( u >= n);
        return u;
    }

#if __FLOAT_WORD_ORDER == __BYTE_ORDER
    // Uniform [0,1)
    double get_double53() {
        union { uint64_t u; double d; } a;
        a.u = get_raw();
        a.u = (a.u >> 12) | UINT64_C(0x3FF0000000000000);
        double q = (a.u&2048) ? (1.0-(DBL_EPSILON/2.0)) : 1.0;
        return a.d-q;
    }

    // Uniform (0,1)
    double get_double52() {
        union { uint64_t u; double d; } a;
        a.u = get_raw();
        a.u = (a.u >> 12) | UINT64_C(0x3FF0000000000000);
        double q = (1.0-(DBL_EPSILON/2.0));
        return a.d-q;
    }
#else
    double get_double53() {
        uint64_t u = get_raw();
        int64_t n = static_cast<int64_t>(u >> 11);
        return n/9007199254740992.0;
    }
    double get_double52() {
        uint64_t u = get_raw();
        int64_t n = static_cast<int64_t>(u >> 11) | 0x1;
        return n/9007199254740992.0;
    }
#endif  
    
    // Uniform [0,max)
    uint64_t operator()() {
        return get_uint64();
    }
    
    // Uniform [0,n) with 64-bits of precision
    uint64_t operator()(uint64_t n) {
        return get_uint64(n);
    }

    static unsigned int create_random_seed();

private:
    uint64_t u,w;
};

// return a 31-bit seed that is not zero
inline unsigned int xorshift64::create_random_seed() {
    // start with some well mixed bits
    unsigned int v = 0x6ba658b3;    
#if __cpluscplus >= 201103L
    // if properly implemented, this will produce 32 random bits
    v += std::random_device{}();
#endif

    // use 5-decimal pid for some randomness after spreading over 32-bits
#if defined(_WIN64) || defined(_WIN32)
    unsigned int p = static_cast<unsigned int>(_getpid());
#else
    unsigned int p = static_cast<unsigned int>(getpid());
#endif
    v +=  p + (p << 15) + (p >> 3);
    v^=(v<<17); v^=(v>>13); v^=(v<<5);
    
    // use current time for more randomness

#if __cpluscplus >= 201103L
    v += static_cast<unsigned int>(std::chrono::high_resolution_clock::now().time_since_epoch().count());
#else
    v += static_cast<unsigned int>(time(NULL));
#endif
    v^=(v<<17); v^=(v>>13); v^=(v<<5);

    // Do four rounds of burn in.
    v^=(v<<17); v^=(v>>13); v^=(v<<5);
    v^=(v<<17); v^=(v>>13); v^=(v<<5);
    v^=(v<<17); v^=(v>>13); v^=(v<<5);
    v^=(v<<17); v^=(v>>13); v^=(v<<5);

    // return a 31-bit seed that is not zero
    v &= 0x7FFFFFFF; 
    return (v == 0) ? 0x6a27d958 : v;
}

#endif