3136 lines
		
	
	
		
			135 KiB
		
	
	
	
		
			Plaintext
		
	
	
	
	
	
			
		
		
	
	
			3136 lines
		
	
	
		
			135 KiB
		
	
	
	
		
			Plaintext
		
	
	
	
	
	
| ///////////////////////////////////////////////////////////////////////////////
 | |
| // Copyright Christopher Kormanyos 2002 - 2013.
 | |
| // Copyright 2011 -2013 John Maddock. 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)
 | |
| //
 | |
| // This work is based on an earlier work:
 | |
| // "Algorithm 910: A Portable C++ Multiple-Precision System for Special-Function Calculations",
 | |
| // in ACM TOMS, {VOL 37, ISSUE 4, (February 2011)} (C) ACM, 2011. http://doi.acm.org/10.1145/1916461.1916469
 | |
| //
 | |
| // Note that there are no "noexcept" specifications on the functions in this file: there are too many
 | |
| // calls to lexical_cast (and similar) to easily analyse the code for correctness. So until compilers
 | |
| // can detect noexcept misuse at compile time, the only realistic option is to simply not use it here.
 | |
| //
 | |
| 
 | |
| #ifndef BOOST_MP_CPP_DEC_FLOAT_BACKEND_HPP
 | |
| #define BOOST_MP_CPP_DEC_FLOAT_BACKEND_HPP
 | |
| 
 | |
| #include <boost/config.hpp>
 | |
| #include <boost/cstdint.hpp>
 | |
| #include <limits>
 | |
| #ifndef BOOST_NO_CXX11_HDR_ARRAY
 | |
| #include <array>
 | |
| #else
 | |
| #include <boost/array.hpp>
 | |
| #endif
 | |
| #include <boost/cstdint.hpp>
 | |
| #include <boost/functional/hash_fwd.hpp>
 | |
| #include <boost/multiprecision/number.hpp>
 | |
| #include <boost/multiprecision/detail/big_lanczos.hpp>
 | |
| #include <boost/multiprecision/detail/dynamic_array.hpp>
 | |
| 
 | |
| //
 | |
| // Headers required for Boost.Math integration:
 | |
| //
 | |
| #include <boost/math/policies/policy.hpp>
 | |
| //
 | |
| // Some includes we need from Boost.Math, since we rely on that library to provide these functions:
 | |
| //
 | |
| #include <boost/math/special_functions/asinh.hpp>
 | |
| #include <boost/math/special_functions/acosh.hpp>
 | |
| #include <boost/math/special_functions/atanh.hpp>
 | |
| #include <boost/math/special_functions/cbrt.hpp>
 | |
| #include <boost/math/special_functions/expm1.hpp>
 | |
| #include <boost/math/special_functions/gamma.hpp>
 | |
| 
 | |
| #ifdef BOOST_MSVC
 | |
| #pragma warning(push)
 | |
| #pragma warning(disable:6326)  // comparison of two constants
 | |
| #endif
 | |
| 
 | |
| namespace boost{
 | |
| namespace multiprecision{
 | |
| namespace backends{
 | |
| 
 | |
| template <unsigned Digits10, class ExponentType = boost::int32_t, class Allocator = void>
 | |
| class cpp_dec_float;
 | |
| 
 | |
| } // namespace
 | |
| 
 | |
| template <unsigned Digits10, class ExponentType, class Allocator>
 | |
| struct number_category<backends::cpp_dec_float<Digits10, ExponentType, Allocator> > : public mpl::int_<number_kind_floating_point>{};
 | |
| 
 | |
| namespace backends{
 | |
| 
 | |
| template <unsigned Digits10, class ExponentType, class Allocator>
 | |
| class cpp_dec_float
 | |
| {
 | |
| private:
 | |
|    static const boost::int32_t cpp_dec_float_digits10_setting = Digits10;
 | |
| 
 | |
|    // We need at least 16-bits in the exponent type to do anything sensible:
 | |
|    BOOST_STATIC_ASSERT_MSG(boost::is_signed<ExponentType>::value, "ExponentType must be a signed built in integer type.");
 | |
|    BOOST_STATIC_ASSERT_MSG(sizeof(ExponentType) > 1, "ExponentType is too small.");
 | |
| 
 | |
| public:
 | |
|    typedef mpl::list<boost::long_long_type> signed_types;
 | |
|    typedef mpl::list<boost::ulong_long_type> unsigned_types;
 | |
|    typedef mpl::list<long double> float_types;
 | |
|    typedef ExponentType exponent_type;
 | |
| 
 | |
|    static const boost::int32_t cpp_dec_float_radix = 10L;
 | |
|    static const boost::int32_t cpp_dec_float_digits10_limit_lo = 9L;
 | |
|    static const boost::int32_t cpp_dec_float_digits10_limit_hi = boost::integer_traits<boost::int32_t>::const_max - 100;
 | |
|    static const boost::int32_t cpp_dec_float_digits10 = ((cpp_dec_float_digits10_setting < cpp_dec_float_digits10_limit_lo) ? cpp_dec_float_digits10_limit_lo : ((cpp_dec_float_digits10_setting > cpp_dec_float_digits10_limit_hi) ? cpp_dec_float_digits10_limit_hi : cpp_dec_float_digits10_setting));
 | |
|    static const ExponentType cpp_dec_float_max_exp10 = (static_cast<ExponentType>(1) << (std::numeric_limits<ExponentType>::digits - 5));
 | |
|    static const ExponentType cpp_dec_float_min_exp10 = -cpp_dec_float_max_exp10;
 | |
|    static const ExponentType cpp_dec_float_max_exp = cpp_dec_float_max_exp10;
 | |
|    static const ExponentType cpp_dec_float_min_exp = cpp_dec_float_min_exp10;
 | |
| 
 | |
|    BOOST_STATIC_ASSERT((cpp_dec_float<Digits10, ExponentType, Allocator>::cpp_dec_float_max_exp10 == -cpp_dec_float<Digits10, ExponentType, Allocator>::cpp_dec_float_min_exp10));
 | |
| 
 | |
| private:
 | |
|    static const boost::int32_t cpp_dec_float_elem_digits10 = 8L;
 | |
|    static const boost::int32_t cpp_dec_float_elem_mask = 100000000L;
 | |
| 
 | |
|    BOOST_STATIC_ASSERT(0 == cpp_dec_float_max_exp10 % cpp_dec_float_elem_digits10);
 | |
| 
 | |
|    // There are three guard limbs.
 | |
|    // 1) The first limb has 'play' from 1...8 decimal digits.
 | |
|    // 2) The last limb also has 'play' from 1...8 decimal digits.
 | |
|    // 3) One limb can get lost when justifying after multiply,
 | |
|    // as only half of the triangle is multiplied and a carry
 | |
|    // from below is missing.
 | |
|    static const boost::int32_t cpp_dec_float_elem_number_request = static_cast<boost::int32_t>((cpp_dec_float_digits10 / cpp_dec_float_elem_digits10) + (((cpp_dec_float_digits10 % cpp_dec_float_elem_digits10) != 0) ? 1 : 0));
 | |
| 
 | |
|    // The number of elements needed (with a minimum of two) plus three added guard limbs.
 | |
|    static const boost::int32_t cpp_dec_float_elem_number = static_cast<boost::int32_t>(((cpp_dec_float_elem_number_request < 2L) ? 2L : cpp_dec_float_elem_number_request) + 3L);
 | |
| 
 | |
| public:
 | |
|    static const boost::int32_t cpp_dec_float_total_digits10 = static_cast<boost::int32_t>(cpp_dec_float_elem_number * cpp_dec_float_elem_digits10);
 | |
| 
 | |
| private:
 | |
| 
 | |
|    typedef enum enum_fpclass_type
 | |
|    {
 | |
|       cpp_dec_float_finite,
 | |
|       cpp_dec_float_inf,
 | |
|       cpp_dec_float_NaN
 | |
|    }
 | |
|    fpclass_type;
 | |
| 
 | |
| #ifndef BOOST_NO_CXX11_HDR_ARRAY
 | |
|    typedef typename mpl::if_<is_void<Allocator>,
 | |
|       std::array<boost::uint32_t, cpp_dec_float_elem_number>,
 | |
|       detail::dynamic_array<boost::uint32_t, cpp_dec_float_elem_number, Allocator>
 | |
|       >::type array_type;
 | |
| #else
 | |
|    typedef typename mpl::if_<is_void<Allocator>,
 | |
|       boost::array<boost::uint32_t, cpp_dec_float_elem_number>,
 | |
|       detail::dynamic_array<boost::uint32_t, cpp_dec_float_elem_number, Allocator>
 | |
|       >::type array_type;
 | |
| #endif
 | |
| 
 | |
|    array_type data;
 | |
|    ExponentType exp;
 | |
|    bool neg;
 | |
|    fpclass_type fpclass;
 | |
|    boost::int32_t prec_elem;
 | |
| 
 | |
|    //
 | |
|    // Special values constructor:
 | |
|    //
 | |
|    cpp_dec_float(fpclass_type c) :
 | |
|       data(),
 | |
|       exp (static_cast<ExponentType>(0)),
 | |
|       neg (false),
 | |
|       fpclass (c),
 | |
|       prec_elem(cpp_dec_float_elem_number) { }
 | |
| 
 | |
|       //
 | |
|       // Static data initializer:
 | |
|       //
 | |
|       struct initializer
 | |
|       {
 | |
|          initializer()
 | |
|          {
 | |
|              cpp_dec_float<Digits10, ExponentType, Allocator>::nan();
 | |
|              cpp_dec_float<Digits10, ExponentType, Allocator>::inf();
 | |
|             (cpp_dec_float<Digits10, ExponentType, Allocator>::min)();
 | |
|             (cpp_dec_float<Digits10, ExponentType, Allocator>::max)();
 | |
|              cpp_dec_float<Digits10, ExponentType, Allocator>::zero();
 | |
|              cpp_dec_float<Digits10, ExponentType, Allocator>::one();
 | |
|              cpp_dec_float<Digits10, ExponentType, Allocator>::two();
 | |
|              cpp_dec_float<Digits10, ExponentType, Allocator>::half();
 | |
|              cpp_dec_float<Digits10, ExponentType, Allocator>::double_min();
 | |
|              cpp_dec_float<Digits10, ExponentType, Allocator>::double_max();
 | |
|              cpp_dec_float<Digits10, ExponentType, Allocator>::long_double_max();
 | |
|              cpp_dec_float<Digits10, ExponentType, Allocator>::long_double_min();
 | |
|              cpp_dec_float<Digits10, ExponentType, Allocator>::long_long_max();
 | |
|              cpp_dec_float<Digits10, ExponentType, Allocator>::long_long_min();
 | |
|              cpp_dec_float<Digits10, ExponentType, Allocator>::ulong_long_max();
 | |
|              cpp_dec_float<Digits10, ExponentType, Allocator>::eps();
 | |
|              cpp_dec_float<Digits10, ExponentType, Allocator>::pow2(0);
 | |
|          }
 | |
|          void do_nothing(){}
 | |
|       };
 | |
| 
 | |
|       static initializer init;
 | |
| 
 | |
| public:
 | |
|    // Constructors
 | |
|    cpp_dec_float() BOOST_MP_NOEXCEPT_IF(noexcept(array_type())) :
 | |
|       data(),
 | |
|       exp (static_cast<ExponentType>(0)),
 | |
|       neg (false),
 | |
|       fpclass (cpp_dec_float_finite),
 | |
|       prec_elem(cpp_dec_float_elem_number) { }
 | |
| 
 | |
|    cpp_dec_float(const char* s) :
 | |
|       data(),
 | |
|       exp (static_cast<ExponentType>(0)),
 | |
|       neg (false),
 | |
|       fpclass (cpp_dec_float_finite),
 | |
|       prec_elem(cpp_dec_float_elem_number)
 | |
|       {
 | |
|          *this = s;
 | |
|       }
 | |
| 
 | |
|    template<class I>
 | |
|    cpp_dec_float(I i, typename enable_if<is_unsigned<I> >::type* = 0) :
 | |
|       data(),
 | |
|       exp (static_cast<ExponentType>(0)),
 | |
|       neg (false),
 | |
|       fpclass (cpp_dec_float_finite),
 | |
|       prec_elem(cpp_dec_float_elem_number)
 | |
|       {
 | |
|          from_unsigned_long_long(i);
 | |
|       }
 | |
| 
 | |
|    template <class I>
 | |
|    cpp_dec_float(I i, typename enable_if<is_signed<I> >::type* = 0) :
 | |
|       data(),
 | |
|       exp (static_cast<ExponentType>(0)),
 | |
|       neg (false),
 | |
|       fpclass (cpp_dec_float_finite),
 | |
|       prec_elem(cpp_dec_float_elem_number)
 | |
|       {
 | |
|          if(i < 0)
 | |
|          {
 | |
|             from_unsigned_long_long(boost::multiprecision::detail::unsigned_abs(i));
 | |
|             negate();
 | |
|          }
 | |
|          else
 | |
|             from_unsigned_long_long(i);
 | |
|       }
 | |
| 
 | |
|    cpp_dec_float(const cpp_dec_float& f) BOOST_MP_NOEXCEPT_IF(noexcept(array_type(std::declval<const array_type&>()))) :
 | |
|       data (f.data),
 | |
|       exp (f.exp),
 | |
|       neg (f.neg),
 | |
|       fpclass (f.fpclass),
 | |
|       prec_elem(f.prec_elem) { }
 | |
| 
 | |
|    template <unsigned D, class ET, class A>
 | |
|    cpp_dec_float(const cpp_dec_float<D, ET, A>& f, typename enable_if_c<D <= Digits10>::type* = 0) :
 | |
|       data(),
 | |
|       exp (f.exp),
 | |
|       neg (f.neg),
 | |
|       fpclass (static_cast<fpclass_type>(static_cast<int>(f.fpclass))),
 | |
|       prec_elem(cpp_dec_float_elem_number)
 | |
|    {
 | |
|       std::copy(f.data.begin(), f.data.begin() + f.prec_elem, data.begin());
 | |
|    }
 | |
|    template <unsigned D, class ET, class A>
 | |
|    explicit cpp_dec_float(const cpp_dec_float<D, ET, A>& f, typename disable_if_c<D <= Digits10>::type* = 0) :
 | |
|       data(),
 | |
|       exp (f.exp),
 | |
|       neg (f.neg),
 | |
|       fpclass (static_cast<fpclass_type>(static_cast<int>(f.fpclass))),
 | |
|       prec_elem(cpp_dec_float_elem_number)
 | |
|    {
 | |
|       // TODO: this doesn't round!
 | |
|       std::copy(f.data.begin(), f.data.begin() + prec_elem, data.begin());
 | |
|    }
 | |
| 
 | |
|    template <class F>
 | |
|    cpp_dec_float(const F val, typename enable_if<is_floating_point<F> >::type* = 0) :
 | |
|       data(),
 | |
|       exp (static_cast<ExponentType>(0)),
 | |
|       neg (false),
 | |
|       fpclass (cpp_dec_float_finite),
 | |
|       prec_elem(cpp_dec_float_elem_number)
 | |
|    {
 | |
|       *this = val;
 | |
|    }
 | |
| 
 | |
|    cpp_dec_float(const double mantissa, const ExponentType exponent);
 | |
| 
 | |
|    std::size_t hash()const
 | |
|    {
 | |
|       std::size_t result = 0;
 | |
|       for(int i = 0; i < prec_elem; ++i)
 | |
|          boost::hash_combine(result, data[i]);
 | |
|       boost::hash_combine(result, exp);
 | |
|       boost::hash_combine(result, neg);
 | |
|       boost::hash_combine(result, fpclass);
 | |
|       return result;
 | |
|    }
 | |
| 
 | |
|    // Specific special values.
 | |
|    static const cpp_dec_float& nan()
 | |
|    {
 | |
|       static const cpp_dec_float val(cpp_dec_float_NaN);
 | |
|       init.do_nothing();
 | |
|       return val;
 | |
|    }
 | |
| 
 | |
|    static const cpp_dec_float& inf()
 | |
|    {
 | |
|       static const cpp_dec_float val(cpp_dec_float_inf);
 | |
|       init.do_nothing();
 | |
|       return val;
 | |
|    }
 | |
| 
 | |
|    static const cpp_dec_float& (max)()
 | |
|    {
 | |
|       init.do_nothing();
 | |
|       static cpp_dec_float val_max = std::string("1.0e" + boost::lexical_cast<std::string>(cpp_dec_float_max_exp10)).c_str();
 | |
|       return val_max;
 | |
|    }
 | |
| 
 | |
|    static const cpp_dec_float& (min)()
 | |
|    {
 | |
|       init.do_nothing();
 | |
|       static cpp_dec_float val_min = std::string("1.0e" + boost::lexical_cast<std::string>(cpp_dec_float_min_exp10)).c_str();
 | |
|       return val_min;
 | |
|    }
 | |
| 
 | |
|    static const cpp_dec_float& zero()
 | |
|    {
 | |
|       init.do_nothing();
 | |
|       static cpp_dec_float val(static_cast<boost::ulong_long_type>(0u));
 | |
|       return val;
 | |
|    }
 | |
| 
 | |
|    static const cpp_dec_float& one()
 | |
|    {
 | |
|       init.do_nothing();
 | |
|       static cpp_dec_float val(static_cast<boost::ulong_long_type>(1u));
 | |
|       return val;
 | |
|    }
 | |
| 
 | |
|    static const cpp_dec_float& two()
 | |
|    {
 | |
|       init.do_nothing();
 | |
|       static cpp_dec_float val(static_cast<boost::ulong_long_type>(2u));
 | |
|       return val;
 | |
|    }
 | |
| 
 | |
|    static const cpp_dec_float& half()
 | |
|    {
 | |
|       init.do_nothing();
 | |
|       static cpp_dec_float val(0.5L);
 | |
|       return val;
 | |
|    }
 | |
| 
 | |
|    static const cpp_dec_float& double_min()
 | |
|    {
 | |
|       init.do_nothing();
 | |
|       static cpp_dec_float val(static_cast<long double>((std::numeric_limits<double>::min)()));
 | |
|       return val;
 | |
|    }
 | |
| 
 | |
|    static const cpp_dec_float& double_max()
 | |
|    {
 | |
|       init.do_nothing();
 | |
|       static cpp_dec_float val(static_cast<long double>((std::numeric_limits<double>::max)()));
 | |
|       return val;
 | |
|    }
 | |
| 
 | |
|    static const cpp_dec_float& long_double_min()
 | |
|    {
 | |
|       init.do_nothing();
 | |
| #ifdef BOOST_MATH_NO_LONG_DOUBLE_MATH_FUNCTIONS
 | |
|       static cpp_dec_float val(static_cast<long double>((std::numeric_limits<double>::min)()));
 | |
| #else
 | |
|       static cpp_dec_float val((std::numeric_limits<long double>::min)());
 | |
| #endif
 | |
|       return val;
 | |
|    }
 | |
| 
 | |
|    static const cpp_dec_float& long_double_max()
 | |
|    {
 | |
|       init.do_nothing();
 | |
| #ifdef BOOST_MATH_NO_LONG_DOUBLE_MATH_FUNCTIONS
 | |
|       static cpp_dec_float val(static_cast<long double>((std::numeric_limits<double>::max)()));
 | |
| #else
 | |
|       static cpp_dec_float val((std::numeric_limits<long double>::max)());
 | |
| #endif
 | |
|       return val;
 | |
|    }
 | |
| 
 | |
|    static const cpp_dec_float& long_long_max()
 | |
|    {
 | |
|       init.do_nothing();
 | |
|       static cpp_dec_float val((std::numeric_limits<boost::long_long_type>::max)());
 | |
|       return val;
 | |
|    }
 | |
| 
 | |
|    static const cpp_dec_float& long_long_min()
 | |
|    {
 | |
|       init.do_nothing();
 | |
|       static cpp_dec_float val((std::numeric_limits<boost::long_long_type>::min)());
 | |
|       return val;
 | |
|    }
 | |
| 
 | |
|    static const cpp_dec_float& ulong_long_max()
 | |
|    {
 | |
|       init.do_nothing();
 | |
|       static cpp_dec_float val((std::numeric_limits<boost::ulong_long_type>::max)());
 | |
|       return val;
 | |
|    }
 | |
| 
 | |
|    static const cpp_dec_float& eps()
 | |
|    {
 | |
|       init.do_nothing();
 | |
|       static cpp_dec_float val(1.0, 1 - static_cast<int>(cpp_dec_float_digits10));
 | |
|       return val;
 | |
|    }
 | |
| 
 | |
|    // Basic operations.
 | |
|    cpp_dec_float& operator=(const cpp_dec_float& v) BOOST_MP_NOEXCEPT_IF(noexcept(std::declval<array_type&>() = std::declval<const array_type&>()))
 | |
|    {
 | |
|       data = v.data;
 | |
|       exp = v.exp;
 | |
|       neg = v.neg;
 | |
|       fpclass = v.fpclass;
 | |
|       prec_elem = v.prec_elem;
 | |
|       return *this;
 | |
|    }
 | |
| 
 | |
|    template <unsigned D>
 | |
|    cpp_dec_float& operator=(const cpp_dec_float<D>& f)
 | |
|    {
 | |
|       exp = f.exp;
 | |
|       neg = f.neg;
 | |
|       fpclass = static_cast<enum_fpclass_type>(static_cast<int>(f.fpclass));
 | |
|       unsigned elems = (std::min)(f.prec_elem, cpp_dec_float_elem_number);
 | |
|       std::copy(f.data.begin(), f.data.begin() + elems, data.begin());
 | |
|       std::fill(data.begin() + elems, data.end(), 0);
 | |
|       prec_elem = cpp_dec_float_elem_number;
 | |
|       return *this;
 | |
|    }
 | |
| 
 | |
|    cpp_dec_float& operator=(boost::long_long_type v)
 | |
|    {
 | |
|       if(v < 0)
 | |
|       {
 | |
|          from_unsigned_long_long(-v);
 | |
|          negate();
 | |
|       }
 | |
|       else
 | |
|          from_unsigned_long_long(v);
 | |
|       return *this;
 | |
|    }
 | |
| 
 | |
|    cpp_dec_float& operator=(boost::ulong_long_type v)
 | |
|    {
 | |
|       from_unsigned_long_long(v);
 | |
|       return *this;
 | |
|    }
 | |
| 
 | |
|    cpp_dec_float& operator=(long double v);
 | |
| 
 | |
|    cpp_dec_float& operator=(const char* v)
 | |
|    {
 | |
|       rd_string(v);
 | |
|       return *this;
 | |
|    }
 | |
| 
 | |
|    cpp_dec_float& operator+=(const cpp_dec_float& v);
 | |
|    cpp_dec_float& operator-=(const cpp_dec_float& v);
 | |
|    cpp_dec_float& operator*=(const cpp_dec_float& v);
 | |
|    cpp_dec_float& operator/=(const cpp_dec_float& v);
 | |
| 
 | |
|    cpp_dec_float& add_unsigned_long_long(const boost::ulong_long_type n)
 | |
|    {
 | |
|       cpp_dec_float t;
 | |
|       t.from_unsigned_long_long(n);
 | |
|       return *this += t;
 | |
|    }
 | |
| 
 | |
|    cpp_dec_float& sub_unsigned_long_long(const boost::ulong_long_type n)
 | |
|    {
 | |
|       cpp_dec_float t;
 | |
|       t.from_unsigned_long_long(n);
 | |
|       return *this -= t;
 | |
|    }
 | |
| 
 | |
|    cpp_dec_float& mul_unsigned_long_long(const boost::ulong_long_type n);
 | |
|    cpp_dec_float& div_unsigned_long_long(const boost::ulong_long_type n);
 | |
| 
 | |
|    // Elementary primitives.
 | |
|    cpp_dec_float& calculate_inv ();
 | |
|    cpp_dec_float& calculate_sqrt();
 | |
| 
 | |
|    void negate()
 | |
|    {
 | |
|       if(!iszero())
 | |
|          neg = !neg;
 | |
|    }
 | |
| 
 | |
|    // Comparison functions
 | |
|    bool isnan    BOOST_PREVENT_MACRO_SUBSTITUTION() const { return (fpclass == cpp_dec_float_NaN); }
 | |
|    bool isinf    BOOST_PREVENT_MACRO_SUBSTITUTION() const { return (fpclass == cpp_dec_float_inf); }
 | |
|    bool isfinite BOOST_PREVENT_MACRO_SUBSTITUTION() const { return (fpclass == cpp_dec_float_finite); }
 | |
| 
 | |
|    bool iszero () const
 | |
|    {
 | |
|       return ((fpclass == cpp_dec_float_finite) && (data[0u] == 0u));
 | |
|    }
 | |
| 
 | |
|    bool isone () const;
 | |
|    bool isint () const;
 | |
|    bool isneg () const { return neg; }
 | |
| 
 | |
|    // Operators pre-increment and pre-decrement
 | |
|    cpp_dec_float& operator++()
 | |
|    {
 | |
|       return *this += one();
 | |
|    }
 | |
| 
 | |
|    cpp_dec_float& operator--()
 | |
|    {
 | |
|       return *this -= one();
 | |
|    }
 | |
| 
 | |
|    std::string str(boost::intmax_t digits, std::ios_base::fmtflags f)const;
 | |
| 
 | |
|    int compare(const cpp_dec_float& v)const;
 | |
| 
 | |
|    template <class V>
 | |
|    int compare(const V& v)const
 | |
|    {
 | |
|       cpp_dec_float<Digits10, ExponentType, Allocator> t;
 | |
|       t = v;
 | |
|       return compare(t);
 | |
|    }
 | |
| 
 | |
|    void swap(cpp_dec_float& v)
 | |
|    {
 | |
|       data.swap(v.data);
 | |
|       std::swap(exp, v.exp);
 | |
|       std::swap(neg, v.neg);
 | |
|       std::swap(fpclass, v.fpclass);
 | |
|       std::swap(prec_elem, v.prec_elem);
 | |
|    }
 | |
| 
 | |
|    double extract_double() const;
 | |
|    long double extract_long_double() const;
 | |
|    boost::long_long_type extract_signed_long_long() const;
 | |
|    boost::ulong_long_type extract_unsigned_long_long() const;
 | |
|    void extract_parts(double& mantissa, ExponentType& exponent) const;
 | |
|    cpp_dec_float extract_integer_part() const;
 | |
| 
 | |
|    void precision(const boost::int32_t prec_digits)
 | |
|    {
 | |
|       if(prec_digits >= cpp_dec_float_total_digits10)
 | |
|       {
 | |
|          prec_elem = cpp_dec_float_elem_number;
 | |
|       }
 | |
|       else
 | |
|       {
 | |
|          const boost::int32_t elems = static_cast<boost::int32_t>( static_cast<boost::int32_t>( (prec_digits + (cpp_dec_float_elem_digits10 / 2)) / cpp_dec_float_elem_digits10)
 | |
|             + static_cast<boost::int32_t>(((prec_digits % cpp_dec_float_elem_digits10) != 0) ? 1 : 0));
 | |
| 
 | |
|          prec_elem = (std::min)(cpp_dec_float_elem_number, (std::max)(elems, static_cast<boost::int32_t>(2)));
 | |
|       }
 | |
|    }
 | |
|    static cpp_dec_float pow2(boost::long_long_type i);
 | |
|    ExponentType order()const
 | |
|    {
 | |
|       const bool bo_order_is_zero = ((!(isfinite)()) || (data[0] == static_cast<boost::uint32_t>(0u)));
 | |
|       //
 | |
|       // Binary search to find the order of the leading term:
 | |
|       //
 | |
|       ExponentType prefix = 0;
 | |
| 
 | |
|       if(data[0] >= 100000UL)
 | |
|       {
 | |
|          if(data[0] >= 10000000UL)
 | |
|          {
 | |
|             if(data[0] >= 100000000UL)
 | |
|             {
 | |
|                if(data[0] >= 1000000000UL)
 | |
|                   prefix = 9;
 | |
|                else
 | |
|                   prefix = 8;
 | |
|             }
 | |
|             else
 | |
|                prefix = 7;
 | |
|          }
 | |
|          else
 | |
|          {
 | |
|             if(data[0] >= 1000000UL)
 | |
|                prefix = 6;
 | |
|             else
 | |
|                prefix = 5;
 | |
|          }
 | |
|       }
 | |
|       else
 | |
|       {
 | |
|          if(data[0] >= 1000UL)
 | |
|          {
 | |
|             if(data[0] >= 10000UL)
 | |
|                prefix = 4;
 | |
|             else
 | |
|                prefix = 3;
 | |
|          }
 | |
|          else
 | |
|          {
 | |
|             if(data[0] >= 100)
 | |
|                prefix = 2;
 | |
|             else if(data[0] >= 10)
 | |
|                prefix = 1;
 | |
|          }
 | |
|       }
 | |
| 
 | |
|       return (bo_order_is_zero ? static_cast<ExponentType>(0) : static_cast<ExponentType>(exp + prefix));
 | |
|    }
 | |
| 
 | |
|    template<class Archive>
 | |
|    void serialize(Archive & ar, const unsigned int /*version*/)
 | |
|    {
 | |
|       for(unsigned i = 0; i < data.size(); ++i)
 | |
|          ar & data[i];
 | |
|       ar & exp;
 | |
|       ar & neg;
 | |
|       ar & fpclass;
 | |
|       ar & prec_elem;
 | |
|    }
 | |
| 
 | |
| private:
 | |
|    static bool data_elem_is_non_zero_predicate(const boost::uint32_t& d) { return (d != static_cast<boost::uint32_t>(0u)); }
 | |
|    static bool data_elem_is_non_nine_predicate(const boost::uint32_t& d) { return (d != static_cast<boost::uint32_t>(cpp_dec_float::cpp_dec_float_elem_mask - 1)); }
 | |
|    static bool char_is_nonzero_predicate(const char& c) { return (c != static_cast<char>('0')); }
 | |
| 
 | |
|    void from_unsigned_long_long(const boost::ulong_long_type u);
 | |
| 
 | |
|    int cmp_data(const array_type& vd) const;
 | |
| 
 | |
| 
 | |
|    static boost::uint32_t mul_loop_uv(boost::uint32_t* const u, const boost::uint32_t* const v, const boost::int32_t p);
 | |
|    static boost::uint32_t mul_loop_n (boost::uint32_t* const u, boost::uint32_t n, const boost::int32_t p);
 | |
|    static boost::uint32_t div_loop_n (boost::uint32_t* const u, boost::uint32_t n, const boost::int32_t p);
 | |
| 
 | |
|    bool rd_string(const char* const s);
 | |
| 
 | |
|    template <unsigned D, class ET, class A>
 | |
|    friend class cpp_dec_float;
 | |
| };
 | |
| 
 | |
| template <unsigned Digits10, class ExponentType, class Allocator>
 | |
| typename cpp_dec_float<Digits10, ExponentType, Allocator>::initializer cpp_dec_float<Digits10, ExponentType, Allocator>::init;
 | |
| 
 | |
| template <unsigned Digits10, class ExponentType, class Allocator>
 | |
| const boost::int32_t cpp_dec_float<Digits10, ExponentType, Allocator>::cpp_dec_float_radix;
 | |
| template <unsigned Digits10, class ExponentType, class Allocator>
 | |
| const boost::int32_t cpp_dec_float<Digits10, ExponentType, Allocator>::cpp_dec_float_digits10_setting;
 | |
| template <unsigned Digits10, class ExponentType, class Allocator>
 | |
| const boost::int32_t cpp_dec_float<Digits10, ExponentType, Allocator>::cpp_dec_float_digits10_limit_lo;
 | |
| template <unsigned Digits10, class ExponentType, class Allocator>
 | |
| const boost::int32_t cpp_dec_float<Digits10, ExponentType, Allocator>::cpp_dec_float_digits10_limit_hi;
 | |
| template <unsigned Digits10, class ExponentType, class Allocator>
 | |
| const boost::int32_t cpp_dec_float<Digits10, ExponentType, Allocator>::cpp_dec_float_digits10;
 | |
| template <unsigned Digits10, class ExponentType, class Allocator>
 | |
| const ExponentType cpp_dec_float<Digits10, ExponentType, Allocator>::cpp_dec_float_max_exp;
 | |
| template <unsigned Digits10, class ExponentType, class Allocator>
 | |
| const ExponentType cpp_dec_float<Digits10, ExponentType, Allocator>::cpp_dec_float_min_exp;
 | |
| template <unsigned Digits10, class ExponentType, class Allocator>
 | |
| const ExponentType cpp_dec_float<Digits10, ExponentType, Allocator>::cpp_dec_float_max_exp10;
 | |
| template <unsigned Digits10, class ExponentType, class Allocator>
 | |
| const ExponentType cpp_dec_float<Digits10, ExponentType, Allocator>::cpp_dec_float_min_exp10;
 | |
| template <unsigned Digits10, class ExponentType, class Allocator>
 | |
| const boost::int32_t cpp_dec_float<Digits10, ExponentType, Allocator>::cpp_dec_float_elem_digits10;
 | |
| template <unsigned Digits10, class ExponentType, class Allocator>
 | |
| const boost::int32_t cpp_dec_float<Digits10, ExponentType, Allocator>::cpp_dec_float_elem_number_request;
 | |
| template <unsigned Digits10, class ExponentType, class Allocator>
 | |
| const boost::int32_t cpp_dec_float<Digits10, ExponentType, Allocator>::cpp_dec_float_elem_number;
 | |
| template <unsigned Digits10, class ExponentType, class Allocator>
 | |
| const boost::int32_t cpp_dec_float<Digits10, ExponentType, Allocator>::cpp_dec_float_elem_mask;
 | |
| 
 | |
| template <unsigned Digits10, class ExponentType, class Allocator>
 | |
| cpp_dec_float<Digits10, ExponentType, Allocator>& cpp_dec_float<Digits10, ExponentType, Allocator>::operator+=(const cpp_dec_float<Digits10, ExponentType, Allocator>& v)
 | |
| {
 | |
|    if((isnan)())
 | |
|    {
 | |
|       return *this;
 | |
|    }
 | |
| 
 | |
|    if((isinf)())
 | |
|    {
 | |
|       if((v.isinf)() && (isneg() != v.isneg()))
 | |
|       {
 | |
|          *this = nan();
 | |
|       }
 | |
|       return *this;
 | |
|    }
 | |
| 
 | |
|    if(iszero())
 | |
|    {
 | |
|       return operator=(v);
 | |
|    }
 | |
| 
 | |
|    if((v.isnan)() || (v.isinf)())
 | |
|    {
 | |
|       *this = v;
 | |
|       return *this;
 | |
|    }
 | |
| 
 | |
|    // Get the offset for the add/sub operation.
 | |
|    static const ExponentType max_delta_exp = static_cast<ExponentType>((cpp_dec_float_elem_number - 1) * cpp_dec_float_elem_digits10);
 | |
| 
 | |
|    const ExponentType ofs_exp = static_cast<ExponentType>(exp - v.exp);
 | |
| 
 | |
|    // Check if the operation is out of range, requiring special handling.
 | |
|    if(v.iszero() || (ofs_exp > max_delta_exp))
 | |
|    {
 | |
|       // Result is *this unchanged since v is negligible compared to *this.
 | |
|       return *this;
 | |
|    }
 | |
|    else if(ofs_exp < -max_delta_exp)
 | |
|    {
 | |
|       // Result is *this = v since *this is negligible compared to v.
 | |
|       return operator=(v);
 | |
|    }
 | |
| 
 | |
|    // Do the add/sub operation.
 | |
| 
 | |
|    typename array_type::iterator p_u = data.begin();
 | |
|    typename array_type::const_iterator p_v = v.data.begin();
 | |
|    bool b_copy = false;
 | |
|    const boost::int32_t ofs = static_cast<boost::int32_t>(static_cast<boost::int32_t>(ofs_exp) / cpp_dec_float_elem_digits10);
 | |
|    array_type n_data;
 | |
| 
 | |
|    if(neg == v.neg)
 | |
|    {
 | |
|       // Add v to *this, where the data array of either *this or v
 | |
|       // might have to be treated with a positive, negative or zero offset.
 | |
|       // The result is stored in *this. The data are added one element
 | |
|       // at a time, each element with carry.
 | |
|       if(ofs >= static_cast<boost::int32_t>(0))
 | |
|       {
 | |
|          std::copy(v.data.begin(), v.data.end() - static_cast<size_t>(ofs), n_data.begin() + static_cast<size_t>(ofs));
 | |
|          std::fill(n_data.begin(), n_data.begin() + static_cast<size_t>(ofs), static_cast<boost::uint32_t>(0u));
 | |
|          p_v = n_data.begin();
 | |
|       }
 | |
|       else
 | |
|       {
 | |
|          std::copy(data.begin(), data.end() - static_cast<size_t>(-ofs), n_data.begin() + static_cast<size_t>(-ofs));
 | |
|          std::fill(n_data.begin(), n_data.begin() + static_cast<size_t>(-ofs), static_cast<boost::uint32_t>(0u));
 | |
|          p_u = n_data.begin();
 | |
|          b_copy = true;
 | |
|       }
 | |
| 
 | |
|       // Addition algorithm
 | |
|       boost::uint32_t carry = static_cast<boost::uint32_t>(0u);
 | |
| 
 | |
|       for(boost::int32_t j = static_cast<boost::int32_t>(cpp_dec_float_elem_number - static_cast<boost::int32_t>(1)); j >= static_cast<boost::int32_t>(0); j--)
 | |
|       {
 | |
|          boost::uint32_t t = static_cast<boost::uint32_t>(static_cast<boost::uint32_t>(p_u[j] + p_v[j]) + carry);
 | |
|          carry = t / static_cast<boost::uint32_t>(cpp_dec_float_elem_mask);
 | |
|          p_u[j] = static_cast<boost::uint32_t>(t - static_cast<boost::uint32_t>(carry * static_cast<boost::uint32_t>(cpp_dec_float_elem_mask)));
 | |
|       }
 | |
| 
 | |
|       if(b_copy)
 | |
|       {
 | |
|          data = n_data;
 | |
|          exp = v.exp;
 | |
|       }
 | |
| 
 | |
|       // There needs to be a carry into the element -1 of the array data
 | |
|       if(carry != static_cast<boost::uint32_t>(0u))
 | |
|       {
 | |
|          std::copy_backward(data.begin(), data.end() - static_cast<std::size_t>(1u), data.end());
 | |
|          data[0] = carry;
 | |
|          exp += static_cast<ExponentType>(cpp_dec_float_elem_digits10);
 | |
|       }
 | |
|    }
 | |
|    else
 | |
|    {
 | |
|       // Subtract v from *this, where the data array of either *this or v
 | |
|       // might have to be treated with a positive, negative or zero offset.
 | |
|       if((ofs > static_cast<boost::int32_t>(0))
 | |
|          || ( (ofs == static_cast<boost::int32_t>(0))
 | |
|          && (cmp_data(v.data) > static_cast<boost::int32_t>(0)))
 | |
|          )
 | |
|       {
 | |
|          // In this case, |u| > |v| and ofs is positive.
 | |
|          // Copy the data of v, shifted down to a lower value
 | |
|          // into the data array m_n. Set the operand pointer p_v
 | |
|          // to point to the copied, shifted data m_n.
 | |
|          std::copy(v.data.begin(), v.data.end() - static_cast<size_t>(ofs), n_data.begin() + static_cast<size_t>(ofs));
 | |
|          std::fill(n_data.begin(), n_data.begin() + static_cast<size_t>(ofs), static_cast<boost::uint32_t>(0u));
 | |
|          p_v = n_data.begin();
 | |
|       }
 | |
|       else
 | |
|       {
 | |
|          if(ofs != static_cast<boost::int32_t>(0))
 | |
|          {
 | |
|             // In this case, |u| < |v| and ofs is negative.
 | |
|             // Shift the data of u down to a lower value.
 | |
|             std::copy_backward(data.begin(), data.end() - static_cast<size_t>(-ofs), data.end());
 | |
|             std::fill(data.begin(), data.begin() + static_cast<size_t>(-ofs), static_cast<boost::uint32_t>(0u));
 | |
|          }
 | |
| 
 | |
|          // Copy the data of v into the data array n_data.
 | |
|          // Set the u-pointer p_u to point to m_n and the
 | |
|          // operand pointer p_v to point to the shifted
 | |
|          // data m_data.
 | |
|          n_data = v.data;
 | |
|          p_u = n_data.begin();
 | |
|          p_v = data.begin();
 | |
|          b_copy = true;
 | |
|       }
 | |
| 
 | |
|       boost::int32_t j;
 | |
| 
 | |
|       // Subtraction algorithm
 | |
|       boost::int32_t borrow = static_cast<boost::int32_t>(0);
 | |
| 
 | |
|       for(j = static_cast<boost::int32_t>(cpp_dec_float_elem_number - static_cast<boost::int32_t>(1)); j >= static_cast<boost::int32_t>(0); j--)
 | |
|       {
 | |
|          boost::int32_t t = static_cast<boost::int32_t>(static_cast<boost::int32_t>( static_cast<boost::int32_t>(p_u[j])
 | |
|             - static_cast<boost::int32_t>(p_v[j])) - borrow);
 | |
| 
 | |
|          // Underflow? Borrow?
 | |
|          if(t < static_cast<boost::int32_t>(0))
 | |
|          {
 | |
|             // Yes, underflow and borrow
 | |
|             t += static_cast<boost::int32_t>(cpp_dec_float_elem_mask);
 | |
|             borrow = static_cast<boost::int32_t>(1);
 | |
|          }
 | |
|          else
 | |
|          {
 | |
|             borrow = static_cast<boost::int32_t>(0);
 | |
|          }
 | |
| 
 | |
|          p_u[j] = static_cast<boost::uint32_t>(static_cast<boost::uint32_t>(t) % static_cast<boost::uint32_t>(cpp_dec_float_elem_mask));
 | |
|       }
 | |
| 
 | |
|       if(b_copy)
 | |
|       {
 | |
|          data = n_data;
 | |
|          exp = v.exp;
 | |
|          neg = v.neg;
 | |
|       }
 | |
| 
 | |
|       // Is it necessary to justify the data?
 | |
|       const typename array_type::const_iterator first_nonzero_elem = std::find_if(data.begin(), data.end(), data_elem_is_non_zero_predicate);
 | |
| 
 | |
|       if(first_nonzero_elem != data.begin())
 | |
|       {
 | |
|          if(first_nonzero_elem == data.end())
 | |
|          {
 | |
|             // This result of the subtraction is exactly zero.
 | |
|             // Reset the sign and the exponent.
 | |
|             neg = false;
 | |
|             exp = static_cast<ExponentType>(0);
 | |
|          }
 | |
|          else
 | |
|          {
 | |
|             // Justify the data
 | |
|             const std::size_t sj = static_cast<std::size_t>(std::distance<typename array_type::const_iterator>(data.begin(), first_nonzero_elem));
 | |
| 
 | |
|             std::copy(data.begin() + static_cast<std::size_t>(sj), data.end(), data.begin());
 | |
|             std::fill(data.end() - sj, data.end(), static_cast<boost::uint32_t>(0u));
 | |
| 
 | |
|             exp -= static_cast<ExponentType>(sj * static_cast<std::size_t>(cpp_dec_float_elem_digits10));
 | |
|          }
 | |
|       }
 | |
|    }
 | |
| 
 | |
|    // Handle underflow.
 | |
|    if(iszero())
 | |
|       return (*this = zero());
 | |
| 
 | |
|    // Check for potential overflow.
 | |
|    const bool b_result_might_overflow  = (exp >= static_cast<ExponentType>(cpp_dec_float_max_exp10));
 | |
| 
 | |
|    // Handle overflow.
 | |
|    if(b_result_might_overflow)
 | |
|    {
 | |
|       const bool b_result_is_neg = neg;
 | |
|       neg = false;
 | |
| 
 | |
|       if(compare((cpp_dec_float::max)()) > 0)
 | |
|         *this = inf();
 | |
| 
 | |
|       neg = b_result_is_neg;
 | |
|    }
 | |
| 
 | |
|    return *this;
 | |
| }
 | |
| 
 | |
| template <unsigned Digits10, class ExponentType, class Allocator>
 | |
| cpp_dec_float<Digits10, ExponentType, Allocator>& cpp_dec_float<Digits10, ExponentType, Allocator>::operator-=(const cpp_dec_float<Digits10, ExponentType, Allocator>& v)
 | |
| {
 | |
|    // Use *this - v = -(-*this + v).
 | |
|    negate();
 | |
|    *this += v;
 | |
|    negate();
 | |
|    return *this;
 | |
| }
 | |
| 
 | |
| template <unsigned Digits10, class ExponentType, class Allocator>
 | |
| cpp_dec_float<Digits10, ExponentType, Allocator>& cpp_dec_float<Digits10, ExponentType, Allocator>::operator*=(const cpp_dec_float<Digits10, ExponentType, Allocator>& v)
 | |
| {
 | |
|    // Evaluate the sign of the result.
 | |
|    const bool b_result_is_neg = (neg != v.neg);
 | |
| 
 | |
|    // Artificially set the sign of the result to be positive.
 | |
|    neg = false;
 | |
| 
 | |
|    // Handle special cases like zero, inf and NaN.
 | |
|    const bool b_u_is_inf  = (isinf)();
 | |
|    const bool b_v_is_inf  = (v.isinf)();
 | |
|    const bool b_u_is_zero = iszero();
 | |
|    const bool b_v_is_zero = v.iszero();
 | |
| 
 | |
|    if(   ((isnan)() || (v.isnan)())
 | |
|       || (b_u_is_inf && b_v_is_zero)
 | |
|       || (b_v_is_inf && b_u_is_zero)
 | |
|       )
 | |
|    {
 | |
|       *this = nan();
 | |
|       return *this;
 | |
|    }
 | |
| 
 | |
|    if(b_u_is_inf || b_v_is_inf)
 | |
|    {
 | |
|       *this = inf();
 | |
|       if(b_result_is_neg)
 | |
|          negate();
 | |
|       return *this;
 | |
|    }
 | |
| 
 | |
|    if(b_u_is_zero || b_v_is_zero)
 | |
|    {
 | |
|       return *this = zero();
 | |
|    }
 | |
| 
 | |
|    // Check for potential overflow or underflow.
 | |
|    const bool b_result_might_overflow  = ((exp + v.exp) >= static_cast<ExponentType>(cpp_dec_float_max_exp10));
 | |
|    const bool b_result_might_underflow = ((exp + v.exp) <= static_cast<ExponentType>(cpp_dec_float_min_exp10));
 | |
| 
 | |
|    // Set the exponent of the result.
 | |
|    exp += v.exp;
 | |
| 
 | |
|    const boost::int32_t prec_mul = (std::min)(prec_elem, v.prec_elem);
 | |
| 
 | |
|    const boost::uint32_t carry = mul_loop_uv(data.data(), v.data.data(), prec_mul);
 | |
| 
 | |
|    // Handle a potential carry.
 | |
|    if(carry != static_cast<boost::uint32_t>(0u))
 | |
|    {
 | |
|       exp += cpp_dec_float_elem_digits10;
 | |
| 
 | |
|       // Shift the result of the multiplication one element to the right...
 | |
|       std::copy_backward(data.begin(),
 | |
|                          data.begin() + static_cast<std::size_t>(prec_elem - static_cast<boost::int32_t>(1)),
 | |
|                          data.begin() + static_cast<std::size_t>(prec_elem));
 | |
| 
 | |
|       // ... And insert the carry.
 | |
|       data.front() = carry;
 | |
|    }
 | |
| 
 | |
|    // Handle overflow.
 | |
|    if(b_result_might_overflow && (compare((cpp_dec_float::max)()) > 0))
 | |
|    {
 | |
|       *this = inf();
 | |
|    }
 | |
| 
 | |
|    // Handle underflow.
 | |
|    if(b_result_might_underflow && (compare((cpp_dec_float::min)()) < 0))
 | |
|    {
 | |
|       *this = zero();
 | |
| 
 | |
|       return *this;
 | |
|    }
 | |
| 
 | |
|    // Set the sign of the result.
 | |
|    neg = b_result_is_neg;
 | |
| 
 | |
|    return *this;
 | |
| }
 | |
| 
 | |
| template <unsigned Digits10, class ExponentType, class Allocator>
 | |
| cpp_dec_float<Digits10, ExponentType, Allocator>& cpp_dec_float<Digits10, ExponentType, Allocator>::operator/=(const cpp_dec_float<Digits10, ExponentType, Allocator>& v)
 | |
| {
 | |
|    if(iszero())
 | |
|    {
 | |
|       if((v.isnan)())
 | |
|       {
 | |
|          return *this = v;
 | |
|       }
 | |
|       else if(v.iszero())
 | |
|       {
 | |
|          return *this = nan();
 | |
|       }
 | |
|    }
 | |
| 
 | |
|    const bool u_and_v_are_finite_and_identical = ( (isfinite)()
 | |
|       && (fpclass == v.fpclass)
 | |
|       && (exp == v.exp)
 | |
|       && (cmp_data(v.data) == static_cast<boost::int32_t>(0)));
 | |
| 
 | |
|    if(u_and_v_are_finite_and_identical)
 | |
|    {
 | |
|       if(neg != v.neg)
 | |
|       {
 | |
|          *this = one();
 | |
|          negate();
 | |
|       }
 | |
|       else
 | |
|          *this = one();
 | |
|       return *this;
 | |
|    }
 | |
|    else
 | |
|    {
 | |
|       cpp_dec_float t(v);
 | |
|       t.calculate_inv();
 | |
|       return operator*=(t);
 | |
|    }
 | |
| }
 | |
| 
 | |
| template <unsigned Digits10, class ExponentType, class Allocator>
 | |
| cpp_dec_float<Digits10, ExponentType, Allocator>& cpp_dec_float<Digits10, ExponentType, Allocator>::mul_unsigned_long_long(const boost::ulong_long_type n)
 | |
| {
 | |
|    // Multiply *this with a constant boost::ulong_long_type.
 | |
| 
 | |
|    // Evaluate the sign of the result.
 | |
|    const bool b_neg = neg;
 | |
| 
 | |
|    // Artificially set the sign of the result to be positive.
 | |
|    neg = false;
 | |
| 
 | |
|    // Handle special cases like zero, inf and NaN.
 | |
|    const bool b_u_is_inf = (isinf)();
 | |
|    const bool b_n_is_zero = (n == static_cast<boost::int32_t>(0));
 | |
| 
 | |
|    if((isnan)() || (b_u_is_inf && b_n_is_zero))
 | |
|    {
 | |
|       return (*this = nan());
 | |
|    }
 | |
| 
 | |
|    if(b_u_is_inf)
 | |
|    {
 | |
|       *this = inf();
 | |
|       if(b_neg)
 | |
|          negate();
 | |
|       return *this;
 | |
|    }
 | |
| 
 | |
|    if(iszero() || b_n_is_zero)
 | |
|    {
 | |
|       // Multiplication by zero.
 | |
|       return *this = zero();
 | |
|    }
 | |
| 
 | |
|    if(n >= static_cast<boost::ulong_long_type>(cpp_dec_float_elem_mask))
 | |
|    {
 | |
|       neg = b_neg;
 | |
|       cpp_dec_float t;
 | |
|       t = n;
 | |
|       return operator*=(t);
 | |
|    }
 | |
| 
 | |
|    if(n == static_cast<boost::ulong_long_type>(1u))
 | |
|    {
 | |
|       neg = b_neg;
 | |
|       return *this;
 | |
|    }
 | |
| 
 | |
|    // Set up the multiplication loop.
 | |
|    const boost::uint32_t nn = static_cast<boost::uint32_t>(n);
 | |
|    const boost::uint32_t carry = mul_loop_n(data.data(), nn, prec_elem);
 | |
| 
 | |
|    // Handle the carry and adjust the exponent.
 | |
|    if(carry != static_cast<boost::uint32_t>(0u))
 | |
|    {
 | |
|       exp += static_cast<ExponentType>(cpp_dec_float_elem_digits10);
 | |
| 
 | |
|       // Shift the result of the multiplication one element to the right.
 | |
|       std::copy_backward(data.begin(),
 | |
|          data.begin() + static_cast<std::size_t>(prec_elem - static_cast<boost::int32_t>(1)),
 | |
|          data.begin() + static_cast<std::size_t>(prec_elem));
 | |
| 
 | |
|       data.front() = static_cast<boost::uint32_t>(carry);
 | |
|    }
 | |
| 
 | |
|    // Check for potential overflow.
 | |
|    const bool b_result_might_overflow = (exp >= cpp_dec_float_max_exp10);
 | |
| 
 | |
|    // Handle overflow.
 | |
|    if(b_result_might_overflow && (compare((cpp_dec_float::max)()) > 0))
 | |
|    {
 | |
|       *this = inf();
 | |
|    }
 | |
| 
 | |
|    // Set the sign.
 | |
|    neg = b_neg;
 | |
| 
 | |
|    return *this;
 | |
| }
 | |
| 
 | |
| template <unsigned Digits10, class ExponentType, class Allocator>
 | |
| cpp_dec_float<Digits10, ExponentType, Allocator>& cpp_dec_float<Digits10, ExponentType, Allocator>::div_unsigned_long_long(const boost::ulong_long_type n)
 | |
| {
 | |
|    // Divide *this by a constant boost::ulong_long_type.
 | |
| 
 | |
|    // Evaluate the sign of the result.
 | |
|    const bool b_neg = neg;
 | |
| 
 | |
|    // Artificially set the sign of the result to be positive.
 | |
|    neg = false;
 | |
| 
 | |
|    // Handle special cases like zero, inf and NaN.
 | |
|    if((isnan)())
 | |
|    {
 | |
|       return *this;
 | |
|    }
 | |
| 
 | |
|    if((isinf)())
 | |
|    {
 | |
|       *this = inf();
 | |
|       if(b_neg)
 | |
|          negate();
 | |
|       return *this;
 | |
|    }
 | |
| 
 | |
|    if(n == static_cast<boost::ulong_long_type>(0u))
 | |
|    {
 | |
|       // Divide by 0.
 | |
|       if(iszero())
 | |
|       {
 | |
|          *this = nan();
 | |
|          return *this;
 | |
|       }
 | |
|       else
 | |
|       {
 | |
|          *this = inf();
 | |
|          if(isneg())
 | |
|             negate();
 | |
|          return *this;
 | |
|       }
 | |
|    }
 | |
| 
 | |
|    if(iszero())
 | |
|    {
 | |
|       return *this;
 | |
|    }
 | |
| 
 | |
|    if(n >= static_cast<boost::ulong_long_type>(cpp_dec_float_elem_mask))
 | |
|    {
 | |
|       neg = b_neg;
 | |
|       cpp_dec_float t;
 | |
|       t = n;
 | |
|       return operator/=(t);
 | |
|    }
 | |
| 
 | |
|    const boost::uint32_t nn = static_cast<boost::uint32_t>(n);
 | |
| 
 | |
|    if(nn > static_cast<boost::uint32_t>(1u))
 | |
|    {
 | |
|       // Do the division loop.
 | |
|       const boost::uint32_t prev = div_loop_n(data.data(), nn, prec_elem);
 | |
| 
 | |
|       // Determine if one leading zero is in the result data.
 | |
|       if(data[0] == static_cast<boost::uint32_t>(0u))
 | |
|       {
 | |
|          // Adjust the exponent
 | |
|          exp -= static_cast<ExponentType>(cpp_dec_float_elem_digits10);
 | |
| 
 | |
|          // Shift result of the division one element to the left.
 | |
|          std::copy(data.begin() + static_cast<std::size_t>(1u),
 | |
|             data.begin() + static_cast<std::size_t>(prec_elem - static_cast<boost::int32_t>(1)),
 | |
|             data.begin());
 | |
| 
 | |
|          data[prec_elem - static_cast<boost::int32_t>(1)] = static_cast<boost::uint32_t>(static_cast<boost::uint64_t>(prev * static_cast<boost::uint64_t>(cpp_dec_float_elem_mask)) / nn);
 | |
|       }
 | |
|    }
 | |
| 
 | |
|    // Check for potential underflow.
 | |
|    const bool b_result_might_underflow = (exp <= cpp_dec_float_min_exp10);
 | |
| 
 | |
|    // Handle underflow.
 | |
|    if(b_result_might_underflow && (compare((cpp_dec_float::min)()) < 0))
 | |
|       return (*this = zero());
 | |
| 
 | |
|    // Set the sign of the result.
 | |
|    neg = b_neg;
 | |
| 
 | |
|    return *this;
 | |
| }
 | |
| 
 | |
| template <unsigned Digits10, class ExponentType, class Allocator>
 | |
| cpp_dec_float<Digits10, ExponentType, Allocator>& cpp_dec_float<Digits10, ExponentType, Allocator>::calculate_inv()
 | |
| {
 | |
|    // Compute the inverse of *this.
 | |
|    const bool b_neg = neg;
 | |
| 
 | |
|    neg = false;
 | |
| 
 | |
|    // Handle special cases like zero, inf and NaN.
 | |
|    if(iszero())
 | |
|    {
 | |
|       *this = inf();
 | |
|       if(b_neg)
 | |
|          negate();
 | |
|       return *this;
 | |
|    }
 | |
| 
 | |
|    if((isnan)())
 | |
|    {
 | |
|       return *this;
 | |
|    }
 | |
| 
 | |
|    if((isinf)())
 | |
|    {
 | |
|       return *this = zero();
 | |
|    }
 | |
| 
 | |
|    if(isone())
 | |
|    {
 | |
|       if(b_neg)
 | |
|          negate();
 | |
|       return *this;
 | |
|    }
 | |
| 
 | |
|    // Save the original *this.
 | |
|    cpp_dec_float<Digits10, ExponentType, Allocator> x(*this);
 | |
| 
 | |
|    // Generate the initial estimate using division.
 | |
|    // Extract the mantissa and exponent for a "manual"
 | |
|    // computation of the estimate.
 | |
|    double dd;
 | |
|    ExponentType ne;
 | |
|    x.extract_parts(dd, ne);
 | |
| 
 | |
|    // Do the inverse estimate using double precision estimates of mantissa and exponent.
 | |
|    operator=(cpp_dec_float<Digits10, ExponentType, Allocator>(1.0 / dd, -ne));
 | |
| 
 | |
|    // Compute the inverse of *this. Quadratically convergent Newton-Raphson iteration
 | |
|    // is used. During the iterative steps, the precision of the calculation is limited
 | |
|    // to the minimum required in order to minimize the run-time.
 | |
| 
 | |
|    static const boost::int32_t double_digits10_minus_a_few = std::numeric_limits<double>::digits10 - 3;
 | |
| 
 | |
|    for(boost::int32_t digits = double_digits10_minus_a_few; digits <= cpp_dec_float_total_digits10; digits *= static_cast<boost::int32_t>(2))
 | |
|    {
 | |
|       // Adjust precision of the terms.
 | |
|       precision(static_cast<boost::int32_t>((digits + 10) * static_cast<boost::int32_t>(2)));
 | |
|       x.precision(static_cast<boost::int32_t>((digits + 10) * static_cast<boost::int32_t>(2)));
 | |
| 
 | |
|       // Next iteration.
 | |
|       cpp_dec_float t(*this);
 | |
|       t *= x;
 | |
|       t -= two();
 | |
|       t.negate();
 | |
|       *this *= t;
 | |
|    }
 | |
| 
 | |
|    neg = b_neg;
 | |
| 
 | |
|    prec_elem = cpp_dec_float_elem_number;
 | |
| 
 | |
|    return *this;
 | |
| }
 | |
| 
 | |
| template <unsigned Digits10, class ExponentType, class Allocator>
 | |
| cpp_dec_float<Digits10, ExponentType, Allocator>& cpp_dec_float<Digits10, ExponentType, Allocator>::calculate_sqrt()
 | |
| {
 | |
|    // Compute the square root of *this.
 | |
| 
 | |
|    if(isneg() || (!(isfinite)()))
 | |
|    {
 | |
|       *this = nan();
 | |
|       return *this;
 | |
|    }
 | |
| 
 | |
|    if(iszero() || isone())
 | |
|    {
 | |
|       return *this;
 | |
|    }
 | |
| 
 | |
|    // Save the original *this.
 | |
|    cpp_dec_float<Digits10, ExponentType, Allocator> x(*this);
 | |
| 
 | |
|    // Generate the initial estimate using division.
 | |
|    // Extract the mantissa and exponent for a "manual"
 | |
|    // computation of the estimate.
 | |
|    double dd;
 | |
|    ExponentType ne;
 | |
|    extract_parts(dd, ne);
 | |
| 
 | |
|    // Force the exponent to be an even multiple of two.
 | |
|    if((ne % static_cast<ExponentType>(2)) != static_cast<ExponentType>(0))
 | |
|    {
 | |
|       ++ne;
 | |
|       dd /= 10.0;
 | |
|    }
 | |
| 
 | |
|    // Setup the iteration.
 | |
|    // Estimate the square root using simple manipulations.
 | |
|    const double sqd = std::sqrt(dd);
 | |
| 
 | |
|    *this = cpp_dec_float<Digits10, ExponentType, Allocator>(sqd, static_cast<ExponentType>(ne / static_cast<ExponentType>(2)));
 | |
| 
 | |
|    // Estimate 1.0 / (2.0 * x0) using simple manipulations.
 | |
|    cpp_dec_float<Digits10, ExponentType, Allocator> vi(0.5 / sqd, static_cast<ExponentType>(-ne / static_cast<ExponentType>(2)));
 | |
| 
 | |
|    // Compute the square root of x. Coupled Newton iteration
 | |
|    // as described in "Pi Unleashed" is used. During the
 | |
|    // iterative steps, the precision of the calculation is
 | |
|    // limited to the minimum required in order to minimize
 | |
|    // the run-time.
 | |
|    //
 | |
|    // Book references:
 | |
|    // http://www.jjj.de/pibook/pibook.html
 | |
|    // http://www.amazon.com/exec/obidos/tg/detail/-/3540665722/qid=1035535482/sr=8-7/ref=sr_8_7/104-3357872-6059916?v=glance&n=507846
 | |
| 
 | |
|    static const boost::uint32_t double_digits10_minus_a_few = std::numeric_limits<double>::digits10 - 3;
 | |
| 
 | |
|    for(boost::int32_t digits = double_digits10_minus_a_few; digits <= cpp_dec_float_total_digits10; digits *= 2u)
 | |
|    {
 | |
|       // Adjust precision of the terms.
 | |
|       precision((digits + 10) * 2);
 | |
|       vi.precision((digits + 10) * 2);
 | |
| 
 | |
|       // Next iteration of vi
 | |
|       cpp_dec_float t(*this);
 | |
|       t *= vi;
 | |
|       t.negate();
 | |
|       t.mul_unsigned_long_long(2u);
 | |
|       t += one();
 | |
|       t *= vi;
 | |
|       vi += t;
 | |
| 
 | |
|       // Next iteration of *this
 | |
|       t = *this;
 | |
|       t *= *this;
 | |
|       t.negate();
 | |
|       t += x;
 | |
|       t *= vi;
 | |
|       *this += t;
 | |
|    }
 | |
| 
 | |
|    prec_elem = cpp_dec_float_elem_number;
 | |
| 
 | |
|    return *this;
 | |
| }
 | |
| 
 | |
| template <unsigned Digits10, class ExponentType, class Allocator>
 | |
| int cpp_dec_float<Digits10, ExponentType, Allocator>::cmp_data(const array_type& vd) const
 | |
| {
 | |
|    // Compare the data of *this with those of v.
 | |
|    // Return +1 for *this > v
 | |
|    // 0 for *this = v
 | |
|    // -1 for *this < v
 | |
| 
 | |
|    const std::pair<typename array_type::const_iterator, typename array_type::const_iterator> mismatch_pair = std::mismatch(data.begin(), data.end(), vd.begin());
 | |
| 
 | |
|    const bool is_equal = ((mismatch_pair.first == data.end()) && (mismatch_pair.second == vd.end()));
 | |
| 
 | |
|    if(is_equal)
 | |
|    {
 | |
|       return 0;
 | |
|    }
 | |
|    else
 | |
|    {
 | |
|       return ((*mismatch_pair.first > *mismatch_pair.second) ? 1 : -1);
 | |
|    }
 | |
| }
 | |
| 
 | |
| template <unsigned Digits10, class ExponentType, class Allocator>
 | |
| int cpp_dec_float<Digits10, ExponentType, Allocator>::compare(const cpp_dec_float& v) const
 | |
| {
 | |
|    // Compare v with *this.
 | |
|    // Return +1 for *this > v
 | |
|    // 0 for *this = v
 | |
|    // -1 for *this < v
 | |
| 
 | |
|    // Handle all non-finite cases.
 | |
|    if((!(isfinite)()) || (!(v.isfinite)()))
 | |
|    {
 | |
|       // NaN can never equal NaN. Return an implementation-dependent
 | |
|       // signed result. Also note that comparison of NaN with NaN
 | |
|       // using operators greater-than or less-than is undefined.
 | |
|       if((isnan)() || (v.isnan)()) { return ((isnan)() ? 1 : -1); }
 | |
| 
 | |
|       if((isinf)() && (v.isinf)())
 | |
|       {
 | |
|          // Both *this and v are infinite. They are equal if they have the same sign.
 | |
|          // Otherwise, *this is less than v if and only if *this is negative.
 | |
|          return ((neg == v.neg) ? 0 : (neg ? -1 : 1));
 | |
|       }
 | |
| 
 | |
|       if((isinf)())
 | |
|       {
 | |
|          // *this is infinite, but v is finite.
 | |
|          // So negative infinite *this is less than any finite v.
 | |
|          // Whereas positive infinite *this is greater than any finite v.
 | |
|          return (isneg() ? -1 : 1);
 | |
|       }
 | |
|       else
 | |
|       {
 | |
|          // *this is finite, and v is infinite.
 | |
|          // So any finite *this is greater than negative infinite v.
 | |
|          // Whereas any finite *this is less than positive infinite v.
 | |
|          return (v.neg ? 1 : -1);
 | |
|       }
 | |
|    }
 | |
| 
 | |
|    // And now handle all *finite* cases.
 | |
|    if(iszero())
 | |
|    {
 | |
|       // The value of *this is zero and v is either zero or non-zero.
 | |
|       return (v.iszero() ? 0
 | |
|          : (v.neg ? 1 : -1));
 | |
|    }
 | |
|    else if(v.iszero())
 | |
|    {
 | |
|       // The value of v is zero and *this is non-zero.
 | |
|       return (neg ? -1 : 1);
 | |
|    }
 | |
|    else
 | |
|    {
 | |
|       // Both *this and v are non-zero.
 | |
| 
 | |
|       if(neg != v.neg)
 | |
|       {
 | |
|          // The signs are different.
 | |
|          return (neg ? -1 : 1);
 | |
|       }
 | |
|       else if(exp != v.exp)
 | |
|       {
 | |
|          // The signs are the same and the exponents are different.
 | |
|          const int val_cexpression = ((exp < v.exp) ? 1 : -1);
 | |
| 
 | |
|          return (neg ? val_cexpression : -val_cexpression);
 | |
|       }
 | |
|       else
 | |
|       {
 | |
|          // The signs are the same and the exponents are the same.
 | |
|          // Compare the data.
 | |
|          const int val_cmp_data = cmp_data(v.data);
 | |
| 
 | |
|          return ((!neg) ? val_cmp_data : -val_cmp_data);
 | |
|       }
 | |
|    }
 | |
| }
 | |
| 
 | |
| template <unsigned Digits10, class ExponentType, class Allocator>
 | |
| bool cpp_dec_float<Digits10, ExponentType, Allocator>::isone() const
 | |
| {
 | |
|    // Check if the value of *this is identically 1 or very close to 1.
 | |
| 
 | |
|    const bool not_negative_and_is_finite = ((!neg) && (isfinite)());
 | |
| 
 | |
|    if(not_negative_and_is_finite)
 | |
|    {
 | |
|       if((data[0u] == static_cast<boost::uint32_t>(1u)) && (exp == static_cast<ExponentType>(0)))
 | |
|       {
 | |
|          const typename array_type::const_iterator it_non_zero = std::find_if(data.begin(), data.end(), data_elem_is_non_zero_predicate);
 | |
|          return (it_non_zero == data.end());
 | |
|       }
 | |
|       else if((data[0u] == static_cast<boost::uint32_t>(cpp_dec_float_elem_mask - 1)) && (exp == static_cast<ExponentType>(-cpp_dec_float_elem_digits10)))
 | |
|       {
 | |
|          const typename array_type::const_iterator it_non_nine = std::find_if(data.begin(), data.end(), data_elem_is_non_nine_predicate);
 | |
|          return (it_non_nine == data.end());
 | |
|       }
 | |
|    }
 | |
| 
 | |
|    return false;
 | |
| }
 | |
| 
 | |
| template <unsigned Digits10, class ExponentType, class Allocator>
 | |
| bool cpp_dec_float<Digits10, ExponentType, Allocator>::isint() const
 | |
| {
 | |
|    if(fpclass != cpp_dec_float_finite) { return false; }
 | |
| 
 | |
|    if(iszero()) { return true; }
 | |
| 
 | |
|    if(exp < static_cast<ExponentType>(0)) { return false; } // |*this| < 1.
 | |
| 
 | |
|    const typename array_type::size_type offset_decimal_part = static_cast<typename array_type::size_type>(exp / cpp_dec_float_elem_digits10) + 1u;
 | |
| 
 | |
|    if(offset_decimal_part >= static_cast<typename array_type::size_type>(cpp_dec_float_elem_number))
 | |
|    {
 | |
|       // The number is too large to resolve the integer part.
 | |
|       // It considered to be a pure integer.
 | |
|       return true;
 | |
|    }
 | |
| 
 | |
|    typename array_type::const_iterator it_non_zero = std::find_if(data.begin() + offset_decimal_part, data.end(), data_elem_is_non_zero_predicate);
 | |
| 
 | |
|    return (it_non_zero == data.end());
 | |
| }
 | |
| 
 | |
| template <unsigned Digits10, class ExponentType, class Allocator>
 | |
| void cpp_dec_float<Digits10, ExponentType, Allocator>::extract_parts(double& mantissa, ExponentType& exponent) const
 | |
| {
 | |
|    // Extract the approximate parts mantissa and base-10 exponent from the input cpp_dec_float<Digits10, ExponentType, Allocator> value x.
 | |
| 
 | |
|    // Extracts the mantissa and exponent.
 | |
|    exponent = exp;
 | |
| 
 | |
|    boost::uint32_t p10 = static_cast<boost::uint32_t>(1u);
 | |
|    boost::uint32_t test = data[0u];
 | |
| 
 | |
|    for(;;)
 | |
|    {
 | |
|       test /= static_cast<boost::uint32_t>(10u);
 | |
| 
 | |
|       if(test == static_cast<boost::uint32_t>(0u))
 | |
|       {
 | |
|          break;
 | |
|       }
 | |
| 
 | |
|       p10 *= static_cast<boost::uint32_t>(10u);
 | |
|       ++exponent;
 | |
|    }
 | |
| 
 | |
|    // Establish the upper bound of limbs for extracting the double.
 | |
|    const int max_elem_in_double_count = static_cast<int>(static_cast<boost::int32_t>(std::numeric_limits<double>::digits10) / cpp_dec_float_elem_digits10)
 | |
|                                          + (static_cast<int>(static_cast<boost::int32_t>(std::numeric_limits<double>::digits10) % cpp_dec_float_elem_digits10) != 0 ? 1 : 0)
 | |
|                                          + 1;
 | |
| 
 | |
|    // And make sure this upper bound stays within bounds of the elems.
 | |
|    const std::size_t max_elem_extract_count = static_cast<std::size_t>((std::min)(static_cast<boost::int32_t>(max_elem_in_double_count), cpp_dec_float_elem_number));
 | |
| 
 | |
|    // Extract into the mantissa the first limb, extracted as a double.
 | |
|    mantissa = static_cast<double>(data[0]);
 | |
|    double scale = 1.0;
 | |
| 
 | |
|    // Extract the rest of the mantissa piecewise from the limbs.
 | |
|    for(std::size_t i = 1u; i < max_elem_extract_count; i++)
 | |
|    {
 | |
|      scale /= static_cast<double>(cpp_dec_float_elem_mask);
 | |
|      mantissa += (static_cast<double>(data[i]) * scale);
 | |
|    }
 | |
| 
 | |
|    mantissa /= static_cast<double>(p10);
 | |
| 
 | |
|    if(neg) { mantissa = -mantissa; }
 | |
| }
 | |
| 
 | |
| template <unsigned Digits10, class ExponentType, class Allocator>
 | |
| double cpp_dec_float<Digits10, ExponentType, Allocator>::extract_double() const
 | |
| {
 | |
|    // Returns the double conversion of a cpp_dec_float<Digits10, ExponentType, Allocator>.
 | |
| 
 | |
|    // Check for non-normal cpp_dec_float<Digits10, ExponentType, Allocator>.
 | |
|    if(!(isfinite)())
 | |
|    {
 | |
|       if((isnan)())
 | |
|       {
 | |
|          return std::numeric_limits<double>::quiet_NaN();
 | |
|       }
 | |
|       else
 | |
|       {
 | |
|          return ((!neg) ? std::numeric_limits<double>::infinity()
 | |
|             : -std::numeric_limits<double>::infinity());
 | |
|       }
 | |
|    }
 | |
| 
 | |
|    cpp_dec_float<Digits10, ExponentType, Allocator> xx(*this);
 | |
|    if(xx.isneg())
 | |
|       xx.negate();
 | |
| 
 | |
|    // Check if *this cpp_dec_float<Digits10, ExponentType, Allocator> is zero.
 | |
|    if(iszero() || (xx.compare(double_min()) < 0))
 | |
|    {
 | |
|       return 0.0;
 | |
|    }
 | |
| 
 | |
|    // Check if *this cpp_dec_float<Digits10, ExponentType, Allocator> exceeds the maximum of double.
 | |
|    if(xx.compare(double_max()) > 0)
 | |
|    {
 | |
|       return ((!neg) ? std::numeric_limits<double>::infinity()
 | |
|          : -std::numeric_limits<double>::infinity());
 | |
|    }
 | |
| 
 | |
|    std::stringstream ss;
 | |
| 
 | |
|    ss << str(std::numeric_limits<double>::digits10 + (2 + 1), std::ios_base::scientific);
 | |
| 
 | |
|    double d;
 | |
|    ss >> d;
 | |
| 
 | |
|    return d;
 | |
| }
 | |
| 
 | |
| template <unsigned Digits10, class ExponentType, class Allocator>
 | |
| long double cpp_dec_float<Digits10, ExponentType, Allocator>::extract_long_double() const
 | |
| {
 | |
|    // Returns the long double conversion of a cpp_dec_float<Digits10, ExponentType, Allocator>.
 | |
| 
 | |
|    // Check if *this cpp_dec_float<Digits10, ExponentType, Allocator> is subnormal.
 | |
|    if(!(isfinite)())
 | |
|    {
 | |
|       if((isnan)())
 | |
|       {
 | |
|          return std::numeric_limits<long double>::quiet_NaN();
 | |
|       }
 | |
|       else
 | |
|       {
 | |
|          return ((!neg) ? std::numeric_limits<long double>::infinity()
 | |
|             : -std::numeric_limits<long double>::infinity());
 | |
|       }
 | |
|    }
 | |
| 
 | |
|    cpp_dec_float<Digits10, ExponentType, Allocator> xx(*this);
 | |
|    if(xx.isneg())
 | |
|       xx.negate();
 | |
| 
 | |
|    // Check if *this cpp_dec_float<Digits10, ExponentType, Allocator> is zero.
 | |
|    if(iszero() || (xx.compare(long_double_min()) < 0))
 | |
|    {
 | |
|       return static_cast<long double>(0.0);
 | |
|    }
 | |
| 
 | |
|    // Check if *this cpp_dec_float<Digits10, ExponentType, Allocator> exceeds the maximum of double.
 | |
|    if(xx.compare(long_double_max()) > 0)
 | |
|    {
 | |
|       return ((!neg) ? std::numeric_limits<long double>::infinity()
 | |
|          : -std::numeric_limits<long double>::infinity());
 | |
|    }
 | |
| 
 | |
|    std::stringstream ss;
 | |
| 
 | |
|    ss << str(std::numeric_limits<long double>::digits10 + (2 + 1), std::ios_base::scientific);
 | |
| 
 | |
|    long double ld;
 | |
|    ss >> ld;
 | |
| 
 | |
|    return ld;
 | |
| }
 | |
| 
 | |
| template <unsigned Digits10, class ExponentType, class Allocator>
 | |
| boost::long_long_type cpp_dec_float<Digits10, ExponentType, Allocator>::extract_signed_long_long() const
 | |
| {
 | |
|    // Extracts a signed long long from *this.
 | |
|    // If (x > maximum of long long) or (x < minimum of long long),
 | |
|    // then the maximum or minimum of long long is returned accordingly.
 | |
| 
 | |
|    if(exp < static_cast<ExponentType>(0))
 | |
|    {
 | |
|       return static_cast<boost::long_long_type>(0);
 | |
|    }
 | |
| 
 | |
|    const bool b_neg = isneg();
 | |
| 
 | |
|    boost::ulong_long_type val;
 | |
| 
 | |
|    if((!b_neg) && (compare(long_long_max()) > 0))
 | |
|    {
 | |
|       return (std::numeric_limits<boost::long_long_type>::max)();
 | |
|    }
 | |
|    else if(b_neg && (compare(long_long_min()) < 0))
 | |
|    {
 | |
|       return (std::numeric_limits<boost::long_long_type>::min)();
 | |
|    }
 | |
|    else
 | |
|    {
 | |
|       // Extract the data into an boost::ulong_long_type value.
 | |
|       cpp_dec_float<Digits10, ExponentType, Allocator> xn(extract_integer_part());
 | |
|       if(xn.isneg())
 | |
|          xn.negate();
 | |
| 
 | |
|       val = static_cast<boost::ulong_long_type>(xn.data[0]);
 | |
| 
 | |
|       const boost::int32_t imax = (std::min)(static_cast<boost::int32_t>(static_cast<boost::int32_t>(xn.exp) / cpp_dec_float_elem_digits10), static_cast<boost::int32_t>(cpp_dec_float_elem_number - static_cast<boost::int32_t>(1)));
 | |
| 
 | |
|       for(boost::int32_t i = static_cast<boost::int32_t>(1); i <= imax; i++)
 | |
|       {
 | |
|          val *= static_cast<boost::ulong_long_type>(cpp_dec_float_elem_mask);
 | |
|          val += static_cast<boost::ulong_long_type>(xn.data[i]);
 | |
|       }
 | |
|    }
 | |
| 
 | |
|    if (!b_neg)
 | |
|    {
 | |
|       return static_cast<boost::long_long_type>(val);
 | |
|    }
 | |
|    else
 | |
|    {
 | |
|       // This strange expression avoids a hardware trap in the corner case
 | |
|       // that val is the most negative value permitted in boost::long_long_type.
 | |
|       // See https://svn.boost.org/trac/boost/ticket/9740.
 | |
|       //
 | |
|       boost::long_long_type sval = static_cast<boost::long_long_type>(val - 1);
 | |
|       sval = -sval;
 | |
|       --sval;
 | |
|       return sval;
 | |
|    }
 | |
| }
 | |
| 
 | |
| template <unsigned Digits10, class ExponentType, class Allocator>
 | |
| boost::ulong_long_type cpp_dec_float<Digits10, ExponentType, Allocator>::extract_unsigned_long_long() const
 | |
| {
 | |
|    // Extracts an boost::ulong_long_type from *this.
 | |
|    // If x exceeds the maximum of boost::ulong_long_type,
 | |
|    // then the maximum of boost::ulong_long_type is returned.
 | |
|    // If x is negative, then the boost::ulong_long_type cast of
 | |
|    // the long long extracted value is returned.
 | |
| 
 | |
|    if(isneg())
 | |
|    {
 | |
|       return static_cast<boost::ulong_long_type>(extract_signed_long_long());
 | |
|    }
 | |
| 
 | |
|    if(exp < static_cast<ExponentType>(0))
 | |
|    {
 | |
|       return static_cast<boost::ulong_long_type>(0u);
 | |
|    }
 | |
| 
 | |
|    const cpp_dec_float<Digits10, ExponentType, Allocator> xn(extract_integer_part());
 | |
| 
 | |
|    boost::ulong_long_type val;
 | |
| 
 | |
|    if(xn.compare(ulong_long_max()) > 0)
 | |
|    {
 | |
|       return (std::numeric_limits<boost::ulong_long_type>::max)();
 | |
|    }
 | |
|    else
 | |
|    {
 | |
|       // Extract the data into an boost::ulong_long_type value.
 | |
|       val = static_cast<boost::ulong_long_type>(xn.data[0]);
 | |
| 
 | |
|       const boost::int32_t imax = (std::min)(static_cast<boost::int32_t>(static_cast<boost::int32_t>(xn.exp) / cpp_dec_float_elem_digits10), static_cast<boost::int32_t>(cpp_dec_float_elem_number - static_cast<boost::int32_t>(1)));
 | |
| 
 | |
|       for(boost::int32_t i = static_cast<boost::int32_t>(1); i <= imax; i++)
 | |
|       {
 | |
|          val *= static_cast<boost::ulong_long_type>(cpp_dec_float_elem_mask);
 | |
|          val += static_cast<boost::ulong_long_type>(xn.data[i]);
 | |
|       }
 | |
|    }
 | |
| 
 | |
|    return val;
 | |
| }
 | |
| 
 | |
| template <unsigned Digits10, class ExponentType, class Allocator>
 | |
| cpp_dec_float<Digits10, ExponentType, Allocator> cpp_dec_float<Digits10, ExponentType, Allocator>::extract_integer_part() const
 | |
| {
 | |
|    // Compute the signed integer part of x.
 | |
| 
 | |
|    if(!(isfinite)())
 | |
|    {
 | |
|       return *this;
 | |
|    }
 | |
| 
 | |
|    if(exp < static_cast<ExponentType>(0))
 | |
|    {
 | |
|       // The absolute value of the number is smaller than 1.
 | |
|       // Thus the integer part is zero.
 | |
|       return zero();
 | |
|    }
 | |
| 
 | |
|    // Truncate the digits from the decimal part, including guard digits
 | |
|    // that do not belong to the integer part.
 | |
| 
 | |
|    // Make a local copy.
 | |
|    cpp_dec_float<Digits10, ExponentType, Allocator> x = *this;
 | |
| 
 | |
|    // Clear out the decimal portion
 | |
|    const size_t first_clear = (static_cast<size_t>(x.exp) / static_cast<size_t>(cpp_dec_float_elem_digits10)) + 1u;
 | |
|    const size_t last_clear = static_cast<size_t>(cpp_dec_float_elem_number);
 | |
| 
 | |
|    if(first_clear < last_clear)
 | |
|       std::fill(x.data.begin() + first_clear, x.data.begin() + last_clear, static_cast<boost::uint32_t>(0u));
 | |
| 
 | |
|    return x;
 | |
| }
 | |
| 
 | |
| template <unsigned Digits10, class ExponentType, class Allocator>
 | |
| std::string cpp_dec_float<Digits10, ExponentType, Allocator>::str(boost::intmax_t number_of_digits, std::ios_base::fmtflags f) const
 | |
| {
 | |
|    if((this->isinf)())
 | |
|    {
 | |
|       if(this->isneg())
 | |
|          return "-inf";
 | |
|       else if(f & std::ios_base::showpos)
 | |
|          return "+inf";
 | |
|       else
 | |
|          return "inf";
 | |
|    }
 | |
|    else if((this->isnan)())
 | |
|    {
 | |
|       return "nan";
 | |
|    }
 | |
| 
 | |
|    std::string str;
 | |
|    boost::intmax_t org_digits(number_of_digits);
 | |
|    ExponentType my_exp = order();
 | |
| 
 | |
|    if(number_of_digits == 0)
 | |
|       number_of_digits = cpp_dec_float_total_digits10;
 | |
| 
 | |
|    if(f & std::ios_base::fixed)
 | |
|    {
 | |
|       number_of_digits += my_exp + 1;
 | |
|    }
 | |
|    else if(f & std::ios_base::scientific)
 | |
|       ++number_of_digits;
 | |
|    // Determine the number of elements needed to provide the requested digits from cpp_dec_float<Digits10, ExponentType, Allocator>.
 | |
|    const std::size_t number_of_elements = (std::min)(static_cast<std::size_t>((number_of_digits / static_cast<std::size_t>(cpp_dec_float_elem_digits10)) + 2u),
 | |
|       static_cast<std::size_t>(cpp_dec_float_elem_number));
 | |
| 
 | |
|    // Extract the remaining digits from cpp_dec_float<Digits10, ExponentType, Allocator> after the decimal point.
 | |
|    str = boost::lexical_cast<std::string>(data[0]);
 | |
| 
 | |
|    // Extract all of the digits from cpp_dec_float<Digits10, ExponentType, Allocator>, beginning with the first data element.
 | |
|    for(std::size_t i = static_cast<std::size_t>(1u); i < number_of_elements; i++)
 | |
|    {
 | |
|       std::stringstream ss;
 | |
| 
 | |
|       ss << std::setw(static_cast<std::streamsize>(cpp_dec_float_elem_digits10))
 | |
|          << std::setfill(static_cast<char>('0'))
 | |
|          << data[i];
 | |
| 
 | |
|       str += ss.str();
 | |
|    }
 | |
| 
 | |
|    bool have_leading_zeros = false;
 | |
| 
 | |
|    if(number_of_digits == 0)
 | |
|    {
 | |
|       // We only get here if the output format is "fixed" and we just need to
 | |
|       // round the first non-zero digit.
 | |
|       number_of_digits -= my_exp + 1; // reset to original value
 | |
|       str.insert(static_cast<std::string::size_type>(0), std::string::size_type(number_of_digits), '0');
 | |
|       have_leading_zeros = true;
 | |
|    }
 | |
| 
 | |
|    if(number_of_digits < 0)
 | |
|    {
 | |
|       str = "0";
 | |
|       if(isneg())
 | |
|          str.insert(static_cast<std::string::size_type>(0), 1, '-');
 | |
|       boost::multiprecision::detail::format_float_string(str, 0, number_of_digits - my_exp - 1, f, this->iszero());
 | |
|       return str;
 | |
|    }
 | |
|    else
 | |
|    {
 | |
|       // Cut the output to the size of the precision.
 | |
|       if(str.length() > static_cast<std::string::size_type>(number_of_digits))
 | |
|       {
 | |
|          // Get the digit after the last needed digit for rounding
 | |
|          const boost::uint32_t round = static_cast<boost::uint32_t>(static_cast<boost::uint32_t>(str[static_cast<std::string::size_type>(number_of_digits)]) - static_cast<boost::uint32_t>('0'));
 | |
| 
 | |
|          bool need_round_up = round >= 5u;
 | |
| 
 | |
|          if(round == 5u)
 | |
|          {
 | |
|             const boost::uint32_t ix = static_cast<boost::uint32_t>(static_cast<boost::uint32_t>(str[static_cast<std::string::size_type>(number_of_digits - 1)]) - static_cast<boost::uint32_t>('0'));
 | |
|             if((ix & 1u) == 0)
 | |
|             {
 | |
|                // We have an even digit followed by a 5, so we might not actually need to round up
 | |
|                // if all the remaining digits are zero:
 | |
|                if(str.find_first_not_of('0', static_cast<std::string::size_type>(number_of_digits + 1)) == std::string::npos)
 | |
|                {
 | |
|                   bool all_zeros = true;
 | |
|                   // No none-zero trailing digits in the string, now check whatever parts we didn't convert to the string:
 | |
|                   for(std::size_t i = number_of_elements; i < data.size(); i++)
 | |
|                   {
 | |
|                      if(data[i])
 | |
|                      {
 | |
|                         all_zeros = false;
 | |
|                         break;
 | |
|                      }
 | |
|                   }
 | |
|                   if(all_zeros)
 | |
|                      need_round_up = false; // tie break - round to even.
 | |
|                }
 | |
|             }
 | |
|          }
 | |
| 
 | |
|          // Truncate the string
 | |
|          str.erase(static_cast<std::string::size_type>(number_of_digits));
 | |
| 
 | |
|          if(need_round_up)
 | |
|          {
 | |
|             std::size_t ix = static_cast<std::size_t>(str.length() - 1u);
 | |
| 
 | |
|             // Every trailing 9 must be rounded up
 | |
|             while(ix && (static_cast<boost::int32_t>(str.at(ix)) - static_cast<boost::int32_t>('0') == static_cast<boost::int32_t>(9)))
 | |
|             {
 | |
|                str.at(ix) = static_cast<char>('0');
 | |
|                --ix;
 | |
|             }
 | |
| 
 | |
|             if(!ix)
 | |
|             {
 | |
|                // There were nothing but trailing nines.
 | |
|                if(static_cast<boost::int32_t>(static_cast<boost::int32_t>(str.at(ix)) - static_cast<boost::int32_t>(0x30)) == static_cast<boost::int32_t>(9))
 | |
|                {
 | |
|                   // Increment up to the next order and adjust exponent.
 | |
|                   str.at(ix) = static_cast<char>('1');
 | |
|                   ++my_exp;
 | |
|                }
 | |
|                else
 | |
|                {
 | |
|                   // Round up this digit.
 | |
|                   ++str.at(ix);
 | |
|                }
 | |
|             }
 | |
|             else
 | |
|             {
 | |
|                // Round up the last digit.
 | |
|                ++str[ix];
 | |
|             }
 | |
|          }
 | |
|       }
 | |
|    }
 | |
| 
 | |
|    if(have_leading_zeros)
 | |
|    {
 | |
|       // We need to take the zeros back out again, and correct the exponent
 | |
|       // if we rounded up:
 | |
|       if(str[std::string::size_type(number_of_digits - 1)] != '0')
 | |
|       {
 | |
|          ++my_exp;
 | |
|          str.erase(0, std::string::size_type(number_of_digits - 1));
 | |
|       }
 | |
|       else
 | |
|          str.erase(0, std::string::size_type(number_of_digits));
 | |
|    }
 | |
| 
 | |
|    if(isneg())
 | |
|       str.insert(static_cast<std::string::size_type>(0), 1, '-');
 | |
| 
 | |
|    boost::multiprecision::detail::format_float_string(str, my_exp, org_digits, f, this->iszero());
 | |
|    return str;
 | |
| }
 | |
| 
 | |
| template <unsigned Digits10, class ExponentType, class Allocator>
 | |
| bool cpp_dec_float<Digits10, ExponentType, Allocator>::rd_string(const char* const s)
 | |
| {
 | |
| #ifndef BOOST_NO_EXCEPTIONS
 | |
|    try{
 | |
| #endif
 | |
| 
 | |
|    std::string str(s);
 | |
| 
 | |
|    // TBD: Using several regular expressions may significantly reduce
 | |
|    // the code complexity (and perhaps the run-time) of rd_string().
 | |
| 
 | |
|    // Get a possible exponent and remove it.
 | |
|    exp = static_cast<ExponentType>(0);
 | |
| 
 | |
|    std::size_t pos;
 | |
| 
 | |
|    if( ((pos = str.find('e')) != std::string::npos)
 | |
|       || ((pos = str.find('E')) != std::string::npos)
 | |
|       )
 | |
|    {
 | |
|       // Remove the exponent part from the string.
 | |
|       exp = boost::lexical_cast<ExponentType>(static_cast<const char*>(str.c_str() + (pos + 1u)));
 | |
|       str = str.substr(static_cast<std::size_t>(0u), pos);
 | |
|    }
 | |
| 
 | |
|    // Get a possible +/- sign and remove it.
 | |
|    neg = false;
 | |
| 
 | |
|    if(str.size())
 | |
|    {
 | |
|       if(str[0] == '-')
 | |
|       {
 | |
|          neg = true;
 | |
|          str.erase(0, 1);
 | |
|       }
 | |
|       else if(str[0] == '+')
 | |
|       {
 | |
|          str.erase(0, 1);
 | |
|       }
 | |
|    }
 | |
|    //
 | |
|    // Special cases for infinities and NaN's:
 | |
|    //
 | |
|    if((str == "inf") || (str == "INF") || (str == "infinity") || (str == "INFINITY"))
 | |
|    {
 | |
|       if(neg)
 | |
|       {
 | |
|          *this = this->inf();
 | |
|          this->negate();
 | |
|       }
 | |
|       else
 | |
|          *this = this->inf();
 | |
|       return true;
 | |
|    }
 | |
|    if((str.size() >= 3) && ((str.substr(0, 3) == "nan") || (str.substr(0, 3) == "NAN") || (str.substr(0, 3) == "NaN")))
 | |
|    {
 | |
|       *this = this->nan();
 | |
|       return true;
 | |
|    }
 | |
| 
 | |
|    // Remove the leading zeros for all input types.
 | |
|    const std::string::iterator fwd_it_leading_zero = std::find_if(str.begin(), str.end(), char_is_nonzero_predicate);
 | |
| 
 | |
|    if(fwd_it_leading_zero != str.begin())
 | |
|    {
 | |
|       if(fwd_it_leading_zero == str.end())
 | |
|       {
 | |
|          // The string contains nothing but leading zeros.
 | |
|          // This string represents zero.
 | |
|          operator=(zero());
 | |
|          return true;
 | |
|       }
 | |
|       else
 | |
|       {
 | |
|          str.erase(str.begin(), fwd_it_leading_zero);
 | |
|       }
 | |
|    }
 | |
| 
 | |
|    // Put the input string into the standard cpp_dec_float<Digits10, ExponentType, Allocator> input form
 | |
|    // aaa.bbbbE+/-n, where aaa has 1...cpp_dec_float_elem_digits10, bbbb has an
 | |
|    // even multiple of cpp_dec_float_elem_digits10 which are possibly zero padded
 | |
|    // on the right-end, and n is a signed 64-bit integer which is an
 | |
|    // even multiple of cpp_dec_float_elem_digits10.
 | |
| 
 | |
|    // Find a possible decimal point.
 | |
|    pos = str.find(static_cast<char>('.'));
 | |
| 
 | |
|    if(pos != std::string::npos)
 | |
|    {
 | |
|       // Remove all trailing insignificant zeros.
 | |
|       const std::string::const_reverse_iterator rit_non_zero = std::find_if(str.rbegin(), str.rend(), char_is_nonzero_predicate);
 | |
| 
 | |
|       if(rit_non_zero != static_cast<std::string::const_reverse_iterator>(str.rbegin()))
 | |
|       {
 | |
|          const std::string::size_type ofs = str.length() - std::distance<std::string::const_reverse_iterator>(str.rbegin(), rit_non_zero);
 | |
|          str.erase(str.begin() + ofs, str.end());
 | |
|       }
 | |
| 
 | |
|       // Check if the input is identically zero.
 | |
|       if(str == std::string("."))
 | |
|       {
 | |
|          operator=(zero());
 | |
|          return true;
 | |
|       }
 | |
| 
 | |
|       // Remove leading significant zeros just after the decimal point
 | |
|       // and adjust the exponent accordingly.
 | |
|       // Note that the while-loop operates only on strings of the form ".000abcd..."
 | |
|       // and peels away the zeros just after the decimal point.
 | |
|       if(str.at(static_cast<std::size_t>(0u)) == static_cast<char>('.'))
 | |
|       {
 | |
|          const std::string::iterator it_non_zero = std::find_if(str.begin() + 1u, str.end(), char_is_nonzero_predicate);
 | |
| 
 | |
|          std::size_t delta_exp = static_cast<std::size_t>(0u);
 | |
| 
 | |
|          if(str.at(static_cast<std::size_t>(1u)) == static_cast<char>('0'))
 | |
|          {
 | |
|             delta_exp = std::distance<std::string::const_iterator>(str.begin() + 1u, it_non_zero);
 | |
|          }
 | |
| 
 | |
|          // Bring one single digit into the mantissa and adjust the exponent accordingly.
 | |
|          str.erase(str.begin(), it_non_zero);
 | |
|          str.insert(static_cast<std::string::size_type>(1u), ".");
 | |
|          exp -= static_cast<ExponentType>(delta_exp + 1u);
 | |
|       }
 | |
|    }
 | |
|    else
 | |
|    {
 | |
|       // Input string has no decimal point: Append decimal point.
 | |
|       str.append(".");
 | |
|    }
 | |
| 
 | |
|    // Shift the decimal point such that the exponent is an even multiple of cpp_dec_float_elem_digits10.
 | |
|    std::size_t n_shift = static_cast<std::size_t>(0u);
 | |
|    const std::size_t n_exp_rem = static_cast<std::size_t>(exp % static_cast<ExponentType>(cpp_dec_float_elem_digits10));
 | |
| 
 | |
|    if((exp % static_cast<ExponentType>(cpp_dec_float_elem_digits10)) != static_cast<ExponentType>(0))
 | |
|    {
 | |
|       n_shift = ((exp < static_cast<ExponentType>(0))
 | |
|          ? static_cast<std::size_t>(n_exp_rem + static_cast<std::size_t>(cpp_dec_float_elem_digits10))
 | |
|          : static_cast<std::size_t>(n_exp_rem));
 | |
|    }
 | |
| 
 | |
|    // Make sure that there are enough digits for the decimal point shift.
 | |
|    pos = str.find(static_cast<char>('.'));
 | |
| 
 | |
|    std::size_t pos_plus_one = static_cast<std::size_t>(pos + 1u);
 | |
| 
 | |
|    if((str.length() - pos_plus_one) < n_shift)
 | |
|    {
 | |
|       const std::size_t sz = static_cast<std::size_t>(n_shift - (str.length() - pos_plus_one));
 | |
| 
 | |
|       str.append(std::string(sz, static_cast<char>('0')));
 | |
|    }
 | |
| 
 | |
|    // Do the decimal point shift.
 | |
|    if(n_shift != static_cast<std::size_t>(0u))
 | |
|    {
 | |
|       str.insert(static_cast<std::string::size_type>(pos_plus_one + n_shift), ".");
 | |
| 
 | |
|       str.erase(pos, static_cast<std::string::size_type>(1u));
 | |
| 
 | |
|       exp -= static_cast<ExponentType>(n_shift);
 | |
|    }
 | |
| 
 | |
|    // Cut the size of the mantissa to <= cpp_dec_float_elem_digits10.
 | |
|    pos = str.find(static_cast<char>('.'));
 | |
|    pos_plus_one = static_cast<std::size_t>(pos + 1u);
 | |
| 
 | |
|    if(pos > static_cast<std::size_t>(cpp_dec_float_elem_digits10))
 | |
|    {
 | |
|       const boost::int32_t n_pos = static_cast<boost::int32_t>(pos);
 | |
|       const boost::int32_t n_rem_is_zero = ((static_cast<boost::int32_t>(n_pos % cpp_dec_float_elem_digits10) == static_cast<boost::int32_t>(0)) ? static_cast<boost::int32_t>(1) : static_cast<boost::int32_t>(0));
 | |
|       const boost::int32_t n = static_cast<boost::int32_t>(static_cast<boost::int32_t>(n_pos / cpp_dec_float_elem_digits10) - n_rem_is_zero);
 | |
| 
 | |
|       str.insert(static_cast<std::size_t>(static_cast<boost::int32_t>(n_pos - static_cast<boost::int32_t>(n * cpp_dec_float_elem_digits10))), ".");
 | |
| 
 | |
|       str.erase(pos_plus_one, static_cast<std::size_t>(1u));
 | |
| 
 | |
|       exp += static_cast<ExponentType>(static_cast<ExponentType>(n) * static_cast<ExponentType>(cpp_dec_float_elem_digits10));
 | |
|    }
 | |
| 
 | |
|    // Pad the decimal part such that its value is an even
 | |
|    // multiple of cpp_dec_float_elem_digits10.
 | |
|    pos = str.find(static_cast<char>('.'));
 | |
|    pos_plus_one = static_cast<std::size_t>(pos + 1u);
 | |
| 
 | |
|    const boost::int32_t n_dec = static_cast<boost::int32_t>(static_cast<boost::int32_t>(str.length() - 1u) - static_cast<boost::int32_t>(pos));
 | |
|    const boost::int32_t n_rem = static_cast<boost::int32_t>(n_dec % cpp_dec_float_elem_digits10);
 | |
| 
 | |
|    boost::int32_t n_cnt = ((n_rem != static_cast<boost::int32_t>(0))
 | |
|                              ? static_cast<boost::int32_t>(cpp_dec_float_elem_digits10 - n_rem)
 | |
|                              : static_cast<boost::int32_t>(0));
 | |
| 
 | |
|    if(n_cnt != static_cast<boost::int32_t>(0))
 | |
|    {
 | |
|       str.append(static_cast<std::size_t>(n_cnt), static_cast<char>('0'));
 | |
|    }
 | |
| 
 | |
|    // Truncate decimal part if it is too long.
 | |
|    const std::size_t max_dec = static_cast<std::size_t>((cpp_dec_float_elem_number - 1) * cpp_dec_float_elem_digits10);
 | |
| 
 | |
|    if(static_cast<std::size_t>(str.length() - pos) > max_dec)
 | |
|    {
 | |
|       str = str.substr(static_cast<std::size_t>(0u),
 | |
|          static_cast<std::size_t>(pos_plus_one + max_dec));
 | |
|    }
 | |
| 
 | |
|    // Now the input string has the standard cpp_dec_float<Digits10, ExponentType, Allocator> input form.
 | |
|    // (See the comment above.)
 | |
| 
 | |
|    // Set all the data elements to 0.
 | |
|    std::fill(data.begin(), data.end(), static_cast<boost::uint32_t>(0u));
 | |
| 
 | |
|    // Extract the data.
 | |
| 
 | |
|    // First get the digits to the left of the decimal point...
 | |
|    data[0u] = boost::lexical_cast<boost::uint32_t>(str.substr(static_cast<std::size_t>(0u), pos));
 | |
| 
 | |
|    // ...then get the remaining digits to the right of the decimal point.
 | |
|    const std::string::size_type i_end = ((str.length() - pos_plus_one) / static_cast<std::string::size_type>(cpp_dec_float_elem_digits10));
 | |
| 
 | |
|    for(std::string::size_type i = static_cast<std::string::size_type>(0u); i < i_end; i++)
 | |
|    {
 | |
|       const std::string::const_iterator it = str.begin()
 | |
|          + pos_plus_one
 | |
|          + (i * static_cast<std::string::size_type>(cpp_dec_float_elem_digits10));
 | |
| 
 | |
|       data[i + 1u] = boost::lexical_cast<boost::uint32_t>(std::string(it, it + static_cast<std::string::size_type>(cpp_dec_float_elem_digits10)));
 | |
|    }
 | |
| 
 | |
|    // Check for overflow...
 | |
|    if(exp > cpp_dec_float_max_exp10)
 | |
|    {
 | |
|       const bool b_result_is_neg = neg;
 | |
| 
 | |
|       *this = inf();
 | |
|       if(b_result_is_neg)
 | |
|          negate();
 | |
|    }
 | |
| 
 | |
|    // ...and check for underflow.
 | |
|    if(exp <= cpp_dec_float_min_exp10)
 | |
|    {
 | |
|       if(exp == cpp_dec_float_min_exp10)
 | |
|       {
 | |
|          // Check for identity with the minimum value.
 | |
|          cpp_dec_float<Digits10, ExponentType, Allocator> test = *this;
 | |
| 
 | |
|          test.exp = static_cast<ExponentType>(0);
 | |
| 
 | |
|          if(test.isone())
 | |
|          {
 | |
|             *this = zero();
 | |
|          }
 | |
|       }
 | |
|       else
 | |
|       {
 | |
|          *this = zero();
 | |
|       }
 | |
|    }
 | |
| 
 | |
| #ifndef BOOST_NO_EXCEPTIONS
 | |
|    }
 | |
|    catch(const bad_lexical_cast&)
 | |
|    {
 | |
|       // Rethrow with better error message:
 | |
|       std::string msg = "Unable to parse the string \"";
 | |
|       msg += s;
 | |
|       msg += "\" as a floating point value.";
 | |
|       throw std::runtime_error(msg);
 | |
|    }
 | |
| #endif
 | |
|    return true;
 | |
| }
 | |
| 
 | |
| template <unsigned Digits10, class ExponentType, class Allocator>
 | |
| cpp_dec_float<Digits10, ExponentType, Allocator>::cpp_dec_float(const double mantissa, const ExponentType exponent)
 | |
|  : data (),
 | |
|    exp  (static_cast<ExponentType>(0)),
 | |
|    neg  (false),
 | |
|    fpclass (cpp_dec_float_finite),
 | |
|    prec_elem(cpp_dec_float_elem_number)
 | |
| {
 | |
|    // Create *this cpp_dec_float<Digits10, ExponentType, Allocator> from a given mantissa and exponent.
 | |
|    // Note: This constructor does not maintain the full precision of double.
 | |
| 
 | |
|    const bool mantissa_is_iszero = (::fabs(mantissa) < ((std::numeric_limits<double>::min)() * (1.0 + std::numeric_limits<double>::epsilon())));
 | |
| 
 | |
|    if(mantissa_is_iszero)
 | |
|    {
 | |
|       std::fill(data.begin(), data.end(), static_cast<boost::uint32_t>(0u));
 | |
|       return;
 | |
|    }
 | |
| 
 | |
|    const bool b_neg = (mantissa < 0.0);
 | |
| 
 | |
|    double d = ((!b_neg) ? mantissa : -mantissa);
 | |
|    ExponentType e = exponent;
 | |
| 
 | |
|    while(d > 10.0) { d /= 10.0; ++e; }
 | |
|    while(d < 1.0) { d *= 10.0; --e; }
 | |
| 
 | |
|    boost::int32_t shift = static_cast<boost::int32_t>(e % static_cast<boost::int32_t>(cpp_dec_float_elem_digits10));
 | |
| 
 | |
|    while(static_cast<boost::int32_t>(shift-- % cpp_dec_float_elem_digits10) != static_cast<boost::int32_t>(0))
 | |
|    {
 | |
|       d *= 10.0;
 | |
|       --e;
 | |
|    }
 | |
| 
 | |
|    exp = e;
 | |
|    neg = b_neg;
 | |
| 
 | |
|    std::fill(data.begin(), data.end(), static_cast<boost::uint32_t>(0u));
 | |
| 
 | |
|    static const boost::int32_t digit_ratio = static_cast<boost::int32_t>(static_cast<boost::int32_t>(std::numeric_limits<double>::digits10) / static_cast<boost::int32_t>(cpp_dec_float_elem_digits10));
 | |
|    static const boost::int32_t digit_loops = static_cast<boost::int32_t>(digit_ratio + static_cast<boost::int32_t>(2));
 | |
| 
 | |
|    for(boost::int32_t i = static_cast<boost::int32_t>(0); i < digit_loops; i++)
 | |
|    {
 | |
|       boost::uint32_t n = static_cast<boost::uint32_t>(static_cast<boost::uint64_t>(d));
 | |
|       data[i] = static_cast<boost::uint32_t>(n);
 | |
|       d -= static_cast<double>(n);
 | |
|       d *= static_cast<double>(cpp_dec_float_elem_mask);
 | |
|    }
 | |
| }
 | |
| 
 | |
| template <unsigned Digits10, class ExponentType, class Allocator>
 | |
| cpp_dec_float<Digits10, ExponentType, Allocator>& cpp_dec_float<Digits10, ExponentType, Allocator>::operator= (long double a)
 | |
| {
 | |
|    // Christopher Kormanyos's original code used a cast to boost::long_long_type here, but that fails
 | |
|    // when long double has more digits than a boost::long_long_type.
 | |
|    using std::frexp;
 | |
|    using std::ldexp;
 | |
|    using std::floor;
 | |
| 
 | |
|    if(a == 0)
 | |
|       return *this = zero();
 | |
| 
 | |
|    if(a == 1)
 | |
|       return *this = one();
 | |
| 
 | |
|    if((boost::math::isinf)(a))
 | |
|    {
 | |
|       *this = inf();
 | |
|       if(a < 0)
 | |
|          this->negate();
 | |
|       return *this;
 | |
|    }
 | |
| 
 | |
|    if((boost::math::isnan)(a))
 | |
|       return *this = nan();
 | |
| 
 | |
|    int e;
 | |
|    long double f, term;
 | |
|    *this = zero();
 | |
| 
 | |
|    f = frexp(a, &e);
 | |
|    // See https://svn.boost.org/trac/boost/ticket/10924 for an example of why this may go wrong:
 | |
|    BOOST_ASSERT((boost::math::isfinite)(f));
 | |
| 
 | |
|    static const int shift = std::numeric_limits<int>::digits - 1;
 | |
| 
 | |
|    while(f)
 | |
|    {
 | |
|       // extract int sized bits from f:
 | |
|       f = ldexp(f, shift);
 | |
|       BOOST_ASSERT((boost::math::isfinite)(f));
 | |
|       term = floor(f);
 | |
|       e -= shift;
 | |
|       *this *= pow2(shift);
 | |
|       if(term > 0)
 | |
|          add_unsigned_long_long(static_cast<unsigned>(term));
 | |
|       else
 | |
|          sub_unsigned_long_long(static_cast<unsigned>(-term));
 | |
|       f -= term;
 | |
|    }
 | |
| 
 | |
|    if(e != 0)
 | |
|       *this *= pow2(e);
 | |
| 
 | |
|    return *this;
 | |
| }
 | |
| 
 | |
| template <unsigned Digits10, class ExponentType, class Allocator>
 | |
| void cpp_dec_float<Digits10, ExponentType, Allocator>::from_unsigned_long_long(const boost::ulong_long_type u)
 | |
| {
 | |
|    std::fill(data.begin(), data.end(), static_cast<boost::uint32_t>(0u));
 | |
| 
 | |
|    exp = static_cast<ExponentType>(0);
 | |
|    neg = false;
 | |
|    fpclass = cpp_dec_float_finite;
 | |
|    prec_elem = cpp_dec_float_elem_number;
 | |
| 
 | |
|    std::size_t i =static_cast<std::size_t>(0u);
 | |
| 
 | |
|    boost::ulong_long_type uu = u;
 | |
| 
 | |
|    boost::uint32_t temp[(std::numeric_limits<boost::ulong_long_type>::digits10 / static_cast<int>(cpp_dec_float_elem_digits10)) + 3] = { static_cast<boost::uint32_t>(0u) };
 | |
| 
 | |
|    while(uu != static_cast<boost::ulong_long_type>(0u))
 | |
|    {
 | |
|       temp[i] = static_cast<boost::uint32_t>(uu % static_cast<boost::ulong_long_type>(cpp_dec_float_elem_mask));
 | |
|       uu = static_cast<boost::ulong_long_type>(uu / static_cast<boost::ulong_long_type>(cpp_dec_float_elem_mask));
 | |
|       ++i;
 | |
|    }
 | |
| 
 | |
|    if(i > static_cast<std::size_t>(1u))
 | |
|    {
 | |
|       exp += static_cast<ExponentType>((i - 1u) * static_cast<std::size_t>(cpp_dec_float_elem_digits10));
 | |
|    }
 | |
| 
 | |
|    std::reverse(temp, temp + i);
 | |
|    std::copy(temp, temp + (std::min)(i, static_cast<std::size_t>(cpp_dec_float_elem_number)), data.begin());
 | |
| }
 | |
| 
 | |
| template <unsigned Digits10, class ExponentType, class Allocator>
 | |
| boost::uint32_t cpp_dec_float<Digits10, ExponentType, Allocator>::mul_loop_uv(boost::uint32_t* const u, const boost::uint32_t* const v, const boost::int32_t p)
 | |
| {
 | |
|    //
 | |
|    // There is a limit on how many limbs this algorithm can handle without dropping digits
 | |
|    // due to overflow in the carry, it is:
 | |
|    //
 | |
|    // FLOOR( (2^64 - 1) / (10^8 * 10^8) ) == 1844
 | |
|    //
 | |
|    BOOST_STATIC_ASSERT_MSG(cpp_dec_float_elem_number < 1800, "Too many limbs in the data type for the multiplication algorithm - unsupported precision in cpp_dec_float.");
 | |
| 
 | |
|    boost::uint64_t carry = static_cast<boost::uint64_t>(0u);
 | |
| 
 | |
|    for(boost::int32_t j = static_cast<boost::int32_t>(p - 1u); j >= static_cast<boost::int32_t>(0); j--)
 | |
|    {
 | |
|      boost::uint64_t sum = carry;
 | |
| 
 | |
|      for(boost::int32_t i = j; i >= static_cast<boost::int32_t>(0); i--)
 | |
|      {
 | |
|        sum += static_cast<boost::uint64_t>(u[j - i] * static_cast<boost::uint64_t>(v[i]));
 | |
|      }
 | |
| 
 | |
|      u[j] = static_cast<boost::uint32_t>(sum % static_cast<boost::uint32_t>(cpp_dec_float_elem_mask));
 | |
|      carry = static_cast<boost::uint64_t>(sum / static_cast<boost::uint32_t>(cpp_dec_float_elem_mask));
 | |
|    }
 | |
| 
 | |
|    return static_cast<boost::uint32_t>(carry);
 | |
| }
 | |
| 
 | |
| template <unsigned Digits10, class ExponentType, class Allocator>
 | |
| boost::uint32_t cpp_dec_float<Digits10, ExponentType, Allocator>::mul_loop_n(boost::uint32_t* const u, boost::uint32_t n, const boost::int32_t p)
 | |
| {
 | |
|    boost::uint64_t carry = static_cast<boost::uint64_t>(0u);
 | |
| 
 | |
|    // Multiplication loop.
 | |
|    for(boost::int32_t j = p - 1; j >= static_cast<boost::int32_t>(0); j--)
 | |
|    {
 | |
|       const boost::uint64_t t = static_cast<boost::uint64_t>(carry + static_cast<boost::uint64_t>(u[j] * static_cast<boost::uint64_t>(n)));
 | |
|       carry = static_cast<boost::uint64_t>(t / static_cast<boost::uint32_t>(cpp_dec_float_elem_mask));
 | |
|       u[j] = static_cast<boost::uint32_t>(t - static_cast<boost::uint64_t>(static_cast<boost::uint32_t>(cpp_dec_float_elem_mask) * static_cast<boost::uint64_t>(carry)));
 | |
|    }
 | |
| 
 | |
|    return static_cast<boost::uint32_t>(carry);
 | |
| }
 | |
| 
 | |
| template <unsigned Digits10, class ExponentType, class Allocator>
 | |
| boost::uint32_t cpp_dec_float<Digits10, ExponentType, Allocator>::div_loop_n(boost::uint32_t* const u, boost::uint32_t n, const boost::int32_t p)
 | |
| {
 | |
|    boost::uint64_t prev = static_cast<boost::uint64_t>(0u);
 | |
| 
 | |
|    for(boost::int32_t j = static_cast<boost::int32_t>(0); j < p; j++)
 | |
|    {
 | |
|       const boost::uint64_t t = static_cast<boost::uint64_t>(u[j] + static_cast<boost::uint64_t>(prev * static_cast<boost::uint32_t>(cpp_dec_float_elem_mask)));
 | |
|       u[j] = static_cast<boost::uint32_t>(t / n);
 | |
|       prev = static_cast<boost::uint64_t>(t - static_cast<boost::uint64_t>(n * static_cast<boost::uint64_t>(u[j])));
 | |
|    }
 | |
| 
 | |
|    return static_cast<boost::uint32_t>(prev);
 | |
| }
 | |
| 
 | |
| template <unsigned Digits10, class ExponentType, class Allocator>
 | |
| cpp_dec_float<Digits10, ExponentType, Allocator> cpp_dec_float<Digits10, ExponentType, Allocator>::pow2(const boost::long_long_type p)
 | |
| {
 | |
|    // Create a static const table of p^2 for -128 < p < +128.
 | |
|    // Note: The size of this table must be odd-numbered and
 | |
|    // symmetric about 0.
 | |
|    init.do_nothing();
 | |
|    static const boost::array<cpp_dec_float<Digits10, ExponentType, Allocator>, 255u> p2_data =
 | |
|    {{
 | |
|        cpp_dec_float("5.877471754111437539843682686111228389093327783860437607543758531392086297273635864257812500000000000e-39"),
 | |
|        cpp_dec_float("1.175494350822287507968736537222245677818665556772087521508751706278417259454727172851562500000000000e-38"),
 | |
|        cpp_dec_float("2.350988701644575015937473074444491355637331113544175043017503412556834518909454345703125000000000000e-38"),
 | |
|        cpp_dec_float("4.701977403289150031874946148888982711274662227088350086035006825113669037818908691406250000000000000e-38"),
 | |
|        cpp_dec_float("9.403954806578300063749892297777965422549324454176700172070013650227338075637817382812500000000000000e-38"),
 | |
|        cpp_dec_float("1.880790961315660012749978459555593084509864890835340034414002730045467615127563476562500000000000000e-37"),
 | |
|        cpp_dec_float("3.761581922631320025499956919111186169019729781670680068828005460090935230255126953125000000000000000e-37"),
 | |
|        cpp_dec_float("7.523163845262640050999913838222372338039459563341360137656010920181870460510253906250000000000000000e-37"),
 | |
|        cpp_dec_float("1.504632769052528010199982767644474467607891912668272027531202184036374092102050781250000000000000000e-36"),
 | |
|        cpp_dec_float("3.009265538105056020399965535288948935215783825336544055062404368072748184204101562500000000000000000e-36"),
 | |
|        cpp_dec_float("6.018531076210112040799931070577897870431567650673088110124808736145496368408203125000000000000000000e-36"),
 | |
|        cpp_dec_float("1.203706215242022408159986214115579574086313530134617622024961747229099273681640625000000000000000000e-35"),
 | |
|        cpp_dec_float("2.407412430484044816319972428231159148172627060269235244049923494458198547363281250000000000000000000e-35"),
 | |
|        cpp_dec_float("4.814824860968089632639944856462318296345254120538470488099846988916397094726562500000000000000000000e-35"),
 | |
|        cpp_dec_float("9.629649721936179265279889712924636592690508241076940976199693977832794189453125000000000000000000000e-35"),
 | |
|        cpp_dec_float("1.925929944387235853055977942584927318538101648215388195239938795566558837890625000000000000000000000e-34"),
 | |
|        cpp_dec_float("3.851859888774471706111955885169854637076203296430776390479877591133117675781250000000000000000000000e-34"),
 | |
|        cpp_dec_float("7.703719777548943412223911770339709274152406592861552780959755182266235351562500000000000000000000000e-34"),
 | |
|        cpp_dec_float("1.540743955509788682444782354067941854830481318572310556191951036453247070312500000000000000000000000e-33"),
 | |
|        cpp_dec_float("3.081487911019577364889564708135883709660962637144621112383902072906494140625000000000000000000000000e-33"),
 | |
|        cpp_dec_float("6.162975822039154729779129416271767419321925274289242224767804145812988281250000000000000000000000000e-33"),
 | |
|        cpp_dec_float("1.232595164407830945955825883254353483864385054857848444953560829162597656250000000000000000000000000e-32"),
 | |
|        cpp_dec_float("2.465190328815661891911651766508706967728770109715696889907121658325195312500000000000000000000000000e-32"),
 | |
|        cpp_dec_float("4.930380657631323783823303533017413935457540219431393779814243316650390625000000000000000000000000000e-32"),
 | |
|        cpp_dec_float("9.860761315262647567646607066034827870915080438862787559628486633300781250000000000000000000000000000e-32"),
 | |
|        cpp_dec_float("1.972152263052529513529321413206965574183016087772557511925697326660156250000000000000000000000000000e-31"),
 | |
|        cpp_dec_float("3.944304526105059027058642826413931148366032175545115023851394653320312500000000000000000000000000000e-31"),
 | |
|        cpp_dec_float("7.888609052210118054117285652827862296732064351090230047702789306640625000000000000000000000000000000e-31"),
 | |
|        cpp_dec_float("1.577721810442023610823457130565572459346412870218046009540557861328125000000000000000000000000000000e-30"),
 | |
|        cpp_dec_float("3.155443620884047221646914261131144918692825740436092019081115722656250000000000000000000000000000000e-30"),
 | |
|        cpp_dec_float("6.310887241768094443293828522262289837385651480872184038162231445312500000000000000000000000000000000e-30"),
 | |
|        cpp_dec_float("1.262177448353618888658765704452457967477130296174436807632446289062500000000000000000000000000000000e-29"),
 | |
|        cpp_dec_float("2.524354896707237777317531408904915934954260592348873615264892578125000000000000000000000000000000000e-29"),
 | |
|        cpp_dec_float("5.048709793414475554635062817809831869908521184697747230529785156250000000000000000000000000000000000e-29"),
 | |
|        cpp_dec_float("1.009741958682895110927012563561966373981704236939549446105957031250000000000000000000000000000000000e-28"),
 | |
|        cpp_dec_float("2.019483917365790221854025127123932747963408473879098892211914062500000000000000000000000000000000000e-28"),
 | |
|        cpp_dec_float("4.038967834731580443708050254247865495926816947758197784423828125000000000000000000000000000000000000e-28"),
 | |
|        cpp_dec_float("8.077935669463160887416100508495730991853633895516395568847656250000000000000000000000000000000000000e-28"),
 | |
|        cpp_dec_float("1.615587133892632177483220101699146198370726779103279113769531250000000000000000000000000000000000000e-27"),
 | |
|        cpp_dec_float("3.231174267785264354966440203398292396741453558206558227539062500000000000000000000000000000000000000e-27"),
 | |
|        cpp_dec_float("6.462348535570528709932880406796584793482907116413116455078125000000000000000000000000000000000000000e-27"),
 | |
|        cpp_dec_float("1.292469707114105741986576081359316958696581423282623291015625000000000000000000000000000000000000000e-26"),
 | |
|        cpp_dec_float("2.584939414228211483973152162718633917393162846565246582031250000000000000000000000000000000000000000e-26"),
 | |
|        cpp_dec_float("5.169878828456422967946304325437267834786325693130493164062500000000000000000000000000000000000000000e-26"),
 | |
|        cpp_dec_float("1.033975765691284593589260865087453566957265138626098632812500000000000000000000000000000000000000000e-25"),
 | |
|        cpp_dec_float("2.067951531382569187178521730174907133914530277252197265625000000000000000000000000000000000000000000e-25"),
 | |
|        cpp_dec_float("4.135903062765138374357043460349814267829060554504394531250000000000000000000000000000000000000000000e-25"),
 | |
|        cpp_dec_float("8.271806125530276748714086920699628535658121109008789062500000000000000000000000000000000000000000000e-25"),
 | |
|        cpp_dec_float("1.654361225106055349742817384139925707131624221801757812500000000000000000000000000000000000000000000e-24"),
 | |
|        cpp_dec_float("3.308722450212110699485634768279851414263248443603515625000000000000000000000000000000000000000000000e-24"),
 | |
|        cpp_dec_float("6.617444900424221398971269536559702828526496887207031250000000000000000000000000000000000000000000000e-24"),
 | |
|        cpp_dec_float("1.323488980084844279794253907311940565705299377441406250000000000000000000000000000000000000000000000e-23"),
 | |
|        cpp_dec_float("2.646977960169688559588507814623881131410598754882812500000000000000000000000000000000000000000000000e-23"),
 | |
|        cpp_dec_float("5.293955920339377119177015629247762262821197509765625000000000000000000000000000000000000000000000000e-23"),
 | |
|        cpp_dec_float("1.058791184067875423835403125849552452564239501953125000000000000000000000000000000000000000000000000e-22"),
 | |
|        cpp_dec_float("2.117582368135750847670806251699104905128479003906250000000000000000000000000000000000000000000000000e-22"),
 | |
|        cpp_dec_float("4.235164736271501695341612503398209810256958007812500000000000000000000000000000000000000000000000000e-22"),
 | |
|        cpp_dec_float("8.470329472543003390683225006796419620513916015625000000000000000000000000000000000000000000000000000e-22"),
 | |
|        cpp_dec_float("1.694065894508600678136645001359283924102783203125000000000000000000000000000000000000000000000000000e-21"),
 | |
|        cpp_dec_float("3.388131789017201356273290002718567848205566406250000000000000000000000000000000000000000000000000000e-21"),
 | |
|        cpp_dec_float("6.776263578034402712546580005437135696411132812500000000000000000000000000000000000000000000000000000e-21"),
 | |
|        cpp_dec_float("1.355252715606880542509316001087427139282226562500000000000000000000000000000000000000000000000000000e-20"),
 | |
|        cpp_dec_float("2.710505431213761085018632002174854278564453125000000000000000000000000000000000000000000000000000000e-20"),
 | |
|        cpp_dec_float("5.421010862427522170037264004349708557128906250000000000000000000000000000000000000000000000000000000e-20"),
 | |
|        cpp_dec_float("1.084202172485504434007452800869941711425781250000000000000000000000000000000000000000000000000000000e-19"),
 | |
|        cpp_dec_float("2.168404344971008868014905601739883422851562500000000000000000000000000000000000000000000000000000000e-19"),
 | |
|        cpp_dec_float("4.336808689942017736029811203479766845703125000000000000000000000000000000000000000000000000000000000e-19"),
 | |
|        cpp_dec_float("8.673617379884035472059622406959533691406250000000000000000000000000000000000000000000000000000000000e-19"),
 | |
|        cpp_dec_float("1.734723475976807094411924481391906738281250000000000000000000000000000000000000000000000000000000000e-18"),
 | |
|        cpp_dec_float("3.469446951953614188823848962783813476562500000000000000000000000000000000000000000000000000000000000e-18"),
 | |
|        cpp_dec_float("6.938893903907228377647697925567626953125000000000000000000000000000000000000000000000000000000000000e-18"),
 | |
|        cpp_dec_float("1.387778780781445675529539585113525390625000000000000000000000000000000000000000000000000000000000000e-17"),
 | |
|        cpp_dec_float("2.775557561562891351059079170227050781250000000000000000000000000000000000000000000000000000000000000e-17"),
 | |
|        cpp_dec_float("5.551115123125782702118158340454101562500000000000000000000000000000000000000000000000000000000000000e-17"),
 | |
|        cpp_dec_float("1.110223024625156540423631668090820312500000000000000000000000000000000000000000000000000000000000000e-16"),
 | |
|        cpp_dec_float("2.220446049250313080847263336181640625000000000000000000000000000000000000000000000000000000000000000e-16"),
 | |
|        cpp_dec_float("4.440892098500626161694526672363281250000000000000000000000000000000000000000000000000000000000000000e-16"),
 | |
|        cpp_dec_float("8.881784197001252323389053344726562500000000000000000000000000000000000000000000000000000000000000000e-16"),
 | |
|        cpp_dec_float("1.776356839400250464677810668945312500000000000000000000000000000000000000000000000000000000000000000e-15"),
 | |
|        cpp_dec_float("3.552713678800500929355621337890625000000000000000000000000000000000000000000000000000000000000000000e-15"),
 | |
|        cpp_dec_float("7.105427357601001858711242675781250000000000000000000000000000000000000000000000000000000000000000000e-15"),
 | |
|        cpp_dec_float("1.421085471520200371742248535156250000000000000000000000000000000000000000000000000000000000000000000e-14"),
 | |
|        cpp_dec_float("2.842170943040400743484497070312500000000000000000000000000000000000000000000000000000000000000000000e-14"),
 | |
|        cpp_dec_float("5.684341886080801486968994140625000000000000000000000000000000000000000000000000000000000000000000000e-14"),
 | |
|        cpp_dec_float("1.136868377216160297393798828125000000000000000000000000000000000000000000000000000000000000000000000e-13"),
 | |
|        cpp_dec_float("2.273736754432320594787597656250000000000000000000000000000000000000000000000000000000000000000000000e-13"),
 | |
|        cpp_dec_float("4.547473508864641189575195312500000000000000000000000000000000000000000000000000000000000000000000000e-13"),
 | |
|        cpp_dec_float("9.094947017729282379150390625000000000000000000000000000000000000000000000000000000000000000000000000e-13"),
 | |
|        cpp_dec_float("1.818989403545856475830078125000000000000000000000000000000000000000000000000000000000000000000000000e-12"),
 | |
|        cpp_dec_float("3.637978807091712951660156250000000000000000000000000000000000000000000000000000000000000000000000000e-12"),
 | |
|        cpp_dec_float("7.275957614183425903320312500000000000000000000000000000000000000000000000000000000000000000000000000e-12"),
 | |
|        cpp_dec_float("1.455191522836685180664062500000000000000000000000000000000000000000000000000000000000000000000000000e-11"),
 | |
|        cpp_dec_float("2.910383045673370361328125000000000000000000000000000000000000000000000000000000000000000000000000000e-11"),
 | |
|        cpp_dec_float("5.820766091346740722656250000000000000000000000000000000000000000000000000000000000000000000000000000e-11"),
 | |
|        cpp_dec_float("1.164153218269348144531250000000000000000000000000000000000000000000000000000000000000000000000000000e-10"),
 | |
|        cpp_dec_float("2.328306436538696289062500000000000000000000000000000000000000000000000000000000000000000000000000000e-10"),
 | |
|        cpp_dec_float("4.656612873077392578125000000000000000000000000000000000000000000000000000000000000000000000000000000e-10"),
 | |
|        cpp_dec_float("9.313225746154785156250000000000000000000000000000000000000000000000000000000000000000000000000000000e-10"),
 | |
|        cpp_dec_float("1.862645149230957031250000000000000000000000000000000000000000000000000000000000000000000000000000000e-9"),
 | |
|        cpp_dec_float("3.725290298461914062500000000000000000000000000000000000000000000000000000000000000000000000000000000e-9"),
 | |
|        cpp_dec_float("7.450580596923828125000000000000000000000000000000000000000000000000000000000000000000000000000000000e-9"),
 | |
|        cpp_dec_float("1.490116119384765625000000000000000000000000000000000000000000000000000000000000000000000000000000000e-8"),
 | |
|        cpp_dec_float("2.980232238769531250000000000000000000000000000000000000000000000000000000000000000000000000000000000e-8"),
 | |
|        cpp_dec_float("5.960464477539062500000000000000000000000000000000000000000000000000000000000000000000000000000000000e-8"),
 | |
|        cpp_dec_float("1.192092895507812500000000000000000000000000000000000000000000000000000000000000000000000000000000000e-7"),
 | |
|        cpp_dec_float("2.384185791015625000000000000000000000000000000000000000000000000000000000000000000000000000000000000e-7"),
 | |
|        cpp_dec_float("4.768371582031250000000000000000000000000000000000000000000000000000000000000000000000000000000000000e-7"),
 | |
|        cpp_dec_float("9.536743164062500000000000000000000000000000000000000000000000000000000000000000000000000000000000000e-7"),
 | |
|        cpp_dec_float("1.907348632812500000000000000000000000000000000000000000000000000000000000000000000000000000000000000e-6"),
 | |
|        cpp_dec_float("3.814697265625000000000000000000000000000000000000000000000000000000000000000000000000000000000000000e-6"),
 | |
|        cpp_dec_float("7.629394531250000000000000000000000000000000000000000000000000000000000000000000000000000000000000000e-6"),
 | |
|        cpp_dec_float("0.000015258789062500000000000000000000000000000000000000000000000000000000000000000000000000000000000000"),
 | |
|        cpp_dec_float("0.000030517578125000000000000000000000000000000000000000000000000000000000000000000000000000000000000000"),
 | |
|        cpp_dec_float("0.000061035156250000000000000000000000000000000000000000000000000000000000000000000000000000000000000000"),
 | |
|        cpp_dec_float("0.000122070312500000000000000000000000000000000000000000000000000000000000000000000000000000000000000000"),
 | |
|        cpp_dec_float("0.000244140625000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000"),
 | |
|        cpp_dec_float("0.000488281250000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000"),
 | |
|        cpp_dec_float("0.000976562500000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000"),
 | |
|        cpp_dec_float("0.001953125000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000"),
 | |
|        cpp_dec_float("0.003906250000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000"),
 | |
|        cpp_dec_float("0.007812500000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000"),
 | |
|        cpp_dec_float("0.01562500000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000"),
 | |
|        cpp_dec_float("0.03125000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000"),
 | |
|        cpp_dec_float("0.06250000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000"),
 | |
|        cpp_dec_float("0.125"),
 | |
|        cpp_dec_float("0.25"),
 | |
|        cpp_dec_float("0.5"),
 | |
|        one(),
 | |
|        two(),
 | |
|        cpp_dec_float(static_cast<boost::ulong_long_type>(4)),
 | |
|        cpp_dec_float(static_cast<boost::ulong_long_type>(8)),
 | |
|        cpp_dec_float(static_cast<boost::ulong_long_type>(16)),
 | |
|        cpp_dec_float(static_cast<boost::ulong_long_type>(32)),
 | |
|        cpp_dec_float(static_cast<boost::ulong_long_type>(64)),
 | |
|        cpp_dec_float(static_cast<boost::ulong_long_type>(128)),
 | |
|        cpp_dec_float(static_cast<boost::ulong_long_type>(256)),
 | |
|        cpp_dec_float(static_cast<boost::ulong_long_type>(512)),
 | |
|        cpp_dec_float(static_cast<boost::ulong_long_type>(1024)),
 | |
|        cpp_dec_float(static_cast<boost::ulong_long_type>(2048)),
 | |
|        cpp_dec_float(static_cast<boost::ulong_long_type>(4096)),
 | |
|        cpp_dec_float(static_cast<boost::ulong_long_type>(8192)),
 | |
|        cpp_dec_float(static_cast<boost::ulong_long_type>(16384)),
 | |
|        cpp_dec_float(static_cast<boost::ulong_long_type>(32768)),
 | |
|        cpp_dec_float(static_cast<boost::ulong_long_type>(65536)),
 | |
|        cpp_dec_float(static_cast<boost::ulong_long_type>(131072)),
 | |
|        cpp_dec_float(static_cast<boost::ulong_long_type>(262144)),
 | |
|        cpp_dec_float(static_cast<boost::ulong_long_type>(524288)),
 | |
|        cpp_dec_float(static_cast<boost::uint64_t>(1uL << 20u)),
 | |
|        cpp_dec_float(static_cast<boost::uint64_t>(1uL << 21u)),
 | |
|        cpp_dec_float(static_cast<boost::uint64_t>(1uL << 22u)),
 | |
|        cpp_dec_float(static_cast<boost::uint64_t>(1uL << 23u)),
 | |
|        cpp_dec_float(static_cast<boost::uint64_t>(1uL << 24u)),
 | |
|        cpp_dec_float(static_cast<boost::uint64_t>(1uL << 25u)),
 | |
|        cpp_dec_float(static_cast<boost::uint64_t>(1uL << 26u)),
 | |
|        cpp_dec_float(static_cast<boost::uint64_t>(1uL << 27u)),
 | |
|        cpp_dec_float(static_cast<boost::uint64_t>(1uL << 28u)),
 | |
|        cpp_dec_float(static_cast<boost::uint64_t>(1uL << 29u)),
 | |
|        cpp_dec_float(static_cast<boost::uint64_t>(1uL << 30u)),
 | |
|        cpp_dec_float(static_cast<boost::uint64_t>(1uL << 31u)),
 | |
|        cpp_dec_float(static_cast<boost::uint64_t>(1uLL << 32u)),
 | |
|        cpp_dec_float(static_cast<boost::uint64_t>(1uLL << 33u)),
 | |
|        cpp_dec_float(static_cast<boost::uint64_t>(1uLL << 34u)),
 | |
|        cpp_dec_float(static_cast<boost::uint64_t>(1uLL << 35u)),
 | |
|        cpp_dec_float(static_cast<boost::uint64_t>(1uLL << 36u)),
 | |
|        cpp_dec_float(static_cast<boost::uint64_t>(1uLL << 37u)),
 | |
|        cpp_dec_float(static_cast<boost::uint64_t>(1uLL << 38u)),
 | |
|        cpp_dec_float(static_cast<boost::uint64_t>(1uLL << 39u)),
 | |
|        cpp_dec_float(static_cast<boost::uint64_t>(1uLL << 40u)),
 | |
|        cpp_dec_float(static_cast<boost::uint64_t>(1uLL << 41u)),
 | |
|        cpp_dec_float(static_cast<boost::uint64_t>(1uLL << 42u)),
 | |
|        cpp_dec_float(static_cast<boost::uint64_t>(1uLL << 43u)),
 | |
|        cpp_dec_float(static_cast<boost::uint64_t>(1uLL << 44u)),
 | |
|        cpp_dec_float(static_cast<boost::uint64_t>(1uLL << 45u)),
 | |
|        cpp_dec_float(static_cast<boost::uint64_t>(1uLL << 46u)),
 | |
|        cpp_dec_float(static_cast<boost::uint64_t>(1uLL << 47u)),
 | |
|        cpp_dec_float(static_cast<boost::uint64_t>(1uLL << 48u)),
 | |
|        cpp_dec_float(static_cast<boost::uint64_t>(1uLL << 49u)),
 | |
|        cpp_dec_float(static_cast<boost::uint64_t>(1uLL << 50u)),
 | |
|        cpp_dec_float(static_cast<boost::uint64_t>(1uLL << 51u)),
 | |
|        cpp_dec_float(static_cast<boost::uint64_t>(1uLL << 52u)),
 | |
|        cpp_dec_float(static_cast<boost::uint64_t>(1uLL << 53u)),
 | |
|        cpp_dec_float(static_cast<boost::uint64_t>(1uLL << 54u)),
 | |
|        cpp_dec_float(static_cast<boost::uint64_t>(1uLL << 55u)),
 | |
|        cpp_dec_float(static_cast<boost::uint64_t>(1uLL << 56u)),
 | |
|        cpp_dec_float(static_cast<boost::uint64_t>(1uLL << 57u)),
 | |
|        cpp_dec_float(static_cast<boost::uint64_t>(1uLL << 58u)),
 | |
|        cpp_dec_float(static_cast<boost::uint64_t>(1uLL << 59u)),
 | |
|        cpp_dec_float(static_cast<boost::uint64_t>(1uLL << 60u)),
 | |
|        cpp_dec_float(static_cast<boost::uint64_t>(1uLL << 61u)),
 | |
|        cpp_dec_float(static_cast<boost::uint64_t>(1uLL << 62u)),
 | |
|        cpp_dec_float(static_cast<boost::uint64_t>(1uLL << 63u)),
 | |
|        cpp_dec_float("1.844674407370955161600000000000000000000000000000000000000000000000000000000000000000000000000000000e19"),
 | |
|        cpp_dec_float("3.689348814741910323200000000000000000000000000000000000000000000000000000000000000000000000000000000e19"),
 | |
|        cpp_dec_float("7.378697629483820646400000000000000000000000000000000000000000000000000000000000000000000000000000000e19"),
 | |
|        cpp_dec_float("1.475739525896764129280000000000000000000000000000000000000000000000000000000000000000000000000000000e20"),
 | |
|        cpp_dec_float("2.951479051793528258560000000000000000000000000000000000000000000000000000000000000000000000000000000e20"),
 | |
|        cpp_dec_float("5.902958103587056517120000000000000000000000000000000000000000000000000000000000000000000000000000000e20"),
 | |
|        cpp_dec_float("1.180591620717411303424000000000000000000000000000000000000000000000000000000000000000000000000000000e21"),
 | |
|        cpp_dec_float("2.361183241434822606848000000000000000000000000000000000000000000000000000000000000000000000000000000e21"),
 | |
|        cpp_dec_float("4.722366482869645213696000000000000000000000000000000000000000000000000000000000000000000000000000000e21"),
 | |
|        cpp_dec_float("9.444732965739290427392000000000000000000000000000000000000000000000000000000000000000000000000000000e21"),
 | |
|        cpp_dec_float("1.888946593147858085478400000000000000000000000000000000000000000000000000000000000000000000000000000e22"),
 | |
|        cpp_dec_float("3.777893186295716170956800000000000000000000000000000000000000000000000000000000000000000000000000000e22"),
 | |
|        cpp_dec_float("7.555786372591432341913600000000000000000000000000000000000000000000000000000000000000000000000000000e22"),
 | |
|        cpp_dec_float("1.511157274518286468382720000000000000000000000000000000000000000000000000000000000000000000000000000e23"),
 | |
|        cpp_dec_float("3.022314549036572936765440000000000000000000000000000000000000000000000000000000000000000000000000000e23"),
 | |
|        cpp_dec_float("6.044629098073145873530880000000000000000000000000000000000000000000000000000000000000000000000000000e23"),
 | |
|        cpp_dec_float("1.208925819614629174706176000000000000000000000000000000000000000000000000000000000000000000000000000e24"),
 | |
|        cpp_dec_float("2.417851639229258349412352000000000000000000000000000000000000000000000000000000000000000000000000000e24"),
 | |
|        cpp_dec_float("4.835703278458516698824704000000000000000000000000000000000000000000000000000000000000000000000000000e24"),
 | |
|        cpp_dec_float("9.671406556917033397649408000000000000000000000000000000000000000000000000000000000000000000000000000e24"),
 | |
|        cpp_dec_float("1.934281311383406679529881600000000000000000000000000000000000000000000000000000000000000000000000000e25"),
 | |
|        cpp_dec_float("3.868562622766813359059763200000000000000000000000000000000000000000000000000000000000000000000000000e25"),
 | |
|        cpp_dec_float("7.737125245533626718119526400000000000000000000000000000000000000000000000000000000000000000000000000e25"),
 | |
|        cpp_dec_float("1.547425049106725343623905280000000000000000000000000000000000000000000000000000000000000000000000000e26"),
 | |
|        cpp_dec_float("3.094850098213450687247810560000000000000000000000000000000000000000000000000000000000000000000000000e26"),
 | |
|        cpp_dec_float("6.189700196426901374495621120000000000000000000000000000000000000000000000000000000000000000000000000e26"),
 | |
|        cpp_dec_float("1.237940039285380274899124224000000000000000000000000000000000000000000000000000000000000000000000000e27"),
 | |
|        cpp_dec_float("2.475880078570760549798248448000000000000000000000000000000000000000000000000000000000000000000000000e27"),
 | |
|        cpp_dec_float("4.951760157141521099596496896000000000000000000000000000000000000000000000000000000000000000000000000e27"),
 | |
|        cpp_dec_float("9.903520314283042199192993792000000000000000000000000000000000000000000000000000000000000000000000000e27"),
 | |
|        cpp_dec_float("1.980704062856608439838598758400000000000000000000000000000000000000000000000000000000000000000000000e28"),
 | |
|        cpp_dec_float("3.961408125713216879677197516800000000000000000000000000000000000000000000000000000000000000000000000e28"),
 | |
|        cpp_dec_float("7.922816251426433759354395033600000000000000000000000000000000000000000000000000000000000000000000000e28"),
 | |
|        cpp_dec_float("1.584563250285286751870879006720000000000000000000000000000000000000000000000000000000000000000000000e29"),
 | |
|        cpp_dec_float("3.169126500570573503741758013440000000000000000000000000000000000000000000000000000000000000000000000e29"),
 | |
|        cpp_dec_float("6.338253001141147007483516026880000000000000000000000000000000000000000000000000000000000000000000000e29"),
 | |
|        cpp_dec_float("1.267650600228229401496703205376000000000000000000000000000000000000000000000000000000000000000000000e30"),
 | |
|        cpp_dec_float("2.535301200456458802993406410752000000000000000000000000000000000000000000000000000000000000000000000e30"),
 | |
|        cpp_dec_float("5.070602400912917605986812821504000000000000000000000000000000000000000000000000000000000000000000000e30"),
 | |
|        cpp_dec_float("1.014120480182583521197362564300800000000000000000000000000000000000000000000000000000000000000000000e31"),
 | |
|        cpp_dec_float("2.028240960365167042394725128601600000000000000000000000000000000000000000000000000000000000000000000e31"),
 | |
|        cpp_dec_float("4.056481920730334084789450257203200000000000000000000000000000000000000000000000000000000000000000000e31"),
 | |
|        cpp_dec_float("8.112963841460668169578900514406400000000000000000000000000000000000000000000000000000000000000000000e31"),
 | |
|        cpp_dec_float("1.622592768292133633915780102881280000000000000000000000000000000000000000000000000000000000000000000e32"),
 | |
|        cpp_dec_float("3.245185536584267267831560205762560000000000000000000000000000000000000000000000000000000000000000000e32"),
 | |
|        cpp_dec_float("6.490371073168534535663120411525120000000000000000000000000000000000000000000000000000000000000000000e32"),
 | |
|        cpp_dec_float("1.298074214633706907132624082305024000000000000000000000000000000000000000000000000000000000000000000e33"),
 | |
|        cpp_dec_float("2.596148429267413814265248164610048000000000000000000000000000000000000000000000000000000000000000000e33"),
 | |
|        cpp_dec_float("5.192296858534827628530496329220096000000000000000000000000000000000000000000000000000000000000000000e33"),
 | |
|        cpp_dec_float("1.038459371706965525706099265844019200000000000000000000000000000000000000000000000000000000000000000e34"),
 | |
|        cpp_dec_float("2.076918743413931051412198531688038400000000000000000000000000000000000000000000000000000000000000000e34"),
 | |
|        cpp_dec_float("4.153837486827862102824397063376076800000000000000000000000000000000000000000000000000000000000000000e34"),
 | |
|        cpp_dec_float("8.307674973655724205648794126752153600000000000000000000000000000000000000000000000000000000000000000e34"),
 | |
|        cpp_dec_float("1.661534994731144841129758825350430720000000000000000000000000000000000000000000000000000000000000000e35"),
 | |
|        cpp_dec_float("3.323069989462289682259517650700861440000000000000000000000000000000000000000000000000000000000000000e35"),
 | |
|        cpp_dec_float("6.646139978924579364519035301401722880000000000000000000000000000000000000000000000000000000000000000e35"),
 | |
|        cpp_dec_float("1.329227995784915872903807060280344576000000000000000000000000000000000000000000000000000000000000000e36"),
 | |
|        cpp_dec_float("2.658455991569831745807614120560689152000000000000000000000000000000000000000000000000000000000000000e36"),
 | |
|        cpp_dec_float("5.316911983139663491615228241121378304000000000000000000000000000000000000000000000000000000000000000e36"),
 | |
|        cpp_dec_float("1.063382396627932698323045648224275660800000000000000000000000000000000000000000000000000000000000000e37"),
 | |
|        cpp_dec_float("2.126764793255865396646091296448551321600000000000000000000000000000000000000000000000000000000000000e37"),
 | |
|        cpp_dec_float("4.253529586511730793292182592897102643200000000000000000000000000000000000000000000000000000000000000e37"),
 | |
|        cpp_dec_float("8.507059173023461586584365185794205286400000000000000000000000000000000000000000000000000000000000000e37"),
 | |
|        cpp_dec_float("1.701411834604692317316873037158841057280000000000000000000000000000000000000000000000000000000000000e38")
 | |
|    }};
 | |
| 
 | |
|    if((p > static_cast<boost::long_long_type>(-128)) && (p < static_cast<boost::long_long_type>(+128)))
 | |
|    {
 | |
|       return p2_data[static_cast<std::size_t>(p + ((p2_data.size() - 1u) / 2u))];
 | |
|    }
 | |
|    else
 | |
|    {
 | |
|       // Compute and return 2^p.
 | |
|       if(p < static_cast<boost::long_long_type>(0))
 | |
|       {
 | |
|          return pow2(static_cast<boost::long_long_type>(-p)).calculate_inv();
 | |
|       }
 | |
|       else
 | |
|       {
 | |
|          cpp_dec_float<Digits10, ExponentType, Allocator> t;
 | |
|          default_ops::detail::pow_imp(t, two(), p, mpl::true_());
 | |
|          return t;
 | |
|       }
 | |
|    }
 | |
| }
 | |
| 
 | |
| 
 | |
| template <unsigned Digits10, class ExponentType, class Allocator>
 | |
| inline void eval_add(cpp_dec_float<Digits10, ExponentType, Allocator>& result, const cpp_dec_float<Digits10, ExponentType, Allocator>& o)
 | |
| {
 | |
|    result += o;
 | |
| }
 | |
| template <unsigned Digits10, class ExponentType, class Allocator>
 | |
| inline void eval_subtract(cpp_dec_float<Digits10, ExponentType, Allocator>& result, const cpp_dec_float<Digits10, ExponentType, Allocator>& o)
 | |
| {
 | |
|    result -= o;
 | |
| }
 | |
| template <unsigned Digits10, class ExponentType, class Allocator>
 | |
| inline void eval_multiply(cpp_dec_float<Digits10, ExponentType, Allocator>& result, const cpp_dec_float<Digits10, ExponentType, Allocator>& o)
 | |
| {
 | |
|    result *= o;
 | |
| }
 | |
| template <unsigned Digits10, class ExponentType, class Allocator>
 | |
| inline void eval_divide(cpp_dec_float<Digits10, ExponentType, Allocator>& result, const cpp_dec_float<Digits10, ExponentType, Allocator>& o)
 | |
| {
 | |
|    result /= o;
 | |
| }
 | |
| 
 | |
| template <unsigned Digits10, class ExponentType, class Allocator>
 | |
| inline void eval_add(cpp_dec_float<Digits10, ExponentType, Allocator>& result, const boost::ulong_long_type& o)
 | |
| {
 | |
|    result.add_unsigned_long_long(o);
 | |
| }
 | |
| template <unsigned Digits10, class ExponentType, class Allocator>
 | |
| inline void eval_subtract(cpp_dec_float<Digits10, ExponentType, Allocator>& result, const boost::ulong_long_type& o)
 | |
| {
 | |
|    result.sub_unsigned_long_long(o);
 | |
| }
 | |
| template <unsigned Digits10, class ExponentType, class Allocator>
 | |
| inline void eval_multiply(cpp_dec_float<Digits10, ExponentType, Allocator>& result, const boost::ulong_long_type& o)
 | |
| {
 | |
|    result.mul_unsigned_long_long(o);
 | |
| }
 | |
| template <unsigned Digits10, class ExponentType, class Allocator>
 | |
| inline void eval_divide(cpp_dec_float<Digits10, ExponentType, Allocator>& result, const boost::ulong_long_type& o)
 | |
| {
 | |
|    result.div_unsigned_long_long(o);
 | |
| }
 | |
| 
 | |
| template <unsigned Digits10, class ExponentType, class Allocator>
 | |
| inline void eval_add(cpp_dec_float<Digits10, ExponentType, Allocator>& result, boost::long_long_type o)
 | |
| {
 | |
|    if(o < 0)
 | |
|       result.sub_unsigned_long_long(boost::multiprecision::detail::unsigned_abs(o));
 | |
|    else
 | |
|       result.add_unsigned_long_long(o);
 | |
| }
 | |
| template <unsigned Digits10, class ExponentType, class Allocator>
 | |
| inline void eval_subtract(cpp_dec_float<Digits10, ExponentType, Allocator>& result, boost::long_long_type o)
 | |
| {
 | |
|    if(o < 0)
 | |
|       result.add_unsigned_long_long(boost::multiprecision::detail::unsigned_abs(o));
 | |
|    else
 | |
|       result.sub_unsigned_long_long(o);
 | |
| }
 | |
| template <unsigned Digits10, class ExponentType, class Allocator>
 | |
| inline void eval_multiply(cpp_dec_float<Digits10, ExponentType, Allocator>& result, boost::long_long_type o)
 | |
| {
 | |
|    if(o < 0)
 | |
|    {
 | |
|       result.mul_unsigned_long_long(boost::multiprecision::detail::unsigned_abs(o));
 | |
|       result.negate();
 | |
|    }
 | |
|    else
 | |
|       result.mul_unsigned_long_long(o);
 | |
| }
 | |
| template <unsigned Digits10, class ExponentType, class Allocator>
 | |
| inline void eval_divide(cpp_dec_float<Digits10, ExponentType, Allocator>& result, boost::long_long_type o)
 | |
| {
 | |
|    if(o < 0)
 | |
|    {
 | |
|       result.div_unsigned_long_long(boost::multiprecision::detail::unsigned_abs(o));
 | |
|       result.negate();
 | |
|    }
 | |
|    else
 | |
|       result.div_unsigned_long_long(o);
 | |
| }
 | |
| 
 | |
| template <unsigned Digits10, class ExponentType, class Allocator>
 | |
| inline void eval_convert_to(boost::ulong_long_type* result, const cpp_dec_float<Digits10, ExponentType, Allocator>& val)
 | |
| {
 | |
|    *result = val.extract_unsigned_long_long();
 | |
| }
 | |
| template <unsigned Digits10, class ExponentType, class Allocator>
 | |
| inline void eval_convert_to(boost::long_long_type* result, const cpp_dec_float<Digits10, ExponentType, Allocator>& val)
 | |
| {
 | |
|    *result = val.extract_signed_long_long();
 | |
| }
 | |
| template <unsigned Digits10, class ExponentType, class Allocator>
 | |
| inline void eval_convert_to(long double* result, cpp_dec_float<Digits10, ExponentType, Allocator>& val)
 | |
| {
 | |
|    *result = val.extract_long_double();
 | |
| }
 | |
| 
 | |
| //
 | |
| // Non member function support:
 | |
| //
 | |
| template <unsigned Digits10, class ExponentType, class Allocator>
 | |
| inline int eval_fpclassify(const cpp_dec_float<Digits10, ExponentType, Allocator>& x)
 | |
| {
 | |
|    if((x.isinf)())
 | |
|       return FP_INFINITE;
 | |
|    if((x.isnan)())
 | |
|       return FP_NAN;
 | |
|    if(x.iszero())
 | |
|       return FP_ZERO;
 | |
|    return FP_NORMAL;
 | |
| }
 | |
| 
 | |
| template <unsigned Digits10, class ExponentType, class Allocator>
 | |
| inline void eval_abs(cpp_dec_float<Digits10, ExponentType, Allocator>& result, const cpp_dec_float<Digits10, ExponentType, Allocator>& x)
 | |
| {
 | |
|    result = x;
 | |
|    if(x.isneg())
 | |
|       result.negate();
 | |
| }
 | |
| 
 | |
| template <unsigned Digits10, class ExponentType, class Allocator>
 | |
| inline void eval_fabs(cpp_dec_float<Digits10, ExponentType, Allocator>& result, const cpp_dec_float<Digits10, ExponentType, Allocator>& x)
 | |
| {
 | |
|    result = x;
 | |
|    if(x.isneg())
 | |
|       result.negate();
 | |
| }
 | |
| 
 | |
| template <unsigned Digits10, class ExponentType, class Allocator>
 | |
| inline void eval_sqrt(cpp_dec_float<Digits10, ExponentType, Allocator>& result, const cpp_dec_float<Digits10, ExponentType, Allocator>& x)
 | |
| {
 | |
|    result = x;
 | |
|    result.calculate_sqrt();
 | |
| }
 | |
| 
 | |
| template <unsigned Digits10, class ExponentType, class Allocator>
 | |
| inline void eval_floor(cpp_dec_float<Digits10, ExponentType, Allocator>& result, const cpp_dec_float<Digits10, ExponentType, Allocator>& x)
 | |
| {
 | |
|    result = x;
 | |
|    if(!(x.isfinite)() || x.isint())
 | |
|    {
 | |
|       return;
 | |
|    }
 | |
| 
 | |
|    if(x.isneg())
 | |
|       result -= cpp_dec_float<Digits10, ExponentType, Allocator>::one();
 | |
|    result = result.extract_integer_part();
 | |
| }
 | |
| 
 | |
| template <unsigned Digits10, class ExponentType, class Allocator>
 | |
| inline void eval_ceil(cpp_dec_float<Digits10, ExponentType, Allocator>& result, const cpp_dec_float<Digits10, ExponentType, Allocator>& x)
 | |
| {
 | |
|    result = x;
 | |
|    if(!(x.isfinite)() || x.isint())
 | |
|    {
 | |
|       return;
 | |
|    }
 | |
| 
 | |
|    if(!x.isneg())
 | |
|       result += cpp_dec_float<Digits10, ExponentType, Allocator>::one();
 | |
|    result = result.extract_integer_part();
 | |
| }
 | |
| 
 | |
| template <unsigned Digits10, class ExponentType, class Allocator>
 | |
| inline void eval_trunc(cpp_dec_float<Digits10, ExponentType, Allocator>& result, const cpp_dec_float<Digits10, ExponentType, Allocator>& x)
 | |
| {
 | |
|    if(!(x.isfinite)())
 | |
|    {
 | |
|       result = boost::math::policies::raise_rounding_error("boost::multiprecision::trunc<%1%>(%1%)", 0, number<cpp_dec_float<Digits10, ExponentType, Allocator> >(x), number<cpp_dec_float<Digits10, ExponentType, Allocator> >(x), boost::math::policies::policy<>()).backend();
 | |
|       return;
 | |
|    }
 | |
|    else if(x.isint())
 | |
|    {
 | |
|       result = x;
 | |
|       return;
 | |
|    }
 | |
|    result = x.extract_integer_part();
 | |
| }
 | |
| 
 | |
| template <unsigned Digits10, class ExponentType, class Allocator>
 | |
| inline ExponentType eval_ilogb(const cpp_dec_float<Digits10, ExponentType, Allocator>& val)
 | |
| {
 | |
|    // Set result, to the exponent of val:
 | |
|    return val.order();
 | |
| }
 | |
| template <unsigned Digits10, class ExponentType, class Allocator, class ArgType>
 | |
| inline void eval_scalbn(cpp_dec_float<Digits10, ExponentType, Allocator>& result, const cpp_dec_float<Digits10, ExponentType, Allocator>& val, ArgType e_)
 | |
| {
 | |
|    using default_ops::eval_multiply;
 | |
|    const ExponentType e = e_;
 | |
|    cpp_dec_float<Digits10, ExponentType, Allocator> t(1.0, e);
 | |
|    eval_multiply(result, val, t);
 | |
| }
 | |
| 
 | |
| template <unsigned Digits10, class ExponentType, class Allocator, class ArgType>
 | |
| inline void eval_ldexp(cpp_dec_float<Digits10, ExponentType, Allocator>& result, const cpp_dec_float<Digits10, ExponentType, Allocator>& x, ArgType e)
 | |
| {
 | |
|    const boost::long_long_type the_exp = static_cast<boost::long_long_type>(e);
 | |
| 
 | |
|    if((the_exp > (std::numeric_limits<ExponentType>::max)()) || (the_exp < (std::numeric_limits<ExponentType>::min)()))
 | |
|       BOOST_THROW_EXCEPTION(std::runtime_error(std::string("Exponent value is out of range.")));
 | |
| 
 | |
|    result = x;
 | |
| 
 | |
|    if ((the_exp > static_cast<boost::long_long_type>(-std::numeric_limits<boost::long_long_type>::digits)) && (the_exp < static_cast<boost::long_long_type>(0)))
 | |
|       result.div_unsigned_long_long(1ULL << static_cast<boost::long_long_type>(-the_exp));
 | |
|    else if((the_exp < static_cast<boost::long_long_type>( std::numeric_limits<boost::long_long_type>::digits)) && (the_exp > static_cast<boost::long_long_type>(0)))
 | |
|       result.mul_unsigned_long_long(1ULL << the_exp);
 | |
|    else if(the_exp != static_cast<boost::long_long_type>(0))
 | |
|       result *= cpp_dec_float<Digits10, ExponentType, Allocator>::pow2(e);
 | |
| }
 | |
| 
 | |
| template <unsigned Digits10, class ExponentType, class Allocator>
 | |
| inline void eval_frexp(cpp_dec_float<Digits10, ExponentType, Allocator>& result, const cpp_dec_float<Digits10, ExponentType, Allocator>& x, ExponentType* e)
 | |
| {
 | |
|    result = x;
 | |
|    if(result.isneg())
 | |
|       result.negate();
 | |
| 
 | |
|    if(result.iszero())
 | |
|    {
 | |
|       *e = 0;
 | |
|       return;
 | |
|    }
 | |
| 
 | |
|    ExponentType t = result.order();
 | |
|    BOOST_MP_USING_ABS
 | |
|    if(abs(t) < ((std::numeric_limits<ExponentType>::max)() / 1000))
 | |
|    {
 | |
|       t *= 1000;
 | |
|       t /= 301;
 | |
|    }
 | |
|    else
 | |
|    {
 | |
|       t /= 301;
 | |
|       t *= 1000;
 | |
|    }
 | |
| 
 | |
|    result *= cpp_dec_float<Digits10, ExponentType, Allocator>::pow2(-t);
 | |
| 
 | |
|    if(result.iszero() || (result.isinf)() || (result.isnan)())
 | |
|    {
 | |
|       // pow2 overflowed, slip the calculation up:
 | |
|       result = x;
 | |
|       if(result.isneg())
 | |
|          result.negate();
 | |
|       t /= 2;
 | |
|       result *= cpp_dec_float<Digits10, ExponentType, Allocator>::pow2(-t);
 | |
|    }
 | |
|    BOOST_MP_USING_ABS
 | |
|    if(abs(result.order()) > 5)
 | |
|    {
 | |
|       // If our first estimate doesn't get close enough then try recursion until we do:
 | |
|       ExponentType e2;
 | |
|       cpp_dec_float<Digits10, ExponentType, Allocator> r2;
 | |
|       eval_frexp(r2, result, &e2);
 | |
|       // overflow protection:
 | |
|       if((t > 0) && (e2 > 0) && (t > (std::numeric_limits<ExponentType>::max)() - e2))
 | |
|          BOOST_THROW_EXCEPTION(std::runtime_error("Exponent is too large to be represented as a power of 2."));
 | |
|       if((t < 0) && (e2 < 0) && (t < (std::numeric_limits<ExponentType>::min)() - e2))
 | |
|          BOOST_THROW_EXCEPTION(std::runtime_error("Exponent is too large to be represented as a power of 2."));
 | |
|       t += e2;
 | |
|       result = r2;
 | |
|    }
 | |
| 
 | |
|    while(result.compare(cpp_dec_float<Digits10, ExponentType, Allocator>::one()) >= 0)
 | |
|    {
 | |
|       result /= cpp_dec_float<Digits10, ExponentType, Allocator>::two();
 | |
|       ++t;
 | |
|    }
 | |
|    while(result.compare(cpp_dec_float<Digits10, ExponentType, Allocator>::half()) < 0)
 | |
|    {
 | |
|       result *= cpp_dec_float<Digits10, ExponentType, Allocator>::two();
 | |
|       --t;
 | |
|    }
 | |
|    *e = t;
 | |
|    if(x.isneg())
 | |
|       result.negate();
 | |
| }
 | |
| 
 | |
| template <unsigned Digits10, class ExponentType, class Allocator>
 | |
| inline typename disable_if<is_same<ExponentType, int> >::type eval_frexp(cpp_dec_float<Digits10, ExponentType, Allocator>& result, const cpp_dec_float<Digits10, ExponentType, Allocator>& x, int* e)
 | |
| {
 | |
|    ExponentType t;
 | |
|    eval_frexp(result, x, &t);
 | |
|    if((t > (std::numeric_limits<int>::max)()) || (t < (std::numeric_limits<int>::min)()))
 | |
|       BOOST_THROW_EXCEPTION(std::runtime_error("Exponent is outside the range of an int"));
 | |
|    *e = static_cast<int>(t);
 | |
| }
 | |
| 
 | |
| template <unsigned Digits10, class ExponentType, class Allocator>
 | |
| inline bool eval_is_zero(const cpp_dec_float<Digits10, ExponentType, Allocator>& val)
 | |
| {
 | |
|    return val.iszero();
 | |
| }
 | |
| template <unsigned Digits10, class ExponentType, class Allocator>
 | |
| inline int eval_get_sign(const cpp_dec_float<Digits10, ExponentType, Allocator>& val)
 | |
| {
 | |
|    return val.iszero() ? 0 : val.isneg() ? -1 : 1;
 | |
| }
 | |
| 
 | |
| template <unsigned Digits10, class ExponentType, class Allocator>
 | |
| inline std::size_t hash_value(const cpp_dec_float<Digits10, ExponentType, Allocator>& val)
 | |
| {
 | |
|    return val.hash();
 | |
| }
 | |
| 
 | |
| } // namespace backends
 | |
| 
 | |
| using boost::multiprecision::backends::cpp_dec_float;
 | |
| 
 | |
| 
 | |
| typedef number<cpp_dec_float<50> > cpp_dec_float_50;
 | |
| typedef number<cpp_dec_float<100> > cpp_dec_float_100;
 | |
| 
 | |
| #ifdef BOOST_NO_SFINAE_EXPR
 | |
| 
 | |
| namespace detail{
 | |
| 
 | |
| template<unsigned D1, class E1, class A1, unsigned D2, class E2, class A2>
 | |
| struct is_explicitly_convertible<cpp_dec_float<D1, E1, A1>, cpp_dec_float<D2, E2, A2> > : public mpl::true_ {};
 | |
| 
 | |
| }
 | |
| 
 | |
| #endif
 | |
| 
 | |
| 
 | |
| }}
 | |
| 
 | |
| namespace std
 | |
| {
 | |
|    template <unsigned Digits10, class ExponentType, class Allocator, boost::multiprecision::expression_template_option ExpressionTemplates>
 | |
|    class numeric_limits<boost::multiprecision::number<boost::multiprecision::cpp_dec_float<Digits10, ExponentType, Allocator>, ExpressionTemplates> >
 | |
|    {
 | |
|    public:
 | |
|       BOOST_STATIC_CONSTEXPR bool is_specialized = true;
 | |
|       BOOST_STATIC_CONSTEXPR bool is_signed = true;
 | |
|       BOOST_STATIC_CONSTEXPR bool is_integer = false;
 | |
|       BOOST_STATIC_CONSTEXPR bool is_exact = false;
 | |
|       BOOST_STATIC_CONSTEXPR bool is_bounded = true;
 | |
|       BOOST_STATIC_CONSTEXPR bool is_modulo = false;
 | |
|       BOOST_STATIC_CONSTEXPR bool is_iec559 = false;
 | |
|       BOOST_STATIC_CONSTEXPR int digits = boost::multiprecision::cpp_dec_float<Digits10, ExponentType, Allocator>::cpp_dec_float_digits10;
 | |
|       BOOST_STATIC_CONSTEXPR int digits10 = boost::multiprecision::cpp_dec_float<Digits10, ExponentType, Allocator>::cpp_dec_float_digits10;
 | |
|       BOOST_STATIC_CONSTEXPR int max_digits10 = boost::multiprecision::cpp_dec_float<Digits10, ExponentType, Allocator>::cpp_dec_float_total_digits10;
 | |
|       BOOST_STATIC_CONSTEXPR ExponentType min_exponent = boost::multiprecision::cpp_dec_float<Digits10, ExponentType, Allocator>::cpp_dec_float_min_exp; // Type differs from int.
 | |
|       BOOST_STATIC_CONSTEXPR ExponentType min_exponent10 = boost::multiprecision::cpp_dec_float<Digits10, ExponentType, Allocator>::cpp_dec_float_min_exp10; // Type differs from int.
 | |
|       BOOST_STATIC_CONSTEXPR ExponentType max_exponent = boost::multiprecision::cpp_dec_float<Digits10, ExponentType, Allocator>::cpp_dec_float_max_exp; // Type differs from int.
 | |
|       BOOST_STATIC_CONSTEXPR ExponentType max_exponent10 = boost::multiprecision::cpp_dec_float<Digits10, ExponentType, Allocator>::cpp_dec_float_max_exp10; // Type differs from int.
 | |
|       BOOST_STATIC_CONSTEXPR int radix = boost::multiprecision::cpp_dec_float<Digits10, ExponentType, Allocator>::cpp_dec_float_radix;
 | |
|       BOOST_STATIC_CONSTEXPR std::float_round_style round_style = std::round_indeterminate;
 | |
|       BOOST_STATIC_CONSTEXPR bool has_infinity = true;
 | |
|       BOOST_STATIC_CONSTEXPR bool has_quiet_NaN = true;
 | |
|       BOOST_STATIC_CONSTEXPR bool has_signaling_NaN = false;
 | |
|       BOOST_STATIC_CONSTEXPR std::float_denorm_style has_denorm = std::denorm_absent;
 | |
|       BOOST_STATIC_CONSTEXPR bool has_denorm_loss = false;
 | |
|       BOOST_STATIC_CONSTEXPR bool traps = false;
 | |
|       BOOST_STATIC_CONSTEXPR bool tinyness_before = false;
 | |
| 
 | |
|       BOOST_STATIC_CONSTEXPR boost::multiprecision::number<boost::multiprecision::cpp_dec_float<Digits10, ExponentType, Allocator>, ExpressionTemplates> (min) () { return (boost::multiprecision::cpp_dec_float<Digits10, ExponentType, Allocator>::min)(); }
 | |
|       BOOST_STATIC_CONSTEXPR boost::multiprecision::number<boost::multiprecision::cpp_dec_float<Digits10, ExponentType, Allocator>, ExpressionTemplates> (max) () { return (boost::multiprecision::cpp_dec_float<Digits10, ExponentType, Allocator>::max)(); }
 | |
|       BOOST_STATIC_CONSTEXPR boost::multiprecision::number<boost::multiprecision::cpp_dec_float<Digits10, ExponentType, Allocator>, ExpressionTemplates> lowest () { return boost::multiprecision::cpp_dec_float<Digits10, ExponentType, Allocator>::zero(); }
 | |
|       BOOST_STATIC_CONSTEXPR boost::multiprecision::number<boost::multiprecision::cpp_dec_float<Digits10, ExponentType, Allocator>, ExpressionTemplates> epsilon () { return boost::multiprecision::cpp_dec_float<Digits10, ExponentType, Allocator>::eps(); }
 | |
|       BOOST_STATIC_CONSTEXPR boost::multiprecision::number<boost::multiprecision::cpp_dec_float<Digits10, ExponentType, Allocator>, ExpressionTemplates> round_error () { return 0.5L; }
 | |
|       BOOST_STATIC_CONSTEXPR boost::multiprecision::number<boost::multiprecision::cpp_dec_float<Digits10, ExponentType, Allocator>, ExpressionTemplates> infinity () { return boost::multiprecision::cpp_dec_float<Digits10, ExponentType, Allocator>::inf(); }
 | |
|       BOOST_STATIC_CONSTEXPR boost::multiprecision::number<boost::multiprecision::cpp_dec_float<Digits10, ExponentType, Allocator>, ExpressionTemplates> quiet_NaN () { return boost::multiprecision::cpp_dec_float<Digits10, ExponentType, Allocator>::nan(); }
 | |
|       BOOST_STATIC_CONSTEXPR boost::multiprecision::number<boost::multiprecision::cpp_dec_float<Digits10, ExponentType, Allocator>, ExpressionTemplates> signaling_NaN() { return boost::multiprecision::cpp_dec_float<Digits10, ExponentType, Allocator>::zero(); }
 | |
|       BOOST_STATIC_CONSTEXPR boost::multiprecision::number<boost::multiprecision::cpp_dec_float<Digits10, ExponentType, Allocator>, ExpressionTemplates> denorm_min () { return boost::multiprecision::cpp_dec_float<Digits10, ExponentType, Allocator>::zero(); }
 | |
|    };
 | |
| 
 | |
| #ifndef BOOST_NO_INCLASS_MEMBER_INITIALIZATION
 | |
| 
 | |
| template <unsigned Digits10, class ExponentType, class Allocator, boost::multiprecision::expression_template_option ExpressionTemplates>
 | |
| BOOST_CONSTEXPR_OR_CONST int numeric_limits<boost::multiprecision::number<boost::multiprecision::cpp_dec_float<Digits10, ExponentType, Allocator>, ExpressionTemplates> >::digits;
 | |
| template <unsigned Digits10, class ExponentType, class Allocator, boost::multiprecision::expression_template_option ExpressionTemplates>
 | |
| BOOST_CONSTEXPR_OR_CONST int numeric_limits<boost::multiprecision::number<boost::multiprecision::cpp_dec_float<Digits10, ExponentType, Allocator>, ExpressionTemplates> >::digits10;
 | |
| template <unsigned Digits10, class ExponentType, class Allocator, boost::multiprecision::expression_template_option ExpressionTemplates>
 | |
| BOOST_CONSTEXPR_OR_CONST int numeric_limits<boost::multiprecision::number<boost::multiprecision::cpp_dec_float<Digits10, ExponentType, Allocator>, ExpressionTemplates> >::max_digits10;
 | |
| template <unsigned Digits10, class ExponentType, class Allocator, boost::multiprecision::expression_template_option ExpressionTemplates>
 | |
| BOOST_CONSTEXPR_OR_CONST bool numeric_limits<boost::multiprecision::number<boost::multiprecision::cpp_dec_float<Digits10, ExponentType, Allocator>, ExpressionTemplates> >::is_signed;
 | |
| template <unsigned Digits10, class ExponentType, class Allocator, boost::multiprecision::expression_template_option ExpressionTemplates>
 | |
| BOOST_CONSTEXPR_OR_CONST bool numeric_limits<boost::multiprecision::number<boost::multiprecision::cpp_dec_float<Digits10, ExponentType, Allocator>, ExpressionTemplates> >::is_integer;
 | |
| template <unsigned Digits10, class ExponentType, class Allocator, boost::multiprecision::expression_template_option ExpressionTemplates>
 | |
| BOOST_CONSTEXPR_OR_CONST bool numeric_limits<boost::multiprecision::number<boost::multiprecision::cpp_dec_float<Digits10, ExponentType, Allocator>, ExpressionTemplates> >::is_exact;
 | |
| template <unsigned Digits10, class ExponentType, class Allocator, boost::multiprecision::expression_template_option ExpressionTemplates>
 | |
| BOOST_CONSTEXPR_OR_CONST int numeric_limits<boost::multiprecision::number<boost::multiprecision::cpp_dec_float<Digits10, ExponentType, Allocator>, ExpressionTemplates> >::radix;
 | |
| template <unsigned Digits10, class ExponentType, class Allocator, boost::multiprecision::expression_template_option ExpressionTemplates>
 | |
| BOOST_CONSTEXPR_OR_CONST ExponentType numeric_limits<boost::multiprecision::number<boost::multiprecision::cpp_dec_float<Digits10, ExponentType, Allocator>, ExpressionTemplates> >::min_exponent;
 | |
| template <unsigned Digits10, class ExponentType, class Allocator, boost::multiprecision::expression_template_option ExpressionTemplates>
 | |
| BOOST_CONSTEXPR_OR_CONST ExponentType numeric_limits<boost::multiprecision::number<boost::multiprecision::cpp_dec_float<Digits10, ExponentType, Allocator>, ExpressionTemplates> >::min_exponent10;
 | |
| template <unsigned Digits10, class ExponentType, class Allocator, boost::multiprecision::expression_template_option ExpressionTemplates>
 | |
| BOOST_CONSTEXPR_OR_CONST ExponentType numeric_limits<boost::multiprecision::number<boost::multiprecision::cpp_dec_float<Digits10, ExponentType, Allocator>, ExpressionTemplates> >::max_exponent;
 | |
| template <unsigned Digits10, class ExponentType, class Allocator, boost::multiprecision::expression_template_option ExpressionTemplates>
 | |
| BOOST_CONSTEXPR_OR_CONST ExponentType numeric_limits<boost::multiprecision::number<boost::multiprecision::cpp_dec_float<Digits10, ExponentType, Allocator>, ExpressionTemplates> >::max_exponent10;
 | |
| template <unsigned Digits10, class ExponentType, class Allocator, boost::multiprecision::expression_template_option ExpressionTemplates>
 | |
| BOOST_CONSTEXPR_OR_CONST bool numeric_limits<boost::multiprecision::number<boost::multiprecision::cpp_dec_float<Digits10, ExponentType, Allocator>, ExpressionTemplates> >::has_infinity;
 | |
| template <unsigned Digits10, class ExponentType, class Allocator, boost::multiprecision::expression_template_option ExpressionTemplates>
 | |
| BOOST_CONSTEXPR_OR_CONST bool numeric_limits<boost::multiprecision::number<boost::multiprecision::cpp_dec_float<Digits10, ExponentType, Allocator>, ExpressionTemplates> >::has_quiet_NaN;
 | |
| template <unsigned Digits10, class ExponentType, class Allocator, boost::multiprecision::expression_template_option ExpressionTemplates>
 | |
| BOOST_CONSTEXPR_OR_CONST bool numeric_limits<boost::multiprecision::number<boost::multiprecision::cpp_dec_float<Digits10, ExponentType, Allocator>, ExpressionTemplates> >::has_signaling_NaN;
 | |
| template <unsigned Digits10, class ExponentType, class Allocator, boost::multiprecision::expression_template_option ExpressionTemplates>
 | |
| BOOST_CONSTEXPR_OR_CONST float_denorm_style numeric_limits<boost::multiprecision::number<boost::multiprecision::cpp_dec_float<Digits10, ExponentType, Allocator>, ExpressionTemplates> >::has_denorm;
 | |
| template <unsigned Digits10, class ExponentType, class Allocator, boost::multiprecision::expression_template_option ExpressionTemplates>
 | |
| BOOST_CONSTEXPR_OR_CONST bool numeric_limits<boost::multiprecision::number<boost::multiprecision::cpp_dec_float<Digits10, ExponentType, Allocator>, ExpressionTemplates> >::has_denorm_loss;
 | |
| template <unsigned Digits10, class ExponentType, class Allocator, boost::multiprecision::expression_template_option ExpressionTemplates>
 | |
| BOOST_CONSTEXPR_OR_CONST bool numeric_limits<boost::multiprecision::number<boost::multiprecision::cpp_dec_float<Digits10, ExponentType, Allocator>, ExpressionTemplates> >::is_iec559;
 | |
| template <unsigned Digits10, class ExponentType, class Allocator, boost::multiprecision::expression_template_option ExpressionTemplates>
 | |
| BOOST_CONSTEXPR_OR_CONST bool numeric_limits<boost::multiprecision::number<boost::multiprecision::cpp_dec_float<Digits10, ExponentType, Allocator>, ExpressionTemplates> >::is_bounded;
 | |
| template <unsigned Digits10, class ExponentType, class Allocator, boost::multiprecision::expression_template_option ExpressionTemplates>
 | |
| BOOST_CONSTEXPR_OR_CONST bool numeric_limits<boost::multiprecision::number<boost::multiprecision::cpp_dec_float<Digits10, ExponentType, Allocator>, ExpressionTemplates> >::is_modulo;
 | |
| template <unsigned Digits10, class ExponentType, class Allocator, boost::multiprecision::expression_template_option ExpressionTemplates>
 | |
| BOOST_CONSTEXPR_OR_CONST bool numeric_limits<boost::multiprecision::number<boost::multiprecision::cpp_dec_float<Digits10, ExponentType, Allocator>, ExpressionTemplates> >::traps;
 | |
| template <unsigned Digits10, class ExponentType, class Allocator, boost::multiprecision::expression_template_option ExpressionTemplates>
 | |
| BOOST_CONSTEXPR_OR_CONST bool numeric_limits<boost::multiprecision::number<boost::multiprecision::cpp_dec_float<Digits10, ExponentType, Allocator>, ExpressionTemplates> >::tinyness_before;
 | |
| template <unsigned Digits10, class ExponentType, class Allocator, boost::multiprecision::expression_template_option ExpressionTemplates>
 | |
| BOOST_CONSTEXPR_OR_CONST float_round_style numeric_limits<boost::multiprecision::number<boost::multiprecision::cpp_dec_float<Digits10, ExponentType, Allocator>, ExpressionTemplates> >::round_style;
 | |
| 
 | |
| #endif
 | |
| }
 | |
| 
 | |
| namespace boost{ namespace math{
 | |
| 
 | |
| namespace policies{
 | |
| 
 | |
| template <unsigned Digits10, class ExponentType, class Allocator, class Policy, boost::multiprecision::expression_template_option ExpressionTemplates>
 | |
| struct precision< boost::multiprecision::number<boost::multiprecision::cpp_dec_float<Digits10, ExponentType, Allocator>, ExpressionTemplates>, Policy>
 | |
| {
 | |
|    // Define a local copy of cpp_dec_float_digits10 because it might differ
 | |
|    // from the template parameter Digits10 for small or large digit counts.
 | |
|    static const boost::int32_t cpp_dec_float_digits10 = boost::multiprecision::cpp_dec_float<Digits10, ExponentType, Allocator>::cpp_dec_float_digits10;
 | |
| 
 | |
|    typedef typename Policy::precision_type precision_type;
 | |
|    typedef digits2<((cpp_dec_float_digits10 + 1LL) * 1000LL) / 301LL> digits_2;
 | |
|    typedef typename mpl::if_c<
 | |
|       ((digits_2::value <= precision_type::value)
 | |
|       || (Policy::precision_type::value <= 0)),
 | |
|       // Default case, full precision for RealType:
 | |
|       digits_2,
 | |
|       // User customized precision:
 | |
|       precision_type
 | |
|    >::type type;
 | |
| };
 | |
| 
 | |
| } // namespace policies
 | |
| 
 | |
| }} // namespaces boost::math
 | |
| 
 | |
| #ifdef BOOST_MSVC
 | |
| #pragma warning(pop)
 | |
| #endif
 | |
| 
 | |
| #endif
 | 
