//---------------------------------------------------------------------------// // Copyright (c) 2013-2014 Kyle Lutz <kyle.r.lutz@gmail.com> // // Distributed under the Boost Software License, Version 1.0 // See accompanying file LICENSE_1_0.txt or copy at // http://www.boost.org/LICENSE_1_0.txt // // See http://boostorg.github.com/compute for more information. //---------------------------------------------------------------------------// #ifndef BOOST_COMPUTE_RANDOM_NORMAL_DISTRIBUTION_HPP #define BOOST_COMPUTE_RANDOM_NORMAL_DISTRIBUTION_HPP #include <limits> #include <boost/assert.hpp> #include <boost/type_traits.hpp> #include <boost/compute/command_queue.hpp> #include <boost/compute/function.hpp> #include <boost/compute/types/fundamental.hpp> #include <boost/compute/type_traits/make_vector_type.hpp> namespace boost { namespace compute { /// \class normal_distribution /// \brief Produces random, normally-distributed floating-point numbers. /// /// The following example shows how to setup a normal distribution to /// produce random \c float values centered at \c 5: /// /// \snippet test/test_normal_distribution.cpp generate /// /// \see default_random_engine, uniform_real_distribution template<class RealType = float> class normal_distribution { public: typedef RealType result_type; /// Creates a new normal distribution producing numbers with the given /// \p mean and \p stddev. normal_distribution(RealType mean = 0.f, RealType stddev = 1.f) : m_mean(mean), m_stddev(stddev) { } /// Destroys the normal distribution object. ~normal_distribution() { } /// Returns the mean value of the distribution. result_type mean() const { return m_mean; } /// Returns the standard-deviation of the distribution. result_type stddev() const { return m_stddev; } /// Returns the minimum value of the distribution. result_type min BOOST_PREVENT_MACRO_SUBSTITUTION () const { return -std::numeric_limits<RealType>::infinity(); } /// Returns the maximum value of the distribution. result_type max BOOST_PREVENT_MACRO_SUBSTITUTION () const { return std::numeric_limits<RealType>::infinity(); } /// Generates normally-distributed floating-point numbers and stores /// them to the range [\p first, \p last). template<class OutputIterator, class Generator> void generate(OutputIterator first, OutputIterator last, Generator &generator, command_queue &queue) { typedef typename make_vector_type<RealType, 2>::type RealType2; size_t count = detail::iterator_range_size(first, last); vector<uint_> tmp(count, queue.get_context()); generator.generate(tmp.begin(), tmp.end(), queue); BOOST_COMPUTE_FUNCTION(RealType2, box_muller, (const uint2_ x), { const RealType one = 1; const RealType two = 2; // Use nextafter to push values down into [0,1) range; without this, floating point rounding can // lead to have x1 = 1, but that would lead to taking the log of 0, which would result in negative // infinities; by pushing the values off 1 towards 0, we ensure this won't happen. const RealType x1 = nextafter(x.x / (RealType) UINT_MAX, (RealType) 0); const RealType x2 = x.y / (RealType) UINT_MAX; const RealType rho = sqrt(-two * log(one-x1)); const RealType z1 = rho * cos(two * M_PI_F * x2); const RealType z2 = rho * sin(two * M_PI_F * x2); return (RealType2)(MEAN, MEAN) + (RealType2)(z1, z2) * (RealType2)(STDDEV, STDDEV); }); box_muller.define("MEAN", boost::lexical_cast<std::string>(m_mean)); box_muller.define("STDDEV", boost::lexical_cast<std::string>(m_stddev)); box_muller.define("RealType", type_name<RealType>()); box_muller.define("RealType2", type_name<RealType2>()); transform( make_buffer_iterator<uint2_>(tmp.get_buffer(), 0), make_buffer_iterator<uint2_>(tmp.get_buffer(), count / 2), make_buffer_iterator<RealType2>(first.get_buffer(), 0), box_muller, queue ); } private: RealType m_mean; RealType m_stddev; BOOST_STATIC_ASSERT_MSG( boost::is_floating_point<RealType>::value, "Template argument must be a floating point type" ); }; } // end compute namespace } // end boost namespace #endif // BOOST_COMPUTE_RANDOM_NORMAL_DISTRIBUTION_HPP