/* -*- mode: c++; tab-width: 4; indent-tabs-mode: nil; c-basic-offset: 4 -*- */

/*

 Copyright (C) 2010 Kakhkhor Abdijalilov

 This file is part of QuantLib, a free-software/open-source library

 for financial quantitative analysts and developers - http://quantlib.org/

 QuantLib is free software: you can redistribute it and/or modify it

 under the terms of the QuantLib license.  You should have received a

 copy of the license along with this program; if not, please email

 <quantlib-dev@lists.sf.net>. The license is also available online at

 <https://www.quantlib.org/license.shtml>.

 This program is distributed in the hope that it will be useful, but WITHOUT

 ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS

 FOR A PARTICULAR PURPOSE.  See the license for more details.

*/

/*! \file zigguratrng.hpp

    \brief Ziggurat random-number generator

*/

#ifndef quantlib_ziggurat_generator_hpp

#define quantlib_ziggurat_generator_hpp

#include <ql/math/randomnumbers/mt19937uniformrng.hpp>

#include <ql/math/randomnumbers/randomsequencegenerator.hpp>

namespace QuantLib {

    //! Ziggurat random-number generator

    /*! This generator returns standard normal variates using the

        Ziggurat method.  The underlying RNG is mt19937 (32 bit

        version). The algorithm is described in Marsaglia and Tsang

        (2000). "The Ziggurat Method for Generating Random

        Variables". Journal of Statistical Software 5 (8).  Note that

        step 2 from the above paper reuses the rightmost 8 bits of the

        random integer, which creates correlation between steps 1 and

        2.  This implementation was written from scratch, following

        Marsaglia and Tsang.  It avoids the correlation by using only

        the leftmost 24 bits of mt19937's output.

        Note that the GNU GSL implementation uses a different value

        for the right-most step. The GSL value is somewhat different

        from the one reported by Marsaglia and Tsang because GSL uses

        a different tail. This implementation uses the same right-most

        step as reported by Marsaglia and Tsang.  The generator was

        put through Marsaglia's Diehard battery of tests and didn't

        exibit any abnormal behavior.

    */

    class ZigguratRng {

      public:

        typedef Sample<Real> sample_type;

        explicit ZigguratRng(unsigned long seed = 0);

        sample_type next() const { return {nextGaussian(), 1.0}; }

      private:

        mutable MersenneTwisterUniformRng mt32_;

        Real nextGaussian() const;

    };

    // RNG traits for Ziggurat generator

    struct Ziggurat {

        // typedefs

        typedef ZigguratRng rng_type;

        typedef RandomSequenceGenerator<rng_type> rsg_type;

        // more traits

        enum { allowsErrorEstimate = 1 };

        // factory

        static rsg_type make_sequence_generator(Size dimension,

                                                BigNatural seed) {

            return rsg_type(dimension, seed);

        }

    };

}

#endif