mirror of
https://github.com/saitohirga/WSJT-X.git
synced 2024-11-19 10:32:02 -05:00
ecaa982b9f
git-svn-id: svn+ssh://svn.code.sf.net/p/wsjt/wsjt/branches/wsjtx@7593 ab8295b8-cf94-4d9e-aec4-7959e3be5d79
2913 lines
107 KiB
C++
2913 lines
107 KiB
C++
///////////////////////////////////////////////////////////////////////////////
|
|
// Copyright 2011 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)
|
|
|
|
#ifndef BOOST_MATH_ER_GMP_BACKEND_HPP
|
|
#define BOOST_MATH_ER_GMP_BACKEND_HPP
|
|
|
|
#include <boost/multiprecision/number.hpp>
|
|
#include <boost/multiprecision/debug_adaptor.hpp>
|
|
#include <boost/multiprecision/detail/integer_ops.hpp>
|
|
#include <boost/multiprecision/detail/big_lanczos.hpp>
|
|
#include <boost/multiprecision/detail/digits.hpp>
|
|
#include <boost/math/special_functions/fpclassify.hpp>
|
|
#include <boost/cstdint.hpp>
|
|
#include <boost/functional/hash_fwd.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:4127)
|
|
#endif
|
|
#include <gmp.h>
|
|
#ifdef BOOST_MSVC
|
|
# pragma warning(pop)
|
|
#endif
|
|
|
|
#if defined(__MPIR_VERSION) && defined(__MPIR_VERSION_MINOR) && defined(__MPIR_VERSION_PATCHLEVEL)
|
|
# define BOOST_MP_MPIR_VERSION (__MPIR_VERSION * 10000 + __MPIR_VERSION_MINOR * 100 + __MPIR_VERSION_PATCHLEVEL)
|
|
#else
|
|
# define BOOST_MP_MPIR_VERSION 0
|
|
#endif
|
|
|
|
#include <cmath>
|
|
#include <limits>
|
|
#include <climits>
|
|
|
|
namespace boost{
|
|
namespace multiprecision{
|
|
namespace backends{
|
|
|
|
#ifdef BOOST_MSVC
|
|
// warning C4127: conditional expression is constant
|
|
#pragma warning(push)
|
|
#pragma warning(disable:4127)
|
|
#endif
|
|
|
|
template <unsigned digits10>
|
|
struct gmp_float;
|
|
struct gmp_int;
|
|
struct gmp_rational;
|
|
|
|
} // namespace backends
|
|
|
|
template<>
|
|
struct number_category<backends::gmp_int> : public mpl::int_<number_kind_integer>{};
|
|
template<>
|
|
struct number_category<backends::gmp_rational> : public mpl::int_<number_kind_rational>{};
|
|
template <unsigned digits10>
|
|
struct number_category<backends::gmp_float<digits10> > : public mpl::int_<number_kind_floating_point>{};
|
|
|
|
namespace backends{
|
|
//
|
|
// Within this file, the only functions we mark as noexcept are those that manipulate
|
|
// (but don't create) an mpf_t. All other types may allocate at pretty much any time
|
|
// via a user-supplied allocator, and therefore throw.
|
|
//
|
|
namespace detail{
|
|
|
|
template <unsigned digits10>
|
|
struct gmp_float_imp
|
|
{
|
|
#ifdef BOOST_HAS_LONG_LONG
|
|
typedef mpl::list<long, boost::long_long_type> signed_types;
|
|
typedef mpl::list<unsigned long, boost::ulong_long_type> unsigned_types;
|
|
#else
|
|
typedef mpl::list<long> signed_types;
|
|
typedef mpl::list<unsigned long> unsigned_types;
|
|
#endif
|
|
typedef mpl::list<double, long double> float_types;
|
|
typedef long exponent_type;
|
|
|
|
gmp_float_imp() BOOST_NOEXCEPT {}
|
|
|
|
gmp_float_imp(const gmp_float_imp& o)
|
|
{
|
|
//
|
|
// We have to do an init followed by a set here, otherwise *this may be at
|
|
// a lower precision than o: seems like mpf_init_set copies just enough bits
|
|
// to get the right value, but if it's then used in further calculations
|
|
// things go badly wrong!!
|
|
//
|
|
mpf_init2(m_data, multiprecision::detail::digits10_2_2(digits10 ? digits10 : get_default_precision()));
|
|
if(o.m_data[0]._mp_d)
|
|
mpf_set(m_data, o.m_data);
|
|
}
|
|
#ifndef BOOST_NO_CXX11_RVALUE_REFERENCES
|
|
gmp_float_imp(gmp_float_imp&& o) BOOST_NOEXCEPT
|
|
{
|
|
m_data[0] = o.m_data[0];
|
|
o.m_data[0]._mp_d = 0;
|
|
}
|
|
#endif
|
|
gmp_float_imp& operator = (const gmp_float_imp& o)
|
|
{
|
|
if(m_data[0]._mp_d == 0)
|
|
mpf_init2(m_data, multiprecision::detail::digits10_2_2(digits10 ? digits10 : get_default_precision()));
|
|
if(o.m_data[0]._mp_d)
|
|
mpf_set(m_data, o.m_data);
|
|
return *this;
|
|
}
|
|
#ifndef BOOST_NO_CXX11_RVALUE_REFERENCES
|
|
gmp_float_imp& operator = (gmp_float_imp&& o) BOOST_NOEXCEPT
|
|
{
|
|
mpf_swap(m_data, o.m_data);
|
|
return *this;
|
|
}
|
|
#endif
|
|
|
|
#ifdef BOOST_HAS_LONG_LONG
|
|
#if defined(ULLONG_MAX) && (ULLONG_MAX == ULONG_MAX)
|
|
gmp_float_imp& operator = (boost::ulong_long_type i)
|
|
{
|
|
*this = static_cast<unsigned long>(i);
|
|
return *this;
|
|
}
|
|
#else
|
|
gmp_float_imp& operator = (boost::ulong_long_type i)
|
|
{
|
|
if(m_data[0]._mp_d == 0)
|
|
mpf_init2(m_data, multiprecision::detail::digits10_2_2(digits10 ? digits10 : get_default_precision()));
|
|
boost::ulong_long_type mask = ((((1uLL << (std::numeric_limits<unsigned long>::digits - 1)) - 1) << 1) | 1uLL);
|
|
unsigned shift = 0;
|
|
mpf_t t;
|
|
mpf_init2(t, multiprecision::detail::digits10_2_2(digits10 ? digits10 : get_default_precision()));
|
|
mpf_set_ui(m_data, 0);
|
|
while(i)
|
|
{
|
|
mpf_set_ui(t, static_cast<unsigned long>(i & mask));
|
|
if(shift)
|
|
mpf_mul_2exp(t, t, shift);
|
|
mpf_add(m_data, m_data, t);
|
|
shift += std::numeric_limits<unsigned long>::digits;
|
|
i >>= std::numeric_limits<unsigned long>::digits;
|
|
}
|
|
mpf_clear(t);
|
|
return *this;
|
|
}
|
|
#endif
|
|
gmp_float_imp& operator = (boost::long_long_type i)
|
|
{
|
|
if(m_data[0]._mp_d == 0)
|
|
mpf_init2(m_data, multiprecision::detail::digits10_2_2(digits10 ? digits10 : get_default_precision()));
|
|
bool neg = i < 0;
|
|
*this = static_cast<boost::ulong_long_type>(boost::multiprecision::detail::unsigned_abs(i));
|
|
if(neg)
|
|
mpf_neg(m_data, m_data);
|
|
return *this;
|
|
}
|
|
#endif
|
|
gmp_float_imp& operator = (unsigned long i)
|
|
{
|
|
if(m_data[0]._mp_d == 0)
|
|
mpf_init2(m_data, multiprecision::detail::digits10_2_2(digits10 ? digits10 : get_default_precision()));
|
|
mpf_set_ui(m_data, i);
|
|
return *this;
|
|
}
|
|
gmp_float_imp& operator = (long i)
|
|
{
|
|
if(m_data[0]._mp_d == 0)
|
|
mpf_init2(m_data, multiprecision::detail::digits10_2_2(digits10 ? digits10 : get_default_precision()));
|
|
mpf_set_si(m_data, i);
|
|
return *this;
|
|
}
|
|
gmp_float_imp& operator = (double d)
|
|
{
|
|
if(m_data[0]._mp_d == 0)
|
|
mpf_init2(m_data, multiprecision::detail::digits10_2_2(digits10 ? digits10 : get_default_precision()));
|
|
mpf_set_d(m_data, d);
|
|
return *this;
|
|
}
|
|
gmp_float_imp& operator = (long double a)
|
|
{
|
|
using std::frexp;
|
|
using std::ldexp;
|
|
using std::floor;
|
|
|
|
if(m_data[0]._mp_d == 0)
|
|
mpf_init2(m_data, multiprecision::detail::digits10_2_2(digits10 ? digits10 : get_default_precision()));
|
|
|
|
if (a == 0) {
|
|
mpf_set_si(m_data, 0);
|
|
return *this;
|
|
}
|
|
|
|
if (a == 1) {
|
|
mpf_set_si(m_data, 1);
|
|
return *this;
|
|
}
|
|
|
|
BOOST_ASSERT(!(boost::math::isinf)(a));
|
|
BOOST_ASSERT(!(boost::math::isnan)(a));
|
|
|
|
int e;
|
|
long double f, term;
|
|
mpf_set_ui(m_data, 0u);
|
|
|
|
f = frexp(a, &e);
|
|
|
|
static const int shift = std::numeric_limits<int>::digits - 1;
|
|
|
|
while(f)
|
|
{
|
|
// extract int sized bits from f:
|
|
f = ldexp(f, shift);
|
|
term = floor(f);
|
|
e -= shift;
|
|
mpf_mul_2exp(m_data, m_data, shift);
|
|
if(term > 0)
|
|
mpf_add_ui(m_data, m_data, static_cast<unsigned>(term));
|
|
else
|
|
mpf_sub_ui(m_data, m_data, static_cast<unsigned>(-term));
|
|
f -= term;
|
|
}
|
|
if(e > 0)
|
|
mpf_mul_2exp(m_data, m_data, e);
|
|
else if(e < 0)
|
|
mpf_div_2exp(m_data, m_data, -e);
|
|
return *this;
|
|
}
|
|
gmp_float_imp& operator = (const char* s)
|
|
{
|
|
if(m_data[0]._mp_d == 0)
|
|
mpf_init2(m_data, multiprecision::detail::digits10_2_2(digits10 ? digits10 : get_default_precision()));
|
|
if(0 != mpf_set_str(m_data, s, 10))
|
|
BOOST_THROW_EXCEPTION(std::runtime_error(std::string("The string \"") + s + std::string("\"could not be interpreted as a valid floating point number.")));
|
|
return *this;
|
|
}
|
|
void swap(gmp_float_imp& o) BOOST_NOEXCEPT
|
|
{
|
|
mpf_swap(m_data, o.m_data);
|
|
}
|
|
std::string str(std::streamsize digits, std::ios_base::fmtflags f)const
|
|
{
|
|
BOOST_ASSERT(m_data[0]._mp_d);
|
|
|
|
bool scientific = (f & std::ios_base::scientific) == std::ios_base::scientific;
|
|
bool fixed = (f & std::ios_base::fixed) == std::ios_base::fixed;
|
|
std::streamsize org_digits(digits);
|
|
|
|
if(scientific && digits)
|
|
++digits;
|
|
|
|
std::string result;
|
|
mp_exp_t e;
|
|
void *(*alloc_func_ptr) (size_t);
|
|
void *(*realloc_func_ptr) (void *, size_t, size_t);
|
|
void (*free_func_ptr) (void *, size_t);
|
|
mp_get_memory_functions(&alloc_func_ptr, &realloc_func_ptr, &free_func_ptr);
|
|
|
|
if(mpf_sgn(m_data) == 0)
|
|
{
|
|
e = 0;
|
|
result = "0";
|
|
if(fixed && digits)
|
|
++digits;
|
|
}
|
|
else
|
|
{
|
|
char* ps = mpf_get_str (0, &e, 10, static_cast<std::size_t>(digits), m_data);
|
|
--e; // To match with what our formatter expects.
|
|
if(fixed && e != -1)
|
|
{
|
|
// Oops we actually need a different number of digits to what we asked for:
|
|
(*free_func_ptr)((void*)ps, std::strlen(ps) + 1);
|
|
digits += e + 1;
|
|
if(digits == 0)
|
|
{
|
|
// We need to get *all* the digits and then possibly round up,
|
|
// we end up with either "0" or "1" as the result.
|
|
ps = mpf_get_str (0, &e, 10, 0, m_data);
|
|
--e;
|
|
unsigned offset = *ps == '-' ? 1 : 0;
|
|
if(ps[offset] > '5')
|
|
{
|
|
++e;
|
|
ps[offset] = '1';
|
|
ps[offset + 1] = 0;
|
|
}
|
|
else if(ps[offset] == '5')
|
|
{
|
|
unsigned i = offset + 1;
|
|
bool round_up = false;
|
|
while(ps[i] != 0)
|
|
{
|
|
if(ps[i] != '0')
|
|
{
|
|
round_up = true;
|
|
break;
|
|
}
|
|
}
|
|
if(round_up)
|
|
{
|
|
++e;
|
|
ps[offset] = '1';
|
|
ps[offset + 1] = 0;
|
|
}
|
|
else
|
|
{
|
|
ps[offset] = '0';
|
|
ps[offset + 1] = 0;
|
|
}
|
|
}
|
|
else
|
|
{
|
|
ps[offset] = '0';
|
|
ps[offset + 1] = 0;
|
|
}
|
|
}
|
|
else if(digits > 0)
|
|
{
|
|
ps = mpf_get_str (0, &e, 10, static_cast<std::size_t>(digits), m_data);
|
|
--e; // To match with what our formatter expects.
|
|
}
|
|
else
|
|
{
|
|
ps = mpf_get_str (0, &e, 10, 1, m_data);
|
|
--e;
|
|
unsigned offset = *ps == '-' ? 1 : 0;
|
|
ps[offset] = '0';
|
|
ps[offset + 1] = 0;
|
|
}
|
|
}
|
|
result = ps;
|
|
(*free_func_ptr)((void*)ps, std::strlen(ps) + 1);
|
|
}
|
|
boost::multiprecision::detail::format_float_string(result, e, org_digits, f, mpf_sgn(m_data) == 0);
|
|
return result;
|
|
}
|
|
~gmp_float_imp() BOOST_NOEXCEPT
|
|
{
|
|
if(m_data[0]._mp_d)
|
|
mpf_clear(m_data);
|
|
}
|
|
void negate() BOOST_NOEXCEPT
|
|
{
|
|
BOOST_ASSERT(m_data[0]._mp_d);
|
|
mpf_neg(m_data, m_data);
|
|
}
|
|
int compare(const gmp_float<digits10>& o)const BOOST_NOEXCEPT
|
|
{
|
|
BOOST_ASSERT(m_data[0]._mp_d && o.m_data[0]._mp_d);
|
|
return mpf_cmp(m_data, o.m_data);
|
|
}
|
|
int compare(long i)const BOOST_NOEXCEPT
|
|
{
|
|
BOOST_ASSERT(m_data[0]._mp_d);
|
|
return mpf_cmp_si(m_data, i);
|
|
}
|
|
int compare(unsigned long i)const BOOST_NOEXCEPT
|
|
{
|
|
BOOST_ASSERT(m_data[0]._mp_d);
|
|
return mpf_cmp_ui(m_data, i);
|
|
}
|
|
template <class V>
|
|
typename enable_if<is_arithmetic<V>, int>::type compare(V v)const
|
|
{
|
|
gmp_float<digits10> d;
|
|
d = v;
|
|
return compare(d);
|
|
}
|
|
mpf_t& data() BOOST_NOEXCEPT
|
|
{
|
|
BOOST_ASSERT(m_data[0]._mp_d);
|
|
return m_data;
|
|
}
|
|
const mpf_t& data()const BOOST_NOEXCEPT
|
|
{
|
|
BOOST_ASSERT(m_data[0]._mp_d);
|
|
return m_data;
|
|
}
|
|
protected:
|
|
mpf_t m_data;
|
|
static unsigned& get_default_precision() BOOST_NOEXCEPT
|
|
{
|
|
static unsigned val = 50;
|
|
return val;
|
|
}
|
|
};
|
|
|
|
} // namespace detail
|
|
|
|
struct gmp_int;
|
|
struct gmp_rational;
|
|
|
|
template <unsigned digits10>
|
|
struct gmp_float : public detail::gmp_float_imp<digits10>
|
|
{
|
|
gmp_float()
|
|
{
|
|
mpf_init2(this->m_data, multiprecision::detail::digits10_2_2(digits10));
|
|
}
|
|
gmp_float(const gmp_float& o) : detail::gmp_float_imp<digits10>(o) {}
|
|
template <unsigned D>
|
|
gmp_float(const gmp_float<D>& o, typename enable_if_c<D <= digits10>::type* = 0);
|
|
template <unsigned D>
|
|
explicit gmp_float(const gmp_float<D>& o, typename disable_if_c<D <= digits10>::type* = 0);
|
|
gmp_float(const gmp_int& o);
|
|
gmp_float(const gmp_rational& o);
|
|
gmp_float(const mpf_t val)
|
|
{
|
|
mpf_init2(this->m_data, multiprecision::detail::digits10_2_2(digits10));
|
|
mpf_set(this->m_data, val);
|
|
}
|
|
gmp_float(const mpz_t val)
|
|
{
|
|
mpf_init2(this->m_data, multiprecision::detail::digits10_2_2(digits10));
|
|
mpf_set_z(this->m_data, val);
|
|
}
|
|
gmp_float(const mpq_t val)
|
|
{
|
|
mpf_init2(this->m_data, multiprecision::detail::digits10_2_2(digits10));
|
|
mpf_set_q(this->m_data, val);
|
|
}
|
|
#ifndef BOOST_NO_CXX11_RVALUE_REFERENCES
|
|
gmp_float(gmp_float&& o) BOOST_NOEXCEPT : detail::gmp_float_imp<digits10>(static_cast<detail::gmp_float_imp<digits10>&&>(o)) {}
|
|
#endif
|
|
gmp_float& operator=(const gmp_float& o)
|
|
{
|
|
*static_cast<detail::gmp_float_imp<digits10>*>(this) = static_cast<detail::gmp_float_imp<digits10> const&>(o);
|
|
return *this;
|
|
}
|
|
#ifndef BOOST_NO_CXX11_RVALUE_REFERENCES
|
|
gmp_float& operator=(gmp_float&& o) BOOST_NOEXCEPT
|
|
{
|
|
*static_cast<detail::gmp_float_imp<digits10>*>(this) = static_cast<detail::gmp_float_imp<digits10>&&>(o);
|
|
return *this;
|
|
}
|
|
#endif
|
|
template <unsigned D>
|
|
gmp_float& operator=(const gmp_float<D>& o);
|
|
gmp_float& operator=(const gmp_int& o);
|
|
gmp_float& operator=(const gmp_rational& o);
|
|
gmp_float& operator=(const mpf_t val)
|
|
{
|
|
if(this->m_data[0]._mp_d == 0)
|
|
mpf_init2(this->m_data, multiprecision::detail::digits10_2_2(digits10));
|
|
mpf_set(this->m_data, val);
|
|
return *this;
|
|
}
|
|
gmp_float& operator=(const mpz_t val)
|
|
{
|
|
if(this->m_data[0]._mp_d == 0)
|
|
mpf_init2(this->m_data, multiprecision::detail::digits10_2_2(digits10));
|
|
mpf_set_z(this->m_data, val);
|
|
return *this;
|
|
}
|
|
gmp_float& operator=(const mpq_t val)
|
|
{
|
|
if(this->m_data[0]._mp_d == 0)
|
|
mpf_init2(this->m_data, multiprecision::detail::digits10_2_2(digits10));
|
|
mpf_set_q(this->m_data, val);
|
|
return *this;
|
|
}
|
|
template <class V>
|
|
gmp_float& operator=(const V& v)
|
|
{
|
|
*static_cast<detail::gmp_float_imp<digits10>*>(this) = v;
|
|
return *this;
|
|
}
|
|
};
|
|
|
|
template <>
|
|
struct gmp_float<0> : public detail::gmp_float_imp<0>
|
|
{
|
|
gmp_float()
|
|
{
|
|
mpf_init2(this->m_data, multiprecision::detail::digits10_2_2(get_default_precision()));
|
|
}
|
|
gmp_float(const mpf_t val)
|
|
{
|
|
mpf_init2(this->m_data, multiprecision::detail::digits10_2_2(get_default_precision()));
|
|
mpf_set(this->m_data, val);
|
|
}
|
|
gmp_float(const mpz_t val)
|
|
{
|
|
mpf_init2(this->m_data, multiprecision::detail::digits10_2_2(get_default_precision()));
|
|
mpf_set_z(this->m_data, val);
|
|
}
|
|
gmp_float(const mpq_t val)
|
|
{
|
|
mpf_init2(this->m_data, multiprecision::detail::digits10_2_2(get_default_precision()));
|
|
mpf_set_q(this->m_data, val);
|
|
}
|
|
gmp_float(const gmp_float& o) : detail::gmp_float_imp<0>(o) {}
|
|
template <unsigned D>
|
|
gmp_float(const gmp_float<D>& o)
|
|
{
|
|
mpf_init2(this->m_data, mpf_get_prec(o.data()));
|
|
mpf_set(this->m_data, o.data());
|
|
}
|
|
#ifndef BOOST_NO_CXX11_RVALUE_REFERENCES
|
|
gmp_float(gmp_float&& o) BOOST_NOEXCEPT : detail::gmp_float_imp<0>(static_cast<detail::gmp_float_imp<0>&&>(o)) {}
|
|
#endif
|
|
gmp_float(const gmp_int& o);
|
|
gmp_float(const gmp_rational& o);
|
|
gmp_float(const gmp_float& o, unsigned digits10)
|
|
{
|
|
mpf_init2(this->m_data, multiprecision::detail::digits10_2_2(digits10));
|
|
mpf_set(this->m_data, o.data());
|
|
}
|
|
|
|
gmp_float& operator=(const gmp_float& o)
|
|
{
|
|
*static_cast<detail::gmp_float_imp<0>*>(this) = static_cast<detail::gmp_float_imp<0> const&>(o);
|
|
return *this;
|
|
}
|
|
#ifndef BOOST_NO_CXX11_RVALUE_REFERENCES
|
|
gmp_float& operator=(gmp_float&& o) BOOST_NOEXCEPT
|
|
{
|
|
*static_cast<detail::gmp_float_imp<0>*>(this) = static_cast<detail::gmp_float_imp<0> &&>(o);
|
|
return *this;
|
|
}
|
|
#endif
|
|
template <unsigned D>
|
|
gmp_float& operator=(const gmp_float<D>& o)
|
|
{
|
|
if(this->m_data[0]._mp_d == 0)
|
|
{
|
|
mpf_init2(this->m_data, mpf_get_prec(o.data()));
|
|
}
|
|
else
|
|
{
|
|
mpf_set_prec(this->m_data, mpf_get_prec(o.data()));
|
|
}
|
|
mpf_set(this->m_data, o.data());
|
|
return *this;
|
|
}
|
|
gmp_float& operator=(const gmp_int& o);
|
|
gmp_float& operator=(const gmp_rational& o);
|
|
gmp_float& operator=(const mpf_t val)
|
|
{
|
|
if(this->m_data[0]._mp_d == 0)
|
|
mpf_init2(this->m_data, multiprecision::detail::digits10_2_2(get_default_precision()));
|
|
mpf_set(this->m_data, val);
|
|
return *this;
|
|
}
|
|
gmp_float& operator=(const mpz_t val)
|
|
{
|
|
if(this->m_data[0]._mp_d == 0)
|
|
mpf_init2(this->m_data, multiprecision::detail::digits10_2_2(get_default_precision()));
|
|
mpf_set_z(this->m_data, val);
|
|
return *this;
|
|
}
|
|
gmp_float& operator=(const mpq_t val)
|
|
{
|
|
if(this->m_data[0]._mp_d == 0)
|
|
mpf_init2(this->m_data, multiprecision::detail::digits10_2_2(get_default_precision()));
|
|
mpf_set_q(this->m_data, val);
|
|
return *this;
|
|
}
|
|
template <class V>
|
|
gmp_float& operator=(const V& v)
|
|
{
|
|
*static_cast<detail::gmp_float_imp<0>*>(this) = v;
|
|
return *this;
|
|
}
|
|
static unsigned default_precision() BOOST_NOEXCEPT
|
|
{
|
|
return get_default_precision();
|
|
}
|
|
static void default_precision(unsigned v) BOOST_NOEXCEPT
|
|
{
|
|
get_default_precision() = v;
|
|
}
|
|
unsigned precision()const BOOST_NOEXCEPT
|
|
{
|
|
return static_cast<unsigned>(multiprecision::detail::digits2_2_10(static_cast<unsigned long>(mpf_get_prec(this->m_data))));
|
|
}
|
|
void precision(unsigned digits10) BOOST_NOEXCEPT
|
|
{
|
|
mpf_set_prec(this->m_data, multiprecision::detail::digits10_2_2(digits10));
|
|
}
|
|
};
|
|
|
|
template <unsigned digits10, class T>
|
|
inline typename enable_if_c<is_arithmetic<T>::value, bool>::type eval_eq(const gmp_float<digits10>& a, const T& b) BOOST_NOEXCEPT
|
|
{
|
|
return a.compare(b) == 0;
|
|
}
|
|
template <unsigned digits10, class T>
|
|
inline typename enable_if_c<is_arithmetic<T>::value, bool>::type eval_lt(const gmp_float<digits10>& a, const T& b) BOOST_NOEXCEPT
|
|
{
|
|
return a.compare(b) < 0;
|
|
}
|
|
template <unsigned digits10, class T>
|
|
inline typename enable_if_c<is_arithmetic<T>::value, bool>::type eval_gt(const gmp_float<digits10>& a, const T& b) BOOST_NOEXCEPT
|
|
{
|
|
return a.compare(b) > 0;
|
|
}
|
|
|
|
template <unsigned D1, unsigned D2>
|
|
inline void eval_add(gmp_float<D1>& result, const gmp_float<D2>& o)
|
|
{
|
|
mpf_add(result.data(), result.data(), o.data());
|
|
}
|
|
template <unsigned D1, unsigned D2>
|
|
inline void eval_subtract(gmp_float<D1>& result, const gmp_float<D2>& o)
|
|
{
|
|
mpf_sub(result.data(), result.data(), o.data());
|
|
}
|
|
template <unsigned D1, unsigned D2>
|
|
inline void eval_multiply(gmp_float<D1>& result, const gmp_float<D2>& o)
|
|
{
|
|
mpf_mul(result.data(), result.data(), o.data());
|
|
}
|
|
template <unsigned digits10>
|
|
inline bool eval_is_zero(const gmp_float<digits10>& val) BOOST_NOEXCEPT
|
|
{
|
|
return mpf_sgn(val.data()) == 0;
|
|
}
|
|
template <unsigned D1, unsigned D2>
|
|
inline void eval_divide(gmp_float<D1>& result, const gmp_float<D2>& o)
|
|
{
|
|
if(eval_is_zero(o))
|
|
BOOST_THROW_EXCEPTION(std::overflow_error("Division by zero."));
|
|
mpf_div(result.data(), result.data(), o.data());
|
|
}
|
|
template <unsigned digits10>
|
|
inline void eval_add(gmp_float<digits10>& result, unsigned long i)
|
|
{
|
|
mpf_add_ui(result.data(), result.data(), i);
|
|
}
|
|
template <unsigned digits10>
|
|
inline void eval_subtract(gmp_float<digits10>& result, unsigned long i)
|
|
{
|
|
mpf_sub_ui(result.data(), result.data(), i);
|
|
}
|
|
template <unsigned digits10>
|
|
inline void eval_multiply(gmp_float<digits10>& result, unsigned long i)
|
|
{
|
|
mpf_mul_ui(result.data(), result.data(), i);
|
|
}
|
|
template <unsigned digits10>
|
|
inline void eval_divide(gmp_float<digits10>& result, unsigned long i)
|
|
{
|
|
if(i == 0)
|
|
BOOST_THROW_EXCEPTION(std::overflow_error("Division by zero."));
|
|
mpf_div_ui(result.data(), result.data(), i);
|
|
}
|
|
template <unsigned digits10>
|
|
inline void eval_add(gmp_float<digits10>& result, long i)
|
|
{
|
|
if(i > 0)
|
|
mpf_add_ui(result.data(), result.data(), i);
|
|
else
|
|
mpf_sub_ui(result.data(), result.data(), boost::multiprecision::detail::unsigned_abs(i));
|
|
}
|
|
template <unsigned digits10>
|
|
inline void eval_subtract(gmp_float<digits10>& result, long i)
|
|
{
|
|
if(i > 0)
|
|
mpf_sub_ui(result.data(), result.data(), i);
|
|
else
|
|
mpf_add_ui(result.data(), result.data(), boost::multiprecision::detail::unsigned_abs(i));
|
|
}
|
|
template <unsigned digits10>
|
|
inline void eval_multiply(gmp_float<digits10>& result, long i)
|
|
{
|
|
mpf_mul_ui(result.data(), result.data(), boost::multiprecision::detail::unsigned_abs(i));
|
|
if(i < 0)
|
|
mpf_neg(result.data(), result.data());
|
|
}
|
|
template <unsigned digits10>
|
|
inline void eval_divide(gmp_float<digits10>& result, long i)
|
|
{
|
|
if(i == 0)
|
|
BOOST_THROW_EXCEPTION(std::overflow_error("Division by zero."));
|
|
mpf_div_ui(result.data(), result.data(), boost::multiprecision::detail::unsigned_abs(i));
|
|
if(i < 0)
|
|
mpf_neg(result.data(), result.data());
|
|
}
|
|
//
|
|
// Specialised 3 arg versions of the basic operators:
|
|
//
|
|
template <unsigned D1, unsigned D2, unsigned D3>
|
|
inline void eval_add(gmp_float<D1>& a, const gmp_float<D2>& x, const gmp_float<D3>& y)
|
|
{
|
|
mpf_add(a.data(), x.data(), y.data());
|
|
}
|
|
template <unsigned D1, unsigned D2>
|
|
inline void eval_add(gmp_float<D1>& a, const gmp_float<D2>& x, unsigned long y)
|
|
{
|
|
mpf_add_ui(a.data(), x.data(), y);
|
|
}
|
|
template <unsigned D1, unsigned D2>
|
|
inline void eval_add(gmp_float<D1>& a, const gmp_float<D2>& x, long y)
|
|
{
|
|
if(y < 0)
|
|
mpf_sub_ui(a.data(), x.data(), boost::multiprecision::detail::unsigned_abs(y));
|
|
else
|
|
mpf_add_ui(a.data(), x.data(), y);
|
|
}
|
|
template <unsigned D1, unsigned D2>
|
|
inline void eval_add(gmp_float<D1>& a, unsigned long x, const gmp_float<D2>& y)
|
|
{
|
|
mpf_add_ui(a.data(), y.data(), x);
|
|
}
|
|
template <unsigned D1, unsigned D2>
|
|
inline void eval_add(gmp_float<D1>& a, long x, const gmp_float<D2>& y)
|
|
{
|
|
if(x < 0)
|
|
{
|
|
mpf_ui_sub(a.data(), boost::multiprecision::detail::unsigned_abs(x), y.data());
|
|
mpf_neg(a.data(), a.data());
|
|
}
|
|
else
|
|
mpf_add_ui(a.data(), y.data(), x);
|
|
}
|
|
template <unsigned D1, unsigned D2, unsigned D3>
|
|
inline void eval_subtract(gmp_float<D1>& a, const gmp_float<D2>& x, const gmp_float<D3>& y)
|
|
{
|
|
mpf_sub(a.data(), x.data(), y.data());
|
|
}
|
|
template <unsigned D1, unsigned D2>
|
|
inline void eval_subtract(gmp_float<D1>& a, const gmp_float<D2>& x, unsigned long y)
|
|
{
|
|
mpf_sub_ui(a.data(), x.data(), y);
|
|
}
|
|
template <unsigned D1, unsigned D2>
|
|
inline void eval_subtract(gmp_float<D1>& a, const gmp_float<D2>& x, long y)
|
|
{
|
|
if(y < 0)
|
|
mpf_add_ui(a.data(), x.data(), boost::multiprecision::detail::unsigned_abs(y));
|
|
else
|
|
mpf_sub_ui(a.data(), x.data(), y);
|
|
}
|
|
template <unsigned D1, unsigned D2>
|
|
inline void eval_subtract(gmp_float<D1>& a, unsigned long x, const gmp_float<D2>& y)
|
|
{
|
|
mpf_ui_sub(a.data(), x, y.data());
|
|
}
|
|
template <unsigned D1, unsigned D2>
|
|
inline void eval_subtract(gmp_float<D1>& a, long x, const gmp_float<D2>& y)
|
|
{
|
|
if(x < 0)
|
|
{
|
|
mpf_add_ui(a.data(), y.data(), boost::multiprecision::detail::unsigned_abs(x));
|
|
mpf_neg(a.data(), a.data());
|
|
}
|
|
else
|
|
mpf_ui_sub(a.data(), x, y.data());
|
|
}
|
|
|
|
template <unsigned D1, unsigned D2, unsigned D3>
|
|
inline void eval_multiply(gmp_float<D1>& a, const gmp_float<D2>& x, const gmp_float<D3>& y)
|
|
{
|
|
mpf_mul(a.data(), x.data(), y.data());
|
|
}
|
|
template <unsigned D1, unsigned D2>
|
|
inline void eval_multiply(gmp_float<D1>& a, const gmp_float<D2>& x, unsigned long y)
|
|
{
|
|
mpf_mul_ui(a.data(), x.data(), y);
|
|
}
|
|
template <unsigned D1, unsigned D2>
|
|
inline void eval_multiply(gmp_float<D1>& a, const gmp_float<D2>& x, long y)
|
|
{
|
|
if(y < 0)
|
|
{
|
|
mpf_mul_ui(a.data(), x.data(), boost::multiprecision::detail::unsigned_abs(y));
|
|
a.negate();
|
|
}
|
|
else
|
|
mpf_mul_ui(a.data(), x.data(), y);
|
|
}
|
|
template <unsigned D1, unsigned D2>
|
|
inline void eval_multiply(gmp_float<D1>& a, unsigned long x, const gmp_float<D2>& y)
|
|
{
|
|
mpf_mul_ui(a.data(), y.data(), x);
|
|
}
|
|
template <unsigned D1, unsigned D2>
|
|
inline void eval_multiply(gmp_float<D1>& a, long x, const gmp_float<D2>& y)
|
|
{
|
|
if(x < 0)
|
|
{
|
|
mpf_mul_ui(a.data(), y.data(), boost::multiprecision::detail::unsigned_abs(x));
|
|
mpf_neg(a.data(), a.data());
|
|
}
|
|
else
|
|
mpf_mul_ui(a.data(), y.data(), x);
|
|
}
|
|
|
|
template <unsigned D1, unsigned D2, unsigned D3>
|
|
inline void eval_divide(gmp_float<D1>& a, const gmp_float<D2>& x, const gmp_float<D3>& y)
|
|
{
|
|
if(eval_is_zero(y))
|
|
BOOST_THROW_EXCEPTION(std::overflow_error("Division by zero."));
|
|
mpf_div(a.data(), x.data(), y.data());
|
|
}
|
|
template <unsigned D1, unsigned D2>
|
|
inline void eval_divide(gmp_float<D1>& a, const gmp_float<D2>& x, unsigned long y)
|
|
{
|
|
if(y == 0)
|
|
BOOST_THROW_EXCEPTION(std::overflow_error("Division by zero."));
|
|
mpf_div_ui(a.data(), x.data(), y);
|
|
}
|
|
template <unsigned D1, unsigned D2>
|
|
inline void eval_divide(gmp_float<D1>& a, const gmp_float<D2>& x, long y)
|
|
{
|
|
if(y == 0)
|
|
BOOST_THROW_EXCEPTION(std::overflow_error("Division by zero."));
|
|
if(y < 0)
|
|
{
|
|
mpf_div_ui(a.data(), x.data(), boost::multiprecision::detail::unsigned_abs(y));
|
|
a.negate();
|
|
}
|
|
else
|
|
mpf_div_ui(a.data(), x.data(), y);
|
|
}
|
|
template <unsigned D1, unsigned D2>
|
|
inline void eval_divide(gmp_float<D1>& a, unsigned long x, const gmp_float<D2>& y)
|
|
{
|
|
if(eval_is_zero(y))
|
|
BOOST_THROW_EXCEPTION(std::overflow_error("Division by zero."));
|
|
mpf_ui_div(a.data(), x, y.data());
|
|
}
|
|
template <unsigned D1, unsigned D2>
|
|
inline void eval_divide(gmp_float<D1>& a, long x, const gmp_float<D2>& y)
|
|
{
|
|
if(eval_is_zero(y))
|
|
BOOST_THROW_EXCEPTION(std::overflow_error("Division by zero."));
|
|
if(x < 0)
|
|
{
|
|
mpf_ui_div(a.data(), boost::multiprecision::detail::unsigned_abs(x), y.data());
|
|
mpf_neg(a.data(), a.data());
|
|
}
|
|
else
|
|
mpf_ui_div(a.data(), x, y.data());
|
|
}
|
|
|
|
template <unsigned digits10>
|
|
inline int eval_get_sign(const gmp_float<digits10>& val) BOOST_NOEXCEPT
|
|
{
|
|
return mpf_sgn(val.data());
|
|
}
|
|
|
|
template <unsigned digits10>
|
|
inline void eval_convert_to(unsigned long* result, const gmp_float<digits10>& val) BOOST_NOEXCEPT
|
|
{
|
|
if(0 == mpf_fits_ulong_p(val.data()))
|
|
*result = (std::numeric_limits<unsigned long>::max)();
|
|
else
|
|
*result = mpf_get_ui(val.data());
|
|
}
|
|
template <unsigned digits10>
|
|
inline void eval_convert_to(long* result, const gmp_float<digits10>& val) BOOST_NOEXCEPT
|
|
{
|
|
if(0 == mpf_fits_slong_p(val.data()))
|
|
{
|
|
*result = (std::numeric_limits<unsigned long>::max)();
|
|
*result *= mpf_sgn(val.data());
|
|
}
|
|
else
|
|
*result = mpf_get_si(val.data());
|
|
}
|
|
template <unsigned digits10>
|
|
inline void eval_convert_to(double* result, const gmp_float<digits10>& val) BOOST_NOEXCEPT
|
|
{
|
|
*result = mpf_get_d(val.data());
|
|
}
|
|
#ifdef BOOST_HAS_LONG_LONG
|
|
template <unsigned digits10>
|
|
inline void eval_convert_to(boost::long_long_type* result, const gmp_float<digits10>& val)
|
|
{
|
|
gmp_float<digits10> t(val);
|
|
if(eval_get_sign(t) < 0)
|
|
t.negate();
|
|
|
|
long digits = std::numeric_limits<boost::long_long_type>::digits - std::numeric_limits<long>::digits;
|
|
|
|
if(digits > 0)
|
|
mpf_div_2exp(t.data(), t.data(), digits);
|
|
|
|
if(!mpf_fits_slong_p(t.data()))
|
|
{
|
|
if(eval_get_sign(val) < 0)
|
|
*result = (std::numeric_limits<boost::long_long_type>::min)();
|
|
else
|
|
*result = (std::numeric_limits<boost::long_long_type>::max)();
|
|
return;
|
|
};
|
|
|
|
*result = mpf_get_si(t.data());
|
|
while(digits > 0)
|
|
{
|
|
*result <<= digits;
|
|
digits -= std::numeric_limits<unsigned long>::digits;
|
|
mpf_mul_2exp(t.data(), t.data(), digits >= 0 ? std::numeric_limits<unsigned long>::digits : std::numeric_limits<unsigned long>::digits + digits);
|
|
unsigned long l = mpf_get_ui(t.data());
|
|
if(digits < 0)
|
|
l >>= -digits;
|
|
*result |= l;
|
|
}
|
|
if(eval_get_sign(val) < 0)
|
|
*result = -*result;
|
|
}
|
|
template <unsigned digits10>
|
|
inline void eval_convert_to(boost::ulong_long_type* result, const gmp_float<digits10>& val)
|
|
{
|
|
gmp_float<digits10> t(val);
|
|
|
|
long digits = std::numeric_limits<boost::long_long_type>::digits - std::numeric_limits<long>::digits;
|
|
|
|
if(digits > 0)
|
|
mpf_div_2exp(t.data(), t.data(), digits);
|
|
|
|
if(!mpf_fits_ulong_p(t.data()))
|
|
{
|
|
*result = (std::numeric_limits<boost::long_long_type>::max)();
|
|
return;
|
|
}
|
|
|
|
*result = mpf_get_ui(t.data());
|
|
while(digits > 0)
|
|
{
|
|
*result <<= digits;
|
|
digits -= std::numeric_limits<unsigned long>::digits;
|
|
mpf_mul_2exp(t.data(), t.data(), digits >= 0 ? std::numeric_limits<unsigned long>::digits : std::numeric_limits<unsigned long>::digits + digits);
|
|
unsigned long l = mpf_get_ui(t.data());
|
|
if(digits < 0)
|
|
l >>= -digits;
|
|
*result |= l;
|
|
}
|
|
}
|
|
#endif
|
|
|
|
//
|
|
// Native non-member operations:
|
|
//
|
|
template <unsigned Digits10>
|
|
inline void eval_sqrt(gmp_float<Digits10>& result, const gmp_float<Digits10>& val)
|
|
{
|
|
mpf_sqrt(result.data(), val.data());
|
|
}
|
|
|
|
template <unsigned Digits10>
|
|
inline void eval_abs(gmp_float<Digits10>& result, const gmp_float<Digits10>& val)
|
|
{
|
|
mpf_abs(result.data(), val.data());
|
|
}
|
|
|
|
template <unsigned Digits10>
|
|
inline void eval_fabs(gmp_float<Digits10>& result, const gmp_float<Digits10>& val)
|
|
{
|
|
mpf_abs(result.data(), val.data());
|
|
}
|
|
template <unsigned Digits10>
|
|
inline void eval_ceil(gmp_float<Digits10>& result, const gmp_float<Digits10>& val)
|
|
{
|
|
mpf_ceil(result.data(), val.data());
|
|
}
|
|
template <unsigned Digits10>
|
|
inline void eval_floor(gmp_float<Digits10>& result, const gmp_float<Digits10>& val)
|
|
{
|
|
mpf_floor(result.data(), val.data());
|
|
}
|
|
template <unsigned Digits10>
|
|
inline void eval_trunc(gmp_float<Digits10>& result, const gmp_float<Digits10>& val)
|
|
{
|
|
mpf_trunc(result.data(), val.data());
|
|
}
|
|
template <unsigned Digits10>
|
|
inline void eval_ldexp(gmp_float<Digits10>& result, const gmp_float<Digits10>& val, long e)
|
|
{
|
|
if(e > 0)
|
|
mpf_mul_2exp(result.data(), val.data(), e);
|
|
else if(e < 0)
|
|
mpf_div_2exp(result.data(), val.data(), -e);
|
|
else
|
|
result = val;
|
|
}
|
|
template <unsigned Digits10>
|
|
inline void eval_frexp(gmp_float<Digits10>& result, const gmp_float<Digits10>& val, int* e)
|
|
{
|
|
#if BOOST_MP_MPIR_VERSION >= 20600
|
|
mpir_si v;
|
|
mpf_get_d_2exp(&v, val.data());
|
|
#else
|
|
long v;
|
|
mpf_get_d_2exp(&v, val.data());
|
|
#endif
|
|
*e = v;
|
|
eval_ldexp(result, val, -v);
|
|
}
|
|
template <unsigned Digits10>
|
|
inline void eval_frexp(gmp_float<Digits10>& result, const gmp_float<Digits10>& val, long* e)
|
|
{
|
|
#if BOOST_MP_MPIR_VERSION >= 20600
|
|
mpir_si v;
|
|
mpf_get_d_2exp(&v, val.data());
|
|
*e = v;
|
|
eval_ldexp(result, val, -v);
|
|
#else
|
|
mpf_get_d_2exp(e, val.data());
|
|
eval_ldexp(result, val, -*e);
|
|
#endif
|
|
}
|
|
|
|
template <unsigned Digits10>
|
|
inline std::size_t hash_value(const gmp_float<Digits10>& val)
|
|
{
|
|
std::size_t result = 0;
|
|
for(int i = 0; i < std::abs(val.data()[0]._mp_size); ++i)
|
|
boost::hash_combine(result, val.data()[0]._mp_d[i]);
|
|
boost::hash_combine(result, val.data()[0]._mp_exp);
|
|
boost::hash_combine(result, val.data()[0]._mp_size);
|
|
return result;
|
|
}
|
|
|
|
struct gmp_int
|
|
{
|
|
#ifdef BOOST_HAS_LONG_LONG
|
|
typedef mpl::list<long, boost::long_long_type> signed_types;
|
|
typedef mpl::list<unsigned long, boost::ulong_long_type> unsigned_types;
|
|
#else
|
|
typedef mpl::list<long> signed_types;
|
|
typedef mpl::list<unsigned long> unsigned_types;
|
|
#endif
|
|
typedef mpl::list<double, long double> float_types;
|
|
|
|
gmp_int()
|
|
{
|
|
mpz_init(this->m_data);
|
|
}
|
|
gmp_int(const gmp_int& o)
|
|
{
|
|
if(o.m_data[0]._mp_d)
|
|
mpz_init_set(m_data, o.m_data);
|
|
else
|
|
mpz_init(this->m_data);
|
|
}
|
|
#ifndef BOOST_NO_CXX11_RVALUE_REFERENCES
|
|
gmp_int(gmp_int&& o) BOOST_NOEXCEPT
|
|
{
|
|
m_data[0] = o.m_data[0];
|
|
o.m_data[0]._mp_d = 0;
|
|
}
|
|
#endif
|
|
explicit gmp_int(const mpf_t val)
|
|
{
|
|
mpz_init(this->m_data);
|
|
mpz_set_f(this->m_data, val);
|
|
}
|
|
gmp_int(const mpz_t val)
|
|
{
|
|
mpz_init_set(this->m_data, val);
|
|
}
|
|
explicit gmp_int(const mpq_t val)
|
|
{
|
|
mpz_init(this->m_data);
|
|
mpz_set_q(this->m_data, val);
|
|
}
|
|
template <unsigned Digits10>
|
|
explicit gmp_int(const gmp_float<Digits10>& o)
|
|
{
|
|
mpz_init(this->m_data);
|
|
mpz_set_f(this->m_data, o.data());
|
|
}
|
|
explicit gmp_int(const gmp_rational& o);
|
|
gmp_int& operator = (const gmp_int& o)
|
|
{
|
|
if(m_data[0]._mp_d == 0)
|
|
mpz_init(this->m_data);
|
|
mpz_set(m_data, o.m_data);
|
|
return *this;
|
|
}
|
|
#ifndef BOOST_NO_CXX11_RVALUE_REFERENCES
|
|
gmp_int& operator = (gmp_int&& o) BOOST_NOEXCEPT
|
|
{
|
|
mpz_swap(m_data, o.m_data);
|
|
return *this;
|
|
}
|
|
#endif
|
|
#ifdef BOOST_HAS_LONG_LONG
|
|
#if defined(ULLONG_MAX) && (ULLONG_MAX == ULONG_MAX)
|
|
gmp_int& operator = (boost::ulong_long_type i)
|
|
{
|
|
*this = static_cast<unsigned long>(i);
|
|
return *this;
|
|
}
|
|
#else
|
|
gmp_int& operator = (boost::ulong_long_type i)
|
|
{
|
|
if(m_data[0]._mp_d == 0)
|
|
mpz_init(this->m_data);
|
|
boost::ulong_long_type mask = ((((1uLL << (std::numeric_limits<unsigned long>::digits - 1)) - 1) << 1) | 1uLL);
|
|
unsigned shift = 0;
|
|
mpz_t t;
|
|
mpz_set_ui(m_data, 0);
|
|
mpz_init_set_ui(t, 0);
|
|
while(i)
|
|
{
|
|
mpz_set_ui(t, static_cast<unsigned long>(i & mask));
|
|
if(shift)
|
|
mpz_mul_2exp(t, t, shift);
|
|
mpz_add(m_data, m_data, t);
|
|
shift += std::numeric_limits<unsigned long>::digits;
|
|
i >>= std::numeric_limits<unsigned long>::digits;
|
|
}
|
|
mpz_clear(t);
|
|
return *this;
|
|
}
|
|
#endif
|
|
gmp_int& operator = (boost::long_long_type i)
|
|
{
|
|
if(m_data[0]._mp_d == 0)
|
|
mpz_init(this->m_data);
|
|
bool neg = i < 0;
|
|
*this = boost::multiprecision::detail::unsigned_abs(i);
|
|
if(neg)
|
|
mpz_neg(m_data, m_data);
|
|
return *this;
|
|
}
|
|
#endif
|
|
gmp_int& operator = (unsigned long i)
|
|
{
|
|
if(m_data[0]._mp_d == 0)
|
|
mpz_init(this->m_data);
|
|
mpz_set_ui(m_data, i);
|
|
return *this;
|
|
}
|
|
gmp_int& operator = (long i)
|
|
{
|
|
if(m_data[0]._mp_d == 0)
|
|
mpz_init(this->m_data);
|
|
mpz_set_si(m_data, i);
|
|
return *this;
|
|
}
|
|
gmp_int& operator = (double d)
|
|
{
|
|
if(m_data[0]._mp_d == 0)
|
|
mpz_init(this->m_data);
|
|
mpz_set_d(m_data, d);
|
|
return *this;
|
|
}
|
|
gmp_int& operator = (long double a)
|
|
{
|
|
using std::frexp;
|
|
using std::ldexp;
|
|
using std::floor;
|
|
|
|
if(m_data[0]._mp_d == 0)
|
|
mpz_init(this->m_data);
|
|
|
|
if (a == 0) {
|
|
mpz_set_si(m_data, 0);
|
|
return *this;
|
|
}
|
|
|
|
if (a == 1) {
|
|
mpz_set_si(m_data, 1);
|
|
return *this;
|
|
}
|
|
|
|
BOOST_ASSERT(!(boost::math::isinf)(a));
|
|
BOOST_ASSERT(!(boost::math::isnan)(a));
|
|
|
|
int e;
|
|
long double f, term;
|
|
mpz_set_ui(m_data, 0u);
|
|
|
|
f = frexp(a, &e);
|
|
|
|
static const int shift = std::numeric_limits<int>::digits - 1;
|
|
|
|
while(f)
|
|
{
|
|
// extract int sized bits from f:
|
|
f = ldexp(f, shift);
|
|
term = floor(f);
|
|
e -= shift;
|
|
mpz_mul_2exp(m_data, m_data, shift);
|
|
if(term > 0)
|
|
mpz_add_ui(m_data, m_data, static_cast<unsigned>(term));
|
|
else
|
|
mpz_sub_ui(m_data, m_data, static_cast<unsigned>(-term));
|
|
f -= term;
|
|
}
|
|
if(e > 0)
|
|
mpz_mul_2exp(m_data, m_data, e);
|
|
else if(e < 0)
|
|
mpz_div_2exp(m_data, m_data, -e);
|
|
return *this;
|
|
}
|
|
gmp_int& operator = (const char* s)
|
|
{
|
|
if(m_data[0]._mp_d == 0)
|
|
mpz_init(this->m_data);
|
|
std::size_t n = s ? std::strlen(s) : 0;
|
|
int radix = 10;
|
|
if(n && (*s == '0'))
|
|
{
|
|
if((n > 1) && ((s[1] == 'x') || (s[1] == 'X')))
|
|
{
|
|
radix = 16;
|
|
s +=2;
|
|
n -= 2;
|
|
}
|
|
else
|
|
{
|
|
radix = 8;
|
|
n -= 1;
|
|
}
|
|
}
|
|
if(n)
|
|
{
|
|
if(0 != mpz_set_str(m_data, s, radix))
|
|
BOOST_THROW_EXCEPTION(std::runtime_error(std::string("The string \"") + s + std::string("\"could not be interpreted as a valid integer.")));
|
|
}
|
|
else
|
|
mpz_set_ui(m_data, 0);
|
|
return *this;
|
|
}
|
|
gmp_int& operator=(const mpf_t val)
|
|
{
|
|
if(m_data[0]._mp_d == 0)
|
|
mpz_init(this->m_data);
|
|
mpz_set_f(this->m_data, val);
|
|
return *this;
|
|
}
|
|
gmp_int& operator=(const mpz_t val)
|
|
{
|
|
if(m_data[0]._mp_d == 0)
|
|
mpz_init(this->m_data);
|
|
mpz_set(this->m_data, val);
|
|
return *this;
|
|
}
|
|
gmp_int& operator=(const mpq_t val)
|
|
{
|
|
if(m_data[0]._mp_d == 0)
|
|
mpz_init(this->m_data);
|
|
mpz_set_q(this->m_data, val);
|
|
return *this;
|
|
}
|
|
template <unsigned Digits10>
|
|
gmp_int& operator=(const gmp_float<Digits10>& o)
|
|
{
|
|
if(m_data[0]._mp_d == 0)
|
|
mpz_init(this->m_data);
|
|
mpz_set_f(this->m_data, o.data());
|
|
return *this;
|
|
}
|
|
gmp_int& operator=(const gmp_rational& o);
|
|
void swap(gmp_int& o)
|
|
{
|
|
mpz_swap(m_data, o.m_data);
|
|
}
|
|
std::string str(std::streamsize /*digits*/, std::ios_base::fmtflags f)const
|
|
{
|
|
BOOST_ASSERT(m_data[0]._mp_d);
|
|
|
|
int base = 10;
|
|
if((f & std::ios_base::oct) == std::ios_base::oct)
|
|
base = 8;
|
|
else if((f & std::ios_base::hex) == std::ios_base::hex)
|
|
base = 16;
|
|
//
|
|
// sanity check, bases 8 and 16 are only available for positive numbers:
|
|
//
|
|
if((base != 10) && (mpz_sgn(m_data) < 0))
|
|
BOOST_THROW_EXCEPTION(std::runtime_error("Formatted output in bases 8 or 16 is only available for positive numbers"));
|
|
void *(*alloc_func_ptr) (size_t);
|
|
void *(*realloc_func_ptr) (void *, size_t, size_t);
|
|
void (*free_func_ptr) (void *, size_t);
|
|
const char* ps = mpz_get_str (0, base, m_data);
|
|
std::string s = ps;
|
|
mp_get_memory_functions(&alloc_func_ptr, &realloc_func_ptr, &free_func_ptr);
|
|
(*free_func_ptr)((void*)ps, std::strlen(ps) + 1);
|
|
|
|
if((base != 10) && (f & std::ios_base::showbase))
|
|
{
|
|
int pos = s[0] == '-' ? 1 : 0;
|
|
const char* pp = base == 8 ? "0" : "0x";
|
|
s.insert(static_cast<std::string::size_type>(pos), pp);
|
|
}
|
|
if((f & std::ios_base::showpos) && (s[0] != '-'))
|
|
s.insert(static_cast<std::string::size_type>(0), 1, '+');
|
|
|
|
return s;
|
|
}
|
|
~gmp_int() BOOST_NOEXCEPT
|
|
{
|
|
if(m_data[0]._mp_d)
|
|
mpz_clear(m_data);
|
|
}
|
|
void negate() BOOST_NOEXCEPT
|
|
{
|
|
BOOST_ASSERT(m_data[0]._mp_d);
|
|
mpz_neg(m_data, m_data);
|
|
}
|
|
int compare(const gmp_int& o)const BOOST_NOEXCEPT
|
|
{
|
|
BOOST_ASSERT(m_data[0]._mp_d && o.m_data[0]._mp_d);
|
|
return mpz_cmp(m_data, o.m_data);
|
|
}
|
|
int compare(long i)const BOOST_NOEXCEPT
|
|
{
|
|
BOOST_ASSERT(m_data[0]._mp_d);
|
|
return mpz_cmp_si(m_data, i);
|
|
}
|
|
int compare(unsigned long i)const BOOST_NOEXCEPT
|
|
{
|
|
BOOST_ASSERT(m_data[0]._mp_d);
|
|
return mpz_cmp_ui(m_data, i);
|
|
}
|
|
template <class V>
|
|
int compare(V v)const
|
|
{
|
|
gmp_int d;
|
|
d = v;
|
|
return compare(d);
|
|
}
|
|
mpz_t& data() BOOST_NOEXCEPT
|
|
{
|
|
BOOST_ASSERT(m_data[0]._mp_d);
|
|
return m_data;
|
|
}
|
|
const mpz_t& data()const BOOST_NOEXCEPT
|
|
{
|
|
BOOST_ASSERT(m_data[0]._mp_d);
|
|
return m_data;
|
|
}
|
|
protected:
|
|
mpz_t m_data;
|
|
};
|
|
|
|
template <class T>
|
|
inline typename enable_if<is_arithmetic<T>, bool>::type eval_eq(const gmp_int& a, const T& b)
|
|
{
|
|
return a.compare(b) == 0;
|
|
}
|
|
template <class T>
|
|
inline typename enable_if<is_arithmetic<T>, bool>::type eval_lt(const gmp_int& a, const T& b)
|
|
{
|
|
return a.compare(b) < 0;
|
|
}
|
|
template <class T>
|
|
inline typename enable_if<is_arithmetic<T>, bool>::type eval_gt(const gmp_int& a, const T& b)
|
|
{
|
|
return a.compare(b) > 0;
|
|
}
|
|
|
|
inline bool eval_is_zero(const gmp_int& val)
|
|
{
|
|
return mpz_sgn(val.data()) == 0;
|
|
}
|
|
inline void eval_add(gmp_int& t, const gmp_int& o)
|
|
{
|
|
mpz_add(t.data(), t.data(), o.data());
|
|
}
|
|
inline void eval_multiply_add(gmp_int& t, const gmp_int& a, const gmp_int& b)
|
|
{
|
|
mpz_addmul(t.data(), a.data(), b.data());
|
|
}
|
|
inline void eval_multiply_subtract(gmp_int& t, const gmp_int& a, const gmp_int& b)
|
|
{
|
|
mpz_submul(t.data(), a.data(), b.data());
|
|
}
|
|
inline void eval_subtract(gmp_int& t, const gmp_int& o)
|
|
{
|
|
mpz_sub(t.data(), t.data(), o.data());
|
|
}
|
|
inline void eval_multiply(gmp_int& t, const gmp_int& o)
|
|
{
|
|
mpz_mul(t.data(), t.data(), o.data());
|
|
}
|
|
inline void eval_divide(gmp_int& t, const gmp_int& o)
|
|
{
|
|
if(eval_is_zero(o))
|
|
BOOST_THROW_EXCEPTION(std::overflow_error("Division by zero."));
|
|
mpz_tdiv_q(t.data(), t.data(), o.data());
|
|
}
|
|
inline void eval_modulus(gmp_int& t, const gmp_int& o)
|
|
{
|
|
mpz_tdiv_r(t.data(), t.data(), o.data());
|
|
}
|
|
inline void eval_add(gmp_int& t, unsigned long i)
|
|
{
|
|
mpz_add_ui(t.data(), t.data(), i);
|
|
}
|
|
inline void eval_multiply_add(gmp_int& t, const gmp_int& a, unsigned long i)
|
|
{
|
|
mpz_addmul_ui(t.data(), a.data(), i);
|
|
}
|
|
inline void eval_multiply_subtract(gmp_int& t, const gmp_int& a, unsigned long i)
|
|
{
|
|
mpz_submul_ui(t.data(), a.data(), i);
|
|
}
|
|
inline void eval_subtract(gmp_int& t, unsigned long i)
|
|
{
|
|
mpz_sub_ui(t.data(), t.data(), i);
|
|
}
|
|
inline void eval_multiply(gmp_int& t, unsigned long i)
|
|
{
|
|
mpz_mul_ui(t.data(), t.data(), i);
|
|
}
|
|
inline void eval_modulus(gmp_int& t, unsigned long i)
|
|
{
|
|
mpz_tdiv_r_ui(t.data(), t.data(), i);
|
|
}
|
|
inline void eval_divide(gmp_int& t, unsigned long i)
|
|
{
|
|
if(i == 0)
|
|
BOOST_THROW_EXCEPTION(std::overflow_error("Division by zero."));
|
|
mpz_tdiv_q_ui(t.data(), t.data(), i);
|
|
}
|
|
inline void eval_add(gmp_int& t, long i)
|
|
{
|
|
if(i > 0)
|
|
mpz_add_ui(t.data(), t.data(), i);
|
|
else
|
|
mpz_sub_ui(t.data(), t.data(), boost::multiprecision::detail::unsigned_abs(i));
|
|
}
|
|
inline void eval_multiply_add(gmp_int& t, const gmp_int& a, long i)
|
|
{
|
|
if(i > 0)
|
|
mpz_addmul_ui(t.data(), a.data(), i);
|
|
else
|
|
mpz_submul_ui(t.data(), a.data(), boost::multiprecision::detail::unsigned_abs(i));
|
|
}
|
|
inline void eval_multiply_subtract(gmp_int& t, const gmp_int& a, long i)
|
|
{
|
|
if(i > 0)
|
|
mpz_submul_ui(t.data(), a.data(), i);
|
|
else
|
|
mpz_addmul_ui(t.data(), a.data(), boost::multiprecision::detail::unsigned_abs(i));
|
|
}
|
|
inline void eval_subtract(gmp_int& t, long i)
|
|
{
|
|
if(i > 0)
|
|
mpz_sub_ui(t.data(), t.data(), i);
|
|
else
|
|
mpz_add_ui(t.data(), t.data(), boost::multiprecision::detail::unsigned_abs(i));
|
|
}
|
|
inline void eval_multiply(gmp_int& t, long i)
|
|
{
|
|
mpz_mul_ui(t.data(), t.data(), boost::multiprecision::detail::unsigned_abs(i));
|
|
if(i < 0)
|
|
mpz_neg(t.data(), t.data());
|
|
}
|
|
inline void eval_modulus(gmp_int& t, long i)
|
|
{
|
|
mpz_tdiv_r_ui(t.data(), t.data(), boost::multiprecision::detail::unsigned_abs(i));
|
|
}
|
|
inline void eval_divide(gmp_int& t, long i)
|
|
{
|
|
if(i == 0)
|
|
BOOST_THROW_EXCEPTION(std::overflow_error("Division by zero."));
|
|
mpz_tdiv_q_ui(t.data(), t.data(), boost::multiprecision::detail::unsigned_abs(i));
|
|
if(i < 0)
|
|
mpz_neg(t.data(), t.data());
|
|
}
|
|
template <class UI>
|
|
inline void eval_left_shift(gmp_int& t, UI i)
|
|
{
|
|
mpz_mul_2exp(t.data(), t.data(), static_cast<unsigned long>(i));
|
|
}
|
|
template <class UI>
|
|
inline void eval_right_shift(gmp_int& t, UI i)
|
|
{
|
|
mpz_fdiv_q_2exp(t.data(), t.data(), static_cast<unsigned long>(i));
|
|
}
|
|
template <class UI>
|
|
inline void eval_left_shift(gmp_int& t, const gmp_int& v, UI i)
|
|
{
|
|
mpz_mul_2exp(t.data(), v.data(), static_cast<unsigned long>(i));
|
|
}
|
|
template <class UI>
|
|
inline void eval_right_shift(gmp_int& t, const gmp_int& v, UI i)
|
|
{
|
|
mpz_fdiv_q_2exp(t.data(), v.data(), static_cast<unsigned long>(i));
|
|
}
|
|
|
|
inline void eval_bitwise_and(gmp_int& result, const gmp_int& v)
|
|
{
|
|
mpz_and(result.data(), result.data(), v.data());
|
|
}
|
|
|
|
inline void eval_bitwise_or(gmp_int& result, const gmp_int& v)
|
|
{
|
|
mpz_ior(result.data(), result.data(), v.data());
|
|
}
|
|
|
|
inline void eval_bitwise_xor(gmp_int& result, const gmp_int& v)
|
|
{
|
|
mpz_xor(result.data(), result.data(), v.data());
|
|
}
|
|
|
|
inline void eval_add(gmp_int& t, const gmp_int& p, const gmp_int& o)
|
|
{
|
|
mpz_add(t.data(), p.data(), o.data());
|
|
}
|
|
inline void eval_subtract(gmp_int& t, const gmp_int& p, const gmp_int& o)
|
|
{
|
|
mpz_sub(t.data(), p.data(), o.data());
|
|
}
|
|
inline void eval_multiply(gmp_int& t, const gmp_int& p, const gmp_int& o)
|
|
{
|
|
mpz_mul(t.data(), p.data(), o.data());
|
|
}
|
|
inline void eval_divide(gmp_int& t, const gmp_int& p, const gmp_int& o)
|
|
{
|
|
if(eval_is_zero(o))
|
|
BOOST_THROW_EXCEPTION(std::overflow_error("Division by zero."));
|
|
mpz_tdiv_q(t.data(), p.data(), o.data());
|
|
}
|
|
inline void eval_modulus(gmp_int& t, const gmp_int& p, const gmp_int& o)
|
|
{
|
|
mpz_tdiv_r(t.data(), p.data(), o.data());
|
|
}
|
|
inline void eval_add(gmp_int& t, const gmp_int& p, unsigned long i)
|
|
{
|
|
mpz_add_ui(t.data(), p.data(), i);
|
|
}
|
|
inline void eval_subtract(gmp_int& t, const gmp_int& p, unsigned long i)
|
|
{
|
|
mpz_sub_ui(t.data(), p.data(), i);
|
|
}
|
|
inline void eval_multiply(gmp_int& t, const gmp_int& p, unsigned long i)
|
|
{
|
|
mpz_mul_ui(t.data(), p.data(), i);
|
|
}
|
|
inline void eval_modulus(gmp_int& t, const gmp_int& p, unsigned long i)
|
|
{
|
|
mpz_tdiv_r_ui(t.data(), p.data(), i);
|
|
}
|
|
inline void eval_divide(gmp_int& t, const gmp_int& p, unsigned long i)
|
|
{
|
|
if(i == 0)
|
|
BOOST_THROW_EXCEPTION(std::overflow_error("Division by zero."));
|
|
mpz_tdiv_q_ui(t.data(), p.data(), i);
|
|
}
|
|
inline void eval_add(gmp_int& t, const gmp_int& p, long i)
|
|
{
|
|
if(i > 0)
|
|
mpz_add_ui(t.data(), p.data(), i);
|
|
else
|
|
mpz_sub_ui(t.data(), p.data(), boost::multiprecision::detail::unsigned_abs(i));
|
|
}
|
|
inline void eval_subtract(gmp_int& t, const gmp_int& p, long i)
|
|
{
|
|
if(i > 0)
|
|
mpz_sub_ui(t.data(), p.data(), i);
|
|
else
|
|
mpz_add_ui(t.data(), p.data(), boost::multiprecision::detail::unsigned_abs(i));
|
|
}
|
|
inline void eval_multiply(gmp_int& t, const gmp_int& p, long i)
|
|
{
|
|
mpz_mul_ui(t.data(), p.data(), boost::multiprecision::detail::unsigned_abs(i));
|
|
if(i < 0)
|
|
mpz_neg(t.data(), t.data());
|
|
}
|
|
inline void eval_modulus(gmp_int& t, const gmp_int& p, long i)
|
|
{
|
|
mpz_tdiv_r_ui(t.data(), p.data(), boost::multiprecision::detail::unsigned_abs(i));
|
|
}
|
|
inline void eval_divide(gmp_int& t, const gmp_int& p, long i)
|
|
{
|
|
if(i == 0)
|
|
BOOST_THROW_EXCEPTION(std::overflow_error("Division by zero."));
|
|
mpz_tdiv_q_ui(t.data(), p.data(), boost::multiprecision::detail::unsigned_abs(i));
|
|
if(i < 0)
|
|
mpz_neg(t.data(), t.data());
|
|
}
|
|
|
|
inline void eval_bitwise_and(gmp_int& result, const gmp_int& u, const gmp_int& v)
|
|
{
|
|
mpz_and(result.data(), u.data(), v.data());
|
|
}
|
|
|
|
inline void eval_bitwise_or(gmp_int& result, const gmp_int& u, const gmp_int& v)
|
|
{
|
|
mpz_ior(result.data(), u.data(), v.data());
|
|
}
|
|
|
|
inline void eval_bitwise_xor(gmp_int& result, const gmp_int& u, const gmp_int& v)
|
|
{
|
|
mpz_xor(result.data(), u.data(), v.data());
|
|
}
|
|
|
|
inline void eval_complement(gmp_int& result, const gmp_int& u)
|
|
{
|
|
mpz_com(result.data(), u.data());
|
|
}
|
|
|
|
inline int eval_get_sign(const gmp_int& val)
|
|
{
|
|
return mpz_sgn(val.data());
|
|
}
|
|
inline void eval_convert_to(unsigned long* result, const gmp_int& val)
|
|
{
|
|
if(0 == mpz_fits_ulong_p(val.data()))
|
|
{
|
|
*result = (std::numeric_limits<unsigned long>::max)();
|
|
}
|
|
else
|
|
*result = mpz_get_ui(val.data());
|
|
}
|
|
inline void eval_convert_to(long* result, const gmp_int& val)
|
|
{
|
|
if(0 == mpz_fits_slong_p(val.data()))
|
|
{
|
|
*result = (std::numeric_limits<unsigned long>::max)();
|
|
*result *= mpz_sgn(val.data());
|
|
}
|
|
else
|
|
*result = mpz_get_si(val.data());
|
|
}
|
|
inline void eval_convert_to(double* result, const gmp_int& val)
|
|
{
|
|
*result = mpz_get_d(val.data());
|
|
}
|
|
|
|
inline void eval_abs(gmp_int& result, const gmp_int& val)
|
|
{
|
|
mpz_abs(result.data(), val.data());
|
|
}
|
|
|
|
inline void eval_gcd(gmp_int& result, const gmp_int& a, const gmp_int& b)
|
|
{
|
|
mpz_gcd(result.data(), a.data(), b.data());
|
|
}
|
|
inline void eval_lcm(gmp_int& result, const gmp_int& a, const gmp_int& b)
|
|
{
|
|
mpz_lcm(result.data(), a.data(), b.data());
|
|
}
|
|
template <class I>
|
|
inline typename enable_if_c<(is_unsigned<I>::value && (sizeof(I) <= sizeof(unsigned long)))>::type eval_gcd(gmp_int& result, const gmp_int& a, const I b)
|
|
{
|
|
mpz_gcd_ui(result.data(), a.data(), b);
|
|
}
|
|
template <class I>
|
|
inline typename enable_if_c<(is_unsigned<I>::value && (sizeof(I) <= sizeof(unsigned long)))>::type eval_lcm(gmp_int& result, const gmp_int& a, const I b)
|
|
{
|
|
mpz_lcm_ui(result.data(), a.data(), b);
|
|
}
|
|
template <class I>
|
|
inline typename enable_if_c<(is_signed<I>::value && (sizeof(I) <= sizeof(long)))>::type eval_gcd(gmp_int& result, const gmp_int& a, const I b)
|
|
{
|
|
mpz_gcd_ui(result.data(), a.data(), boost::multiprecision::detail::unsigned_abs(b));
|
|
}
|
|
template <class I>
|
|
inline typename enable_if_c<is_signed<I>::value && ((sizeof(I) <= sizeof(long)))>::type eval_lcm(gmp_int& result, const gmp_int& a, const I b)
|
|
{
|
|
mpz_lcm_ui(result.data(), a.data(), boost::multiprecision::detail::unsigned_abs(b));
|
|
}
|
|
|
|
inline void eval_integer_sqrt(gmp_int& s, gmp_int& r, const gmp_int& x)
|
|
{
|
|
mpz_sqrtrem(s.data(), r.data(), x.data());
|
|
}
|
|
|
|
inline unsigned eval_lsb(const gmp_int& val)
|
|
{
|
|
int c = eval_get_sign(val);
|
|
if(c == 0)
|
|
{
|
|
BOOST_THROW_EXCEPTION(std::range_error("No bits were set in the operand."));
|
|
}
|
|
if(c < 0)
|
|
{
|
|
BOOST_THROW_EXCEPTION(std::range_error("Testing individual bits in negative values is not supported - results are undefined."));
|
|
}
|
|
return static_cast<unsigned>(mpz_scan1(val.data(), 0));
|
|
}
|
|
|
|
inline unsigned eval_msb(const gmp_int& val)
|
|
{
|
|
int c = eval_get_sign(val);
|
|
if(c == 0)
|
|
{
|
|
BOOST_THROW_EXCEPTION(std::range_error("No bits were set in the operand."));
|
|
}
|
|
if(c < 0)
|
|
{
|
|
BOOST_THROW_EXCEPTION(std::range_error("Testing individual bits in negative values is not supported - results are undefined."));
|
|
}
|
|
return static_cast<unsigned>(mpz_sizeinbase(val.data(), 2) - 1);
|
|
}
|
|
|
|
inline bool eval_bit_test(const gmp_int& val, unsigned index)
|
|
{
|
|
return mpz_tstbit(val.data(), index) ? true : false;
|
|
}
|
|
|
|
inline void eval_bit_set(gmp_int& val, unsigned index)
|
|
{
|
|
mpz_setbit(val.data(), index);
|
|
}
|
|
|
|
inline void eval_bit_unset(gmp_int& val, unsigned index)
|
|
{
|
|
mpz_clrbit(val.data(), index);
|
|
}
|
|
|
|
inline void eval_bit_flip(gmp_int& val, unsigned index)
|
|
{
|
|
mpz_combit(val.data(), index);
|
|
}
|
|
|
|
inline void eval_qr(const gmp_int& x, const gmp_int& y,
|
|
gmp_int& q, gmp_int& r)
|
|
{
|
|
mpz_tdiv_qr(q.data(), r.data(), x.data(), y.data());
|
|
}
|
|
|
|
template <class Integer>
|
|
inline typename enable_if<is_unsigned<Integer>, Integer>::type eval_integer_modulus(const gmp_int& x, Integer val)
|
|
{
|
|
if((sizeof(Integer) <= sizeof(long)) || (val <= (std::numeric_limits<unsigned long>::max)()))
|
|
{
|
|
return mpz_tdiv_ui(x.data(), val);
|
|
}
|
|
else
|
|
{
|
|
return default_ops::eval_integer_modulus(x, val);
|
|
}
|
|
}
|
|
template <class Integer>
|
|
inline typename enable_if<is_signed<Integer>, Integer>::type eval_integer_modulus(const gmp_int& x, Integer val)
|
|
{
|
|
return eval_integer_modulus(x, boost::multiprecision::detail::unsigned_abs(val));
|
|
}
|
|
inline void eval_powm(gmp_int& result, const gmp_int& base, const gmp_int& p, const gmp_int& m)
|
|
{
|
|
if(eval_get_sign(p) < 0)
|
|
{
|
|
BOOST_THROW_EXCEPTION(std::runtime_error("powm requires a positive exponent."));
|
|
}
|
|
mpz_powm(result.data(), base.data(), p.data(), m.data());
|
|
}
|
|
|
|
template <class Integer>
|
|
inline typename enable_if<
|
|
mpl::and_<
|
|
is_unsigned<Integer>,
|
|
mpl::bool_<sizeof(Integer) <= sizeof(unsigned long)>
|
|
>
|
|
>::type eval_powm(gmp_int& result, const gmp_int& base, Integer p, const gmp_int& m)
|
|
{
|
|
mpz_powm_ui(result.data(), base.data(), p, m.data());
|
|
}
|
|
template <class Integer>
|
|
inline typename enable_if<
|
|
mpl::and_<
|
|
is_signed<Integer>,
|
|
mpl::bool_<sizeof(Integer) <= sizeof(unsigned long)>
|
|
>
|
|
>::type eval_powm(gmp_int& result, const gmp_int& base, Integer p, const gmp_int& m)
|
|
{
|
|
if(p < 0)
|
|
{
|
|
BOOST_THROW_EXCEPTION(std::runtime_error("powm requires a positive exponent."));
|
|
}
|
|
mpz_powm_ui(result.data(), base.data(), p, m.data());
|
|
}
|
|
|
|
inline std::size_t hash_value(const gmp_int& val)
|
|
{
|
|
// We should really use mpz_limbs_read here, but that's unsupported on older versions:
|
|
std::size_t result = 0;
|
|
for(int i = 0; i < std::abs(val.data()[0]._mp_size); ++i)
|
|
boost::hash_combine(result, val.data()[0]._mp_d[i]);
|
|
boost::hash_combine(result, val.data()[0]._mp_size);
|
|
return result;
|
|
}
|
|
|
|
struct gmp_rational;
|
|
void eval_add(gmp_rational& t, const gmp_rational& o);
|
|
|
|
struct gmp_rational
|
|
{
|
|
#ifdef BOOST_HAS_LONG_LONG
|
|
typedef mpl::list<long, boost::long_long_type> signed_types;
|
|
typedef mpl::list<unsigned long, boost::ulong_long_type> unsigned_types;
|
|
#else
|
|
typedef mpl::list<long> signed_types;
|
|
typedef mpl::list<unsigned long> unsigned_types;
|
|
#endif
|
|
typedef mpl::list<double, long double> float_types;
|
|
|
|
gmp_rational()
|
|
{
|
|
mpq_init(this->m_data);
|
|
}
|
|
gmp_rational(const gmp_rational& o)
|
|
{
|
|
mpq_init(m_data);
|
|
if(o.m_data[0]._mp_num._mp_d)
|
|
mpq_set(m_data, o.m_data);
|
|
}
|
|
gmp_rational(const gmp_int& o)
|
|
{
|
|
mpq_init(m_data);
|
|
mpq_set_z(m_data, o.data());
|
|
}
|
|
#ifndef BOOST_NO_CXX11_RVALUE_REFERENCES
|
|
gmp_rational(gmp_rational&& o) BOOST_NOEXCEPT
|
|
{
|
|
m_data[0] = o.m_data[0];
|
|
o.m_data[0]._mp_num._mp_d = 0;
|
|
o.m_data[0]._mp_den._mp_d = 0;
|
|
}
|
|
#endif
|
|
gmp_rational(const mpq_t o)
|
|
{
|
|
mpq_init(m_data);
|
|
mpq_set(m_data, o);
|
|
}
|
|
gmp_rational(const mpz_t o)
|
|
{
|
|
mpq_init(m_data);
|
|
mpq_set_z(m_data, o);
|
|
}
|
|
gmp_rational& operator = (const gmp_rational& o)
|
|
{
|
|
if(m_data[0]._mp_den._mp_d == 0)
|
|
mpq_init(m_data);
|
|
mpq_set(m_data, o.m_data);
|
|
return *this;
|
|
}
|
|
#ifndef BOOST_NO_CXX11_RVALUE_REFERENCES
|
|
gmp_rational& operator = (gmp_rational&& o) BOOST_NOEXCEPT
|
|
{
|
|
mpq_swap(m_data, o.m_data);
|
|
return *this;
|
|
}
|
|
#endif
|
|
#ifdef BOOST_HAS_LONG_LONG
|
|
#if defined(ULLONG_MAX) && (ULLONG_MAX == ULONG_MAX)
|
|
gmp_rational& operator = (boost::ulong_long_type i)
|
|
{
|
|
*this = static_cast<unsigned long>(i);
|
|
return *this;
|
|
}
|
|
#else
|
|
gmp_rational& operator = (boost::ulong_long_type i)
|
|
{
|
|
if(m_data[0]._mp_den._mp_d == 0)
|
|
mpq_init(m_data);
|
|
gmp_int zi;
|
|
zi = i;
|
|
mpq_set_z(m_data, zi.data());
|
|
return *this;
|
|
}
|
|
gmp_rational& operator = (boost::long_long_type i)
|
|
{
|
|
if(m_data[0]._mp_den._mp_d == 0)
|
|
mpq_init(m_data);
|
|
bool neg = i < 0;
|
|
*this = boost::multiprecision::detail::unsigned_abs(i);
|
|
if(neg)
|
|
mpq_neg(m_data, m_data);
|
|
return *this;
|
|
}
|
|
#endif
|
|
#endif
|
|
gmp_rational& operator = (unsigned long i)
|
|
{
|
|
if(m_data[0]._mp_den._mp_d == 0)
|
|
mpq_init(m_data);
|
|
mpq_set_ui(m_data, i, 1);
|
|
return *this;
|
|
}
|
|
gmp_rational& operator = (long i)
|
|
{
|
|
if(m_data[0]._mp_den._mp_d == 0)
|
|
mpq_init(m_data);
|
|
mpq_set_si(m_data, i, 1);
|
|
return *this;
|
|
}
|
|
gmp_rational& operator = (double d)
|
|
{
|
|
if(m_data[0]._mp_den._mp_d == 0)
|
|
mpq_init(m_data);
|
|
mpq_set_d(m_data, d);
|
|
return *this;
|
|
}
|
|
gmp_rational& operator = (long double a)
|
|
{
|
|
using std::frexp;
|
|
using std::ldexp;
|
|
using std::floor;
|
|
using default_ops::eval_add;
|
|
using default_ops::eval_subtract;
|
|
|
|
if(m_data[0]._mp_den._mp_d == 0)
|
|
mpq_init(m_data);
|
|
|
|
if (a == 0) {
|
|
mpq_set_si(m_data, 0, 1);
|
|
return *this;
|
|
}
|
|
|
|
if (a == 1) {
|
|
mpq_set_si(m_data, 1, 1);
|
|
return *this;
|
|
}
|
|
|
|
BOOST_ASSERT(!(boost::math::isinf)(a));
|
|
BOOST_ASSERT(!(boost::math::isnan)(a));
|
|
|
|
int e;
|
|
long double f, term;
|
|
mpq_set_ui(m_data, 0, 1);
|
|
mpq_set_ui(m_data, 0u, 1);
|
|
gmp_rational t;
|
|
|
|
f = frexp(a, &e);
|
|
|
|
static const int shift = std::numeric_limits<int>::digits - 1;
|
|
|
|
while(f)
|
|
{
|
|
// extract int sized bits from f:
|
|
f = ldexp(f, shift);
|
|
term = floor(f);
|
|
e -= shift;
|
|
mpq_mul_2exp(m_data, m_data, shift);
|
|
t = static_cast<long>(term);
|
|
eval_add(*this, t);
|
|
f -= term;
|
|
}
|
|
if(e > 0)
|
|
mpq_mul_2exp(m_data, m_data, e);
|
|
else if(e < 0)
|
|
mpq_div_2exp(m_data, m_data, -e);
|
|
return *this;
|
|
}
|
|
gmp_rational& operator = (const char* s)
|
|
{
|
|
if(m_data[0]._mp_den._mp_d == 0)
|
|
mpq_init(m_data);
|
|
if(0 != mpq_set_str(m_data, s, 10))
|
|
BOOST_THROW_EXCEPTION(std::runtime_error(std::string("The string \"") + s + std::string("\"could not be interpreted as a valid rational number.")));
|
|
return *this;
|
|
}
|
|
gmp_rational& operator=(const gmp_int& o)
|
|
{
|
|
if(m_data[0]._mp_den._mp_d == 0)
|
|
mpq_init(m_data);
|
|
mpq_set_z(m_data, o.data());
|
|
return *this;
|
|
}
|
|
gmp_rational& operator=(const mpq_t o)
|
|
{
|
|
if(m_data[0]._mp_den._mp_d == 0)
|
|
mpq_init(m_data);
|
|
mpq_set(m_data, o);
|
|
return *this;
|
|
}
|
|
gmp_rational& operator=(const mpz_t o)
|
|
{
|
|
if(m_data[0]._mp_den._mp_d == 0)
|
|
mpq_init(m_data);
|
|
mpq_set_z(m_data, o);
|
|
return *this;
|
|
}
|
|
void swap(gmp_rational& o)
|
|
{
|
|
mpq_swap(m_data, o.m_data);
|
|
}
|
|
std::string str(std::streamsize /*digits*/, std::ios_base::fmtflags /*f*/)const
|
|
{
|
|
BOOST_ASSERT(m_data[0]._mp_num._mp_d);
|
|
// TODO make a better job of this including handling of f!!
|
|
void *(*alloc_func_ptr) (size_t);
|
|
void *(*realloc_func_ptr) (void *, size_t, size_t);
|
|
void (*free_func_ptr) (void *, size_t);
|
|
const char* ps = mpq_get_str (0, 10, m_data);
|
|
std::string s = ps;
|
|
mp_get_memory_functions(&alloc_func_ptr, &realloc_func_ptr, &free_func_ptr);
|
|
(*free_func_ptr)((void*)ps, std::strlen(ps) + 1);
|
|
return s;
|
|
}
|
|
~gmp_rational()
|
|
{
|
|
if(m_data[0]._mp_num._mp_d || m_data[0]._mp_den._mp_d)
|
|
mpq_clear(m_data);
|
|
}
|
|
void negate()
|
|
{
|
|
BOOST_ASSERT(m_data[0]._mp_num._mp_d);
|
|
mpq_neg(m_data, m_data);
|
|
}
|
|
int compare(const gmp_rational& o)const
|
|
{
|
|
BOOST_ASSERT(m_data[0]._mp_num._mp_d && o.m_data[0]._mp_num._mp_d);
|
|
return mpq_cmp(m_data, o.m_data);
|
|
}
|
|
template <class V>
|
|
int compare(V v)const
|
|
{
|
|
gmp_rational d;
|
|
d = v;
|
|
return compare(d);
|
|
}
|
|
int compare(unsigned long v)const
|
|
{
|
|
BOOST_ASSERT(m_data[0]._mp_num._mp_d);
|
|
return mpq_cmp_ui(m_data, v, 1);
|
|
}
|
|
int compare(long v)const
|
|
{
|
|
BOOST_ASSERT(m_data[0]._mp_num._mp_d);
|
|
return mpq_cmp_si(m_data, v, 1);
|
|
}
|
|
mpq_t& data()
|
|
{
|
|
BOOST_ASSERT(m_data[0]._mp_num._mp_d);
|
|
return m_data;
|
|
}
|
|
const mpq_t& data()const
|
|
{
|
|
BOOST_ASSERT(m_data[0]._mp_num._mp_d);
|
|
return m_data;
|
|
}
|
|
protected:
|
|
mpq_t m_data;
|
|
};
|
|
|
|
inline bool eval_is_zero(const gmp_rational& val)
|
|
{
|
|
return mpq_sgn(val.data()) == 0;
|
|
}
|
|
template <class T>
|
|
inline bool eval_eq(gmp_rational& a, const T& b)
|
|
{
|
|
return a.compare(b) == 0;
|
|
}
|
|
template <class T>
|
|
inline bool eval_lt(gmp_rational& a, const T& b)
|
|
{
|
|
return a.compare(b) < 0;
|
|
}
|
|
template <class T>
|
|
inline bool eval_gt(gmp_rational& a, const T& b)
|
|
{
|
|
return a.compare(b) > 0;
|
|
}
|
|
|
|
inline void eval_add(gmp_rational& t, const gmp_rational& o)
|
|
{
|
|
mpq_add(t.data(), t.data(), o.data());
|
|
}
|
|
inline void eval_subtract(gmp_rational& t, const gmp_rational& o)
|
|
{
|
|
mpq_sub(t.data(), t.data(), o.data());
|
|
}
|
|
inline void eval_multiply(gmp_rational& t, const gmp_rational& o)
|
|
{
|
|
mpq_mul(t.data(), t.data(), o.data());
|
|
}
|
|
inline void eval_divide(gmp_rational& t, const gmp_rational& o)
|
|
{
|
|
if(eval_is_zero(o))
|
|
BOOST_THROW_EXCEPTION(std::overflow_error("Division by zero."));
|
|
mpq_div(t.data(), t.data(), o.data());
|
|
}
|
|
inline void eval_add(gmp_rational& t, const gmp_rational& p, const gmp_rational& o)
|
|
{
|
|
mpq_add(t.data(), p.data(), o.data());
|
|
}
|
|
inline void eval_subtract(gmp_rational& t, const gmp_rational& p, const gmp_rational& o)
|
|
{
|
|
mpq_sub(t.data(), p.data(), o.data());
|
|
}
|
|
inline void eval_multiply(gmp_rational& t, const gmp_rational& p, const gmp_rational& o)
|
|
{
|
|
mpq_mul(t.data(), p.data(), o.data());
|
|
}
|
|
inline void eval_divide(gmp_rational& t, const gmp_rational& p, const gmp_rational& o)
|
|
{
|
|
if(eval_is_zero(o))
|
|
BOOST_THROW_EXCEPTION(std::overflow_error("Division by zero."));
|
|
mpq_div(t.data(), p.data(), o.data());
|
|
}
|
|
|
|
inline int eval_get_sign(const gmp_rational& val)
|
|
{
|
|
return mpq_sgn(val.data());
|
|
}
|
|
inline void eval_convert_to(double* result, const gmp_rational& val)
|
|
{
|
|
//
|
|
// This does not round correctly:
|
|
//
|
|
//*result = mpq_get_d(val.data());
|
|
//
|
|
// This does:
|
|
//
|
|
boost::multiprecision::detail::generic_convert_rational_to_float(*result, val);
|
|
}
|
|
|
|
inline void eval_convert_to(long* result, const gmp_rational& val)
|
|
{
|
|
double r;
|
|
eval_convert_to(&r, val);
|
|
*result = static_cast<long>(r);
|
|
}
|
|
|
|
inline void eval_convert_to(unsigned long* result, const gmp_rational& val)
|
|
{
|
|
double r;
|
|
eval_convert_to(&r, val);
|
|
*result = static_cast<long>(r);
|
|
}
|
|
|
|
inline void eval_abs(gmp_rational& result, const gmp_rational& val)
|
|
{
|
|
mpq_abs(result.data(), val.data());
|
|
}
|
|
|
|
inline void assign_components(gmp_rational& result, unsigned long v1, unsigned long v2)
|
|
{
|
|
mpq_set_ui(result.data(), v1, v2);
|
|
mpq_canonicalize(result.data());
|
|
}
|
|
inline void assign_components(gmp_rational& result, long v1, long v2)
|
|
{
|
|
mpq_set_si(result.data(), v1, v2);
|
|
mpq_canonicalize(result.data());
|
|
}
|
|
inline void assign_components(gmp_rational& result, gmp_int const& v1, gmp_int const& v2)
|
|
{
|
|
mpz_set(mpq_numref(result.data()), v1.data());
|
|
mpz_set(mpq_denref(result.data()), v2.data());
|
|
mpq_canonicalize(result.data());
|
|
}
|
|
|
|
inline std::size_t hash_value(const gmp_rational& val)
|
|
{
|
|
std::size_t result = 0;
|
|
for(int i = 0; i < std::abs(val.data()[0]._mp_num._mp_size); ++i)
|
|
boost::hash_combine(result, val.data()[0]._mp_num._mp_d[i]);
|
|
for(int i = 0; i < std::abs(val.data()[0]._mp_den._mp_size); ++i)
|
|
boost::hash_combine(result, val.data()[0]._mp_den._mp_d[i]);
|
|
boost::hash_combine(result, val.data()[0]._mp_num._mp_size);
|
|
return result;
|
|
}
|
|
|
|
//
|
|
// Some member functions that are dependent upon previous code go here:
|
|
//
|
|
template <unsigned Digits10>
|
|
template <unsigned D>
|
|
inline gmp_float<Digits10>::gmp_float(const gmp_float<D>& o, typename enable_if_c<D <= Digits10>::type*)
|
|
{
|
|
mpf_init2(this->m_data, multiprecision::detail::digits10_2_2(Digits10 ? Digits10 : this->get_default_precision()));
|
|
mpf_set(this->m_data, o.data());
|
|
}
|
|
template <unsigned Digits10>
|
|
template <unsigned D>
|
|
inline gmp_float<Digits10>::gmp_float(const gmp_float<D>& o, typename disable_if_c<D <= Digits10>::type*)
|
|
{
|
|
mpf_init2(this->m_data, multiprecision::detail::digits10_2_2(Digits10 ? Digits10 : this->get_default_precision()));
|
|
mpf_set(this->m_data, o.data());
|
|
}
|
|
template <unsigned Digits10>
|
|
inline gmp_float<Digits10>::gmp_float(const gmp_int& o)
|
|
{
|
|
mpf_init2(this->m_data, multiprecision::detail::digits10_2_2(Digits10 ? Digits10 : this->get_default_precision()));
|
|
mpf_set_z(this->data(), o.data());
|
|
}
|
|
template <unsigned Digits10>
|
|
inline gmp_float<Digits10>::gmp_float(const gmp_rational& o)
|
|
{
|
|
mpf_init2(this->m_data, multiprecision::detail::digits10_2_2(Digits10 ? Digits10 : this->get_default_precision()));
|
|
mpf_set_q(this->data(), o.data());
|
|
}
|
|
template <unsigned Digits10>
|
|
template <unsigned D>
|
|
inline gmp_float<Digits10>& gmp_float<Digits10>::operator=(const gmp_float<D>& o)
|
|
{
|
|
if(this->m_data[0]._mp_d == 0)
|
|
mpf_init2(this->m_data, multiprecision::detail::digits10_2_2(Digits10 ? Digits10 : this->get_default_precision()));
|
|
mpf_set(this->m_data, o.data());
|
|
return *this;
|
|
}
|
|
template <unsigned Digits10>
|
|
inline gmp_float<Digits10>& gmp_float<Digits10>::operator=(const gmp_int& o)
|
|
{
|
|
if(this->m_data[0]._mp_d == 0)
|
|
mpf_init2(this->m_data, multiprecision::detail::digits10_2_2(Digits10 ? Digits10 : this->get_default_precision()));
|
|
mpf_set_z(this->data(), o.data());
|
|
return *this;
|
|
}
|
|
template <unsigned Digits10>
|
|
inline gmp_float<Digits10>& gmp_float<Digits10>::operator=(const gmp_rational& o)
|
|
{
|
|
if(this->m_data[0]._mp_d == 0)
|
|
mpf_init2(this->m_data, multiprecision::detail::digits10_2_2(Digits10 ? Digits10 : this->get_default_precision()));
|
|
mpf_set_q(this->data(), o.data());
|
|
return *this;
|
|
}
|
|
inline gmp_float<0>::gmp_float(const gmp_int& o)
|
|
{
|
|
mpf_init2(this->m_data, multiprecision::detail::digits10_2_2(get_default_precision()));
|
|
mpf_set_z(this->data(), o.data());
|
|
}
|
|
inline gmp_float<0>::gmp_float(const gmp_rational& o)
|
|
{
|
|
mpf_init2(this->m_data, multiprecision::detail::digits10_2_2(get_default_precision()));
|
|
mpf_set_q(this->data(), o.data());
|
|
}
|
|
inline gmp_float<0>& gmp_float<0>::operator=(const gmp_int& o)
|
|
{
|
|
if(this->m_data[0]._mp_d == 0)
|
|
mpf_init2(this->m_data, multiprecision::detail::digits10_2_2(this->get_default_precision()));
|
|
mpf_set_z(this->data(), o.data());
|
|
return *this;
|
|
}
|
|
inline gmp_float<0>& gmp_float<0>::operator=(const gmp_rational& o)
|
|
{
|
|
if(this->m_data[0]._mp_d == 0)
|
|
mpf_init2(this->m_data, multiprecision::detail::digits10_2_2(this->get_default_precision()));
|
|
mpf_set_q(this->data(), o.data());
|
|
return *this;
|
|
}
|
|
inline gmp_int::gmp_int(const gmp_rational& o)
|
|
{
|
|
mpz_init(this->m_data);
|
|
mpz_set_q(this->m_data, o.data());
|
|
}
|
|
inline gmp_int& gmp_int::operator=(const gmp_rational& o)
|
|
{
|
|
if(this->m_data[0]._mp_d == 0)
|
|
mpz_init(this->m_data);
|
|
mpz_set_q(this->m_data, o.data());
|
|
return *this;
|
|
}
|
|
|
|
} //namespace backends
|
|
|
|
using boost::multiprecision::backends::gmp_int;
|
|
using boost::multiprecision::backends::gmp_rational;
|
|
using boost::multiprecision::backends::gmp_float;
|
|
|
|
template <>
|
|
struct component_type<number<gmp_rational> >
|
|
{
|
|
typedef number<gmp_int> type;
|
|
};
|
|
|
|
template <expression_template_option ET>
|
|
inline number<gmp_int, ET> numerator(const number<gmp_rational, ET>& val)
|
|
{
|
|
number<gmp_int, ET> result;
|
|
mpz_set(result.backend().data(), (mpq_numref(val.backend().data())));
|
|
return result;
|
|
}
|
|
template <expression_template_option ET>
|
|
inline number<gmp_int, ET> denominator(const number<gmp_rational, ET>& val)
|
|
{
|
|
number<gmp_int, ET> result;
|
|
mpz_set(result.backend().data(), (mpq_denref(val.backend().data())));
|
|
return result;
|
|
}
|
|
|
|
namespace detail{
|
|
|
|
#ifdef BOOST_NO_SFINAE_EXPR
|
|
|
|
template<>
|
|
struct is_explicitly_convertible<canonical<mpf_t, gmp_int>::type, gmp_int> : public mpl::true_ {};
|
|
template<>
|
|
struct is_explicitly_convertible<canonical<mpq_t, gmp_int>::type, gmp_int> : public mpl::true_ {};
|
|
template<unsigned Digits10>
|
|
struct is_explicitly_convertible<gmp_float<Digits10>, gmp_int> : public mpl::true_ {};
|
|
template<>
|
|
struct is_explicitly_convertible<gmp_rational, gmp_int> : public mpl::true_ {};
|
|
template<unsigned D1, unsigned D2>
|
|
struct is_explicitly_convertible<gmp_float<D1>, gmp_float<D2> > : public mpl::true_ {};
|
|
|
|
#endif
|
|
|
|
template <>
|
|
struct digits2<number<gmp_float<0>, et_on> >
|
|
{
|
|
static long value()
|
|
{
|
|
return multiprecision::detail::digits10_2_2(gmp_float<0>::default_precision());
|
|
}
|
|
};
|
|
|
|
template <>
|
|
struct digits2<number<gmp_float<0>, et_off> >
|
|
{
|
|
static long value()
|
|
{
|
|
return multiprecision::detail::digits10_2_2(gmp_float<0>::default_precision());
|
|
}
|
|
};
|
|
|
|
template <>
|
|
struct digits2<number<debug_adaptor<gmp_float<0> >, et_on> >
|
|
{
|
|
static long value()
|
|
{
|
|
return multiprecision::detail::digits10_2_2(gmp_float<0>::default_precision());
|
|
}
|
|
};
|
|
|
|
template <>
|
|
struct digits2<number<debug_adaptor<gmp_float<0> >, et_off> >
|
|
{
|
|
static long value()
|
|
{
|
|
return multiprecision::detail::digits10_2_2(gmp_float<0>::default_precision());
|
|
}
|
|
};
|
|
|
|
}
|
|
|
|
template<>
|
|
struct number_category<detail::canonical<mpz_t, gmp_int>::type> : public mpl::int_<number_kind_integer>{};
|
|
template<>
|
|
struct number_category<detail::canonical<mpq_t, gmp_rational>::type> : public mpl::int_<number_kind_rational>{};
|
|
template<>
|
|
struct number_category<detail::canonical<mpf_t, gmp_float<0> >::type> : public mpl::int_<number_kind_floating_point>{};
|
|
|
|
|
|
typedef number<gmp_float<50> > mpf_float_50;
|
|
typedef number<gmp_float<100> > mpf_float_100;
|
|
typedef number<gmp_float<500> > mpf_float_500;
|
|
typedef number<gmp_float<1000> > mpf_float_1000;
|
|
typedef number<gmp_float<0> > mpf_float;
|
|
typedef number<gmp_int > mpz_int;
|
|
typedef number<gmp_rational > mpq_rational;
|
|
|
|
} // namespace multiprecision
|
|
|
|
namespace math { namespace tools{
|
|
|
|
template <>
|
|
inline int digits<boost::multiprecision::mpf_float>()
|
|
#ifdef BOOST_MATH_NOEXCEPT
|
|
BOOST_NOEXCEPT
|
|
#endif
|
|
{
|
|
return multiprecision::detail::digits10_2_2(boost::multiprecision::mpf_float::default_precision());
|
|
}
|
|
template <>
|
|
inline int digits<boost::multiprecision::number<boost::multiprecision::gmp_float<0>, boost::multiprecision::et_off> >()
|
|
#ifdef BOOST_MATH_NOEXCEPT
|
|
BOOST_NOEXCEPT
|
|
#endif
|
|
{
|
|
return multiprecision::detail::digits10_2_2(boost::multiprecision::mpf_float::default_precision());
|
|
}
|
|
|
|
template <>
|
|
inline boost::multiprecision::mpf_float
|
|
max_value<boost::multiprecision::mpf_float>()
|
|
{
|
|
boost::multiprecision::mpf_float result(0.5);
|
|
mpf_mul_2exp(result.backend().data(), result.backend().data(), (std::numeric_limits<mp_exp_t>::max)() / 64 + 1);
|
|
return result;
|
|
}
|
|
|
|
template <>
|
|
inline boost::multiprecision::mpf_float
|
|
min_value<boost::multiprecision::mpf_float>()
|
|
{
|
|
boost::multiprecision::mpf_float result(0.5);
|
|
mpf_div_2exp(result.backend().data(), result.backend().data(), (std::numeric_limits<mp_exp_t>::min)() / 64 + 1);
|
|
return result;
|
|
}
|
|
|
|
template <>
|
|
inline boost::multiprecision::number<boost::multiprecision::gmp_float<0>, boost::multiprecision::et_off>
|
|
max_value<boost::multiprecision::number<boost::multiprecision::gmp_float<0>, boost::multiprecision::et_off> >()
|
|
{
|
|
boost::multiprecision::number<boost::multiprecision::gmp_float<0>, boost::multiprecision::et_off> result(0.5);
|
|
mpf_mul_2exp(result.backend().data(), result.backend().data(), (std::numeric_limits<mp_exp_t>::max)() / 64 + 1);
|
|
return result;
|
|
}
|
|
|
|
template <>
|
|
inline boost::multiprecision::number<boost::multiprecision::gmp_float<0>, boost::multiprecision::et_off>
|
|
min_value<boost::multiprecision::number<boost::multiprecision::gmp_float<0>, boost::multiprecision::et_off> >()
|
|
{
|
|
boost::multiprecision::number<boost::multiprecision::gmp_float<0>, boost::multiprecision::et_off> result(0.5);
|
|
mpf_div_2exp(result.backend().data(), result.backend().data(), (std::numeric_limits<mp_exp_t>::max)() / 64 + 1);
|
|
return result;
|
|
}
|
|
|
|
template <>
|
|
inline int digits<boost::multiprecision::number<boost::multiprecision::debug_adaptor<boost::multiprecision::mpf_float::backend_type> > >()
|
|
#ifdef BOOST_MATH_NOEXCEPT
|
|
BOOST_NOEXCEPT
|
|
#endif
|
|
{
|
|
return multiprecision::detail::digits10_2_2(boost::multiprecision::number<boost::multiprecision::debug_adaptor<boost::multiprecision::mpf_float::backend_type> >::default_precision());
|
|
}
|
|
template <>
|
|
inline int digits<boost::multiprecision::number<boost::multiprecision::debug_adaptor<boost::multiprecision::gmp_float<0> >, boost::multiprecision::et_off> >()
|
|
#ifdef BOOST_MATH_NOEXCEPT
|
|
BOOST_NOEXCEPT
|
|
#endif
|
|
{
|
|
return multiprecision::detail::digits10_2_2(boost::multiprecision::number<boost::multiprecision::debug_adaptor<boost::multiprecision::mpf_float::backend_type> >::default_precision());
|
|
}
|
|
|
|
template <>
|
|
inline boost::multiprecision::number<boost::multiprecision::debug_adaptor<boost::multiprecision::mpf_float::backend_type> >
|
|
max_value<boost::multiprecision::number<boost::multiprecision::debug_adaptor<boost::multiprecision::mpf_float::backend_type> > >()
|
|
{
|
|
return max_value<boost::multiprecision::mpf_float>().backend();
|
|
}
|
|
|
|
template <>
|
|
inline boost::multiprecision::number<boost::multiprecision::debug_adaptor<boost::multiprecision::mpf_float::backend_type> >
|
|
min_value<boost::multiprecision::number<boost::multiprecision::debug_adaptor<boost::multiprecision::mpf_float::backend_type> > >()
|
|
{
|
|
return min_value<boost::multiprecision::mpf_float>().backend();
|
|
}
|
|
|
|
template <>
|
|
inline boost::multiprecision::number<boost::multiprecision::debug_adaptor<boost::multiprecision::gmp_float<0> >, boost::multiprecision::et_off>
|
|
max_value<boost::multiprecision::number<boost::multiprecision::debug_adaptor<boost::multiprecision::gmp_float<0> >, boost::multiprecision::et_off> >()
|
|
{
|
|
return max_value<boost::multiprecision::mpf_float>().backend();
|
|
}
|
|
|
|
template <>
|
|
inline boost::multiprecision::number<boost::multiprecision::debug_adaptor<boost::multiprecision::gmp_float<0> >, boost::multiprecision::et_off>
|
|
min_value<boost::multiprecision::number<boost::multiprecision::debug_adaptor<boost::multiprecision::gmp_float<0> >, boost::multiprecision::et_off> >()
|
|
{
|
|
return min_value<boost::multiprecision::mpf_float>().backend();
|
|
}
|
|
|
|
|
|
}} // namespaces math::tools
|
|
|
|
|
|
} // namespace boost
|
|
|
|
namespace std{
|
|
|
|
//
|
|
// numeric_limits [partial] specializations for the types declared in this header:
|
|
//
|
|
template<unsigned Digits10, boost::multiprecision::expression_template_option ExpressionTemplates>
|
|
class numeric_limits<boost::multiprecision::number<boost::multiprecision::gmp_float<Digits10>, ExpressionTemplates> >
|
|
{
|
|
typedef boost::multiprecision::number<boost::multiprecision::gmp_float<Digits10>, ExpressionTemplates> number_type;
|
|
public:
|
|
BOOST_STATIC_CONSTEXPR bool is_specialized = true;
|
|
//
|
|
// min and max values chosen so as to not cause segfaults when calling
|
|
// mpf_get_str on 64-bit Linux builds. Possibly we could use larger
|
|
// exponent values elsewhere.
|
|
//
|
|
static number_type (min)()
|
|
{
|
|
initializer.do_nothing();
|
|
static std::pair<bool, number_type> value;
|
|
if(!value.first)
|
|
{
|
|
value.first = true;
|
|
value.second = 1;
|
|
mpf_div_2exp(value.second.backend().data(), value.second.backend().data(), (std::numeric_limits<mp_exp_t>::max)() / 64 + 1);
|
|
}
|
|
return value.second;
|
|
}
|
|
static number_type (max)()
|
|
{
|
|
initializer.do_nothing();
|
|
static std::pair<bool, number_type> value;
|
|
if(!value.first)
|
|
{
|
|
value.first = true;
|
|
value.second = 1;
|
|
mpf_mul_2exp(value.second.backend().data(), value.second.backend().data(), (std::numeric_limits<mp_exp_t>::max)() / 64 + 1);
|
|
}
|
|
return value.second;
|
|
}
|
|
BOOST_STATIC_CONSTEXPR number_type lowest()
|
|
{
|
|
return -(max)();
|
|
}
|
|
BOOST_STATIC_CONSTEXPR int digits = static_cast<int>((Digits10 * 1000L) / 301L + ((Digits10 * 1000L) % 301L ? 2 : 1));
|
|
BOOST_STATIC_CONSTEXPR int digits10 = Digits10;
|
|
// Have to allow for a possible extra limb inside the gmp data structure:
|
|
BOOST_STATIC_CONSTEXPR int max_digits10 = Digits10 + 3 + ((GMP_LIMB_BITS * 301L) / 1000L);
|
|
BOOST_STATIC_CONSTEXPR bool is_signed = true;
|
|
BOOST_STATIC_CONSTEXPR bool is_integer = false;
|
|
BOOST_STATIC_CONSTEXPR bool is_exact = false;
|
|
BOOST_STATIC_CONSTEXPR int radix = 2;
|
|
static number_type epsilon()
|
|
{
|
|
initializer.do_nothing();
|
|
static std::pair<bool, number_type> value;
|
|
if(!value.first)
|
|
{
|
|
value.first = true;
|
|
value.second = 1;
|
|
mpf_div_2exp(value.second.backend().data(), value.second.backend().data(), std::numeric_limits<number_type>::digits - 1);
|
|
}
|
|
return value.second;
|
|
}
|
|
// What value should this be????
|
|
static number_type round_error()
|
|
{
|
|
// returns epsilon/2
|
|
initializer.do_nothing();
|
|
static std::pair<bool, number_type> value;
|
|
if(!value.first)
|
|
{
|
|
value.first = true;
|
|
value.second = 1;
|
|
}
|
|
return value.second;
|
|
}
|
|
BOOST_STATIC_CONSTEXPR long min_exponent = LONG_MIN;
|
|
BOOST_STATIC_CONSTEXPR long min_exponent10 = (LONG_MIN / 1000) * 301L;
|
|
BOOST_STATIC_CONSTEXPR long max_exponent = LONG_MAX;
|
|
BOOST_STATIC_CONSTEXPR long max_exponent10 = (LONG_MAX / 1000) * 301L;
|
|
BOOST_STATIC_CONSTEXPR bool has_infinity = false;
|
|
BOOST_STATIC_CONSTEXPR bool has_quiet_NaN = false;
|
|
BOOST_STATIC_CONSTEXPR bool has_signaling_NaN = false;
|
|
BOOST_STATIC_CONSTEXPR float_denorm_style has_denorm = denorm_absent;
|
|
BOOST_STATIC_CONSTEXPR bool has_denorm_loss = false;
|
|
BOOST_STATIC_CONSTEXPR number_type infinity() { return number_type(); }
|
|
BOOST_STATIC_CONSTEXPR number_type quiet_NaN() { return number_type(); }
|
|
BOOST_STATIC_CONSTEXPR number_type signaling_NaN() { return number_type(); }
|
|
BOOST_STATIC_CONSTEXPR number_type denorm_min() { return number_type(); }
|
|
BOOST_STATIC_CONSTEXPR bool is_iec559 = false;
|
|
BOOST_STATIC_CONSTEXPR bool is_bounded = true;
|
|
BOOST_STATIC_CONSTEXPR bool is_modulo = false;
|
|
BOOST_STATIC_CONSTEXPR bool traps = true;
|
|
BOOST_STATIC_CONSTEXPR bool tinyness_before = false;
|
|
BOOST_STATIC_CONSTEXPR float_round_style round_style = round_indeterminate;
|
|
|
|
private:
|
|
struct data_initializer
|
|
{
|
|
data_initializer()
|
|
{
|
|
std::numeric_limits<boost::multiprecision::number<boost::multiprecision::gmp_float<digits10> > >::epsilon();
|
|
std::numeric_limits<boost::multiprecision::number<boost::multiprecision::gmp_float<digits10> > >::round_error();
|
|
(std::numeric_limits<boost::multiprecision::number<boost::multiprecision::gmp_float<digits10> > >::min)();
|
|
(std::numeric_limits<boost::multiprecision::number<boost::multiprecision::gmp_float<digits10> > >::max)();
|
|
}
|
|
void do_nothing()const{}
|
|
};
|
|
static const data_initializer initializer;
|
|
};
|
|
|
|
template<unsigned Digits10, boost::multiprecision::expression_template_option ExpressionTemplates>
|
|
const typename numeric_limits<boost::multiprecision::number<boost::multiprecision::gmp_float<Digits10>, ExpressionTemplates> >::data_initializer numeric_limits<boost::multiprecision::number<boost::multiprecision::gmp_float<Digits10>, ExpressionTemplates> >::initializer;
|
|
|
|
#ifndef BOOST_NO_INCLASS_MEMBER_INITIALIZATION
|
|
|
|
template <unsigned Digits10, boost::multiprecision::expression_template_option ExpressionTemplates>
|
|
BOOST_CONSTEXPR_OR_CONST int numeric_limits<boost::multiprecision::number<boost::multiprecision::gmp_float<Digits10>, ExpressionTemplates> >::digits;
|
|
template <unsigned Digits10, boost::multiprecision::expression_template_option ExpressionTemplates>
|
|
BOOST_CONSTEXPR_OR_CONST int numeric_limits<boost::multiprecision::number<boost::multiprecision::gmp_float<Digits10>, ExpressionTemplates> >::digits10;
|
|
template <unsigned Digits10, boost::multiprecision::expression_template_option ExpressionTemplates>
|
|
BOOST_CONSTEXPR_OR_CONST int numeric_limits<boost::multiprecision::number<boost::multiprecision::gmp_float<Digits10>, ExpressionTemplates> >::max_digits10;
|
|
template <unsigned Digits10, boost::multiprecision::expression_template_option ExpressionTemplates>
|
|
BOOST_CONSTEXPR_OR_CONST bool numeric_limits<boost::multiprecision::number<boost::multiprecision::gmp_float<Digits10>, ExpressionTemplates> >::is_signed;
|
|
template <unsigned Digits10, boost::multiprecision::expression_template_option ExpressionTemplates>
|
|
BOOST_CONSTEXPR_OR_CONST bool numeric_limits<boost::multiprecision::number<boost::multiprecision::gmp_float<Digits10>, ExpressionTemplates> >::is_integer;
|
|
template <unsigned Digits10, boost::multiprecision::expression_template_option ExpressionTemplates>
|
|
BOOST_CONSTEXPR_OR_CONST bool numeric_limits<boost::multiprecision::number<boost::multiprecision::gmp_float<Digits10>, ExpressionTemplates> >::is_exact;
|
|
template <unsigned Digits10, boost::multiprecision::expression_template_option ExpressionTemplates>
|
|
BOOST_CONSTEXPR_OR_CONST int numeric_limits<boost::multiprecision::number<boost::multiprecision::gmp_float<Digits10>, ExpressionTemplates> >::radix;
|
|
template <unsigned Digits10, boost::multiprecision::expression_template_option ExpressionTemplates>
|
|
BOOST_CONSTEXPR_OR_CONST long numeric_limits<boost::multiprecision::number<boost::multiprecision::gmp_float<Digits10>, ExpressionTemplates> >::min_exponent;
|
|
template <unsigned Digits10, boost::multiprecision::expression_template_option ExpressionTemplates>
|
|
BOOST_CONSTEXPR_OR_CONST long numeric_limits<boost::multiprecision::number<boost::multiprecision::gmp_float<Digits10>, ExpressionTemplates> >::min_exponent10;
|
|
template <unsigned Digits10, boost::multiprecision::expression_template_option ExpressionTemplates>
|
|
BOOST_CONSTEXPR_OR_CONST long numeric_limits<boost::multiprecision::number<boost::multiprecision::gmp_float<Digits10>, ExpressionTemplates> >::max_exponent;
|
|
template <unsigned Digits10, boost::multiprecision::expression_template_option ExpressionTemplates>
|
|
BOOST_CONSTEXPR_OR_CONST long numeric_limits<boost::multiprecision::number<boost::multiprecision::gmp_float<Digits10>, ExpressionTemplates> >::max_exponent10;
|
|
template <unsigned Digits10, boost::multiprecision::expression_template_option ExpressionTemplates>
|
|
BOOST_CONSTEXPR_OR_CONST bool numeric_limits<boost::multiprecision::number<boost::multiprecision::gmp_float<Digits10>, ExpressionTemplates> >::has_infinity;
|
|
template <unsigned Digits10, boost::multiprecision::expression_template_option ExpressionTemplates>
|
|
BOOST_CONSTEXPR_OR_CONST bool numeric_limits<boost::multiprecision::number<boost::multiprecision::gmp_float<Digits10>, ExpressionTemplates> >::has_quiet_NaN;
|
|
template <unsigned Digits10, boost::multiprecision::expression_template_option ExpressionTemplates>
|
|
BOOST_CONSTEXPR_OR_CONST bool numeric_limits<boost::multiprecision::number<boost::multiprecision::gmp_float<Digits10>, ExpressionTemplates> >::has_signaling_NaN;
|
|
template <unsigned Digits10, boost::multiprecision::expression_template_option ExpressionTemplates>
|
|
BOOST_CONSTEXPR_OR_CONST float_denorm_style numeric_limits<boost::multiprecision::number<boost::multiprecision::gmp_float<Digits10>, ExpressionTemplates> >::has_denorm;
|
|
template <unsigned Digits10, boost::multiprecision::expression_template_option ExpressionTemplates>
|
|
BOOST_CONSTEXPR_OR_CONST bool numeric_limits<boost::multiprecision::number<boost::multiprecision::gmp_float<Digits10>, ExpressionTemplates> >::has_denorm_loss;
|
|
template <unsigned Digits10, boost::multiprecision::expression_template_option ExpressionTemplates>
|
|
BOOST_CONSTEXPR_OR_CONST bool numeric_limits<boost::multiprecision::number<boost::multiprecision::gmp_float<Digits10>, ExpressionTemplates> >::is_iec559;
|
|
template <unsigned Digits10, boost::multiprecision::expression_template_option ExpressionTemplates>
|
|
BOOST_CONSTEXPR_OR_CONST bool numeric_limits<boost::multiprecision::number<boost::multiprecision::gmp_float<Digits10>, ExpressionTemplates> >::is_bounded;
|
|
template <unsigned Digits10, boost::multiprecision::expression_template_option ExpressionTemplates>
|
|
BOOST_CONSTEXPR_OR_CONST bool numeric_limits<boost::multiprecision::number<boost::multiprecision::gmp_float<Digits10>, ExpressionTemplates> >::is_modulo;
|
|
template <unsigned Digits10, boost::multiprecision::expression_template_option ExpressionTemplates>
|
|
BOOST_CONSTEXPR_OR_CONST bool numeric_limits<boost::multiprecision::number<boost::multiprecision::gmp_float<Digits10>, ExpressionTemplates> >::traps;
|
|
template <unsigned Digits10, boost::multiprecision::expression_template_option ExpressionTemplates>
|
|
BOOST_CONSTEXPR_OR_CONST bool numeric_limits<boost::multiprecision::number<boost::multiprecision::gmp_float<Digits10>, ExpressionTemplates> >::tinyness_before;
|
|
template <unsigned Digits10, boost::multiprecision::expression_template_option ExpressionTemplates>
|
|
BOOST_CONSTEXPR_OR_CONST float_round_style numeric_limits<boost::multiprecision::number<boost::multiprecision::gmp_float<Digits10>, ExpressionTemplates> >::round_style;
|
|
|
|
#endif
|
|
|
|
template<boost::multiprecision::expression_template_option ExpressionTemplates>
|
|
class numeric_limits<boost::multiprecision::number<boost::multiprecision::gmp_float<0>, ExpressionTemplates> >
|
|
{
|
|
typedef boost::multiprecision::number<boost::multiprecision::gmp_float<0>, ExpressionTemplates> number_type;
|
|
public:
|
|
BOOST_STATIC_CONSTEXPR bool is_specialized = false;
|
|
static number_type (min)() { return number_type(); }
|
|
static number_type (max)() { return number_type(); }
|
|
static number_type lowest() { return number_type(); }
|
|
BOOST_STATIC_CONSTEXPR int digits = 0;
|
|
BOOST_STATIC_CONSTEXPR int digits10 = 0;
|
|
BOOST_STATIC_CONSTEXPR int max_digits10 = 0;
|
|
BOOST_STATIC_CONSTEXPR bool is_signed = false;
|
|
BOOST_STATIC_CONSTEXPR bool is_integer = false;
|
|
BOOST_STATIC_CONSTEXPR bool is_exact = false;
|
|
BOOST_STATIC_CONSTEXPR int radix = 0;
|
|
static number_type epsilon() { return number_type(); }
|
|
static number_type round_error() { return number_type(); }
|
|
BOOST_STATIC_CONSTEXPR int min_exponent = 0;
|
|
BOOST_STATIC_CONSTEXPR int min_exponent10 = 0;
|
|
BOOST_STATIC_CONSTEXPR int max_exponent = 0;
|
|
BOOST_STATIC_CONSTEXPR int max_exponent10 = 0;
|
|
BOOST_STATIC_CONSTEXPR bool has_infinity = false;
|
|
BOOST_STATIC_CONSTEXPR bool has_quiet_NaN = false;
|
|
BOOST_STATIC_CONSTEXPR bool has_signaling_NaN = false;
|
|
BOOST_STATIC_CONSTEXPR float_denorm_style has_denorm = denorm_absent;
|
|
BOOST_STATIC_CONSTEXPR bool has_denorm_loss = false;
|
|
static number_type infinity() { return number_type(); }
|
|
static number_type quiet_NaN() { return number_type(); }
|
|
static number_type signaling_NaN() { return number_type(); }
|
|
static number_type denorm_min() { return number_type(); }
|
|
BOOST_STATIC_CONSTEXPR bool is_iec559 = false;
|
|
BOOST_STATIC_CONSTEXPR bool is_bounded = false;
|
|
BOOST_STATIC_CONSTEXPR bool is_modulo = false;
|
|
BOOST_STATIC_CONSTEXPR bool traps = false;
|
|
BOOST_STATIC_CONSTEXPR bool tinyness_before = false;
|
|
BOOST_STATIC_CONSTEXPR float_round_style round_style = round_indeterminate;
|
|
};
|
|
|
|
#ifndef BOOST_NO_INCLASS_MEMBER_INITIALIZATION
|
|
|
|
template <boost::multiprecision::expression_template_option ExpressionTemplates>
|
|
BOOST_CONSTEXPR_OR_CONST int numeric_limits<boost::multiprecision::number<boost::multiprecision::gmp_float<0>, ExpressionTemplates> >::digits;
|
|
template <boost::multiprecision::expression_template_option ExpressionTemplates>
|
|
BOOST_CONSTEXPR_OR_CONST int numeric_limits<boost::multiprecision::number<boost::multiprecision::gmp_float<0>, ExpressionTemplates> >::digits10;
|
|
template <boost::multiprecision::expression_template_option ExpressionTemplates>
|
|
BOOST_CONSTEXPR_OR_CONST int numeric_limits<boost::multiprecision::number<boost::multiprecision::gmp_float<0>, ExpressionTemplates> >::max_digits10;
|
|
template <boost::multiprecision::expression_template_option ExpressionTemplates>
|
|
BOOST_CONSTEXPR_OR_CONST bool numeric_limits<boost::multiprecision::number<boost::multiprecision::gmp_float<0>, ExpressionTemplates> >::is_signed;
|
|
template <boost::multiprecision::expression_template_option ExpressionTemplates>
|
|
BOOST_CONSTEXPR_OR_CONST bool numeric_limits<boost::multiprecision::number<boost::multiprecision::gmp_float<0>, ExpressionTemplates> >::is_integer;
|
|
template <boost::multiprecision::expression_template_option ExpressionTemplates>
|
|
BOOST_CONSTEXPR_OR_CONST bool numeric_limits<boost::multiprecision::number<boost::multiprecision::gmp_float<0>, ExpressionTemplates> >::is_exact;
|
|
template <boost::multiprecision::expression_template_option ExpressionTemplates>
|
|
BOOST_CONSTEXPR_OR_CONST int numeric_limits<boost::multiprecision::number<boost::multiprecision::gmp_float<0>, ExpressionTemplates> >::radix;
|
|
template <boost::multiprecision::expression_template_option ExpressionTemplates>
|
|
BOOST_CONSTEXPR_OR_CONST int numeric_limits<boost::multiprecision::number<boost::multiprecision::gmp_float<0>, ExpressionTemplates> >::min_exponent;
|
|
template <boost::multiprecision::expression_template_option ExpressionTemplates>
|
|
BOOST_CONSTEXPR_OR_CONST int numeric_limits<boost::multiprecision::number<boost::multiprecision::gmp_float<0>, ExpressionTemplates> >::min_exponent10;
|
|
template <boost::multiprecision::expression_template_option ExpressionTemplates>
|
|
BOOST_CONSTEXPR_OR_CONST int numeric_limits<boost::multiprecision::number<boost::multiprecision::gmp_float<0>, ExpressionTemplates> >::max_exponent;
|
|
template <boost::multiprecision::expression_template_option ExpressionTemplates>
|
|
BOOST_CONSTEXPR_OR_CONST int numeric_limits<boost::multiprecision::number<boost::multiprecision::gmp_float<0>, ExpressionTemplates> >::max_exponent10;
|
|
template <boost::multiprecision::expression_template_option ExpressionTemplates>
|
|
BOOST_CONSTEXPR_OR_CONST bool numeric_limits<boost::multiprecision::number<boost::multiprecision::gmp_float<0>, ExpressionTemplates> >::has_infinity;
|
|
template <boost::multiprecision::expression_template_option ExpressionTemplates>
|
|
BOOST_CONSTEXPR_OR_CONST bool numeric_limits<boost::multiprecision::number<boost::multiprecision::gmp_float<0>, ExpressionTemplates> >::has_quiet_NaN;
|
|
template <boost::multiprecision::expression_template_option ExpressionTemplates>
|
|
BOOST_CONSTEXPR_OR_CONST bool numeric_limits<boost::multiprecision::number<boost::multiprecision::gmp_float<0>, ExpressionTemplates> >::has_signaling_NaN;
|
|
template <boost::multiprecision::expression_template_option ExpressionTemplates>
|
|
BOOST_CONSTEXPR_OR_CONST float_denorm_style numeric_limits<boost::multiprecision::number<boost::multiprecision::gmp_float<0>, ExpressionTemplates> >::has_denorm;
|
|
template <boost::multiprecision::expression_template_option ExpressionTemplates>
|
|
BOOST_CONSTEXPR_OR_CONST bool numeric_limits<boost::multiprecision::number<boost::multiprecision::gmp_float<0>, ExpressionTemplates> >::has_denorm_loss;
|
|
template <boost::multiprecision::expression_template_option ExpressionTemplates>
|
|
BOOST_CONSTEXPR_OR_CONST bool numeric_limits<boost::multiprecision::number<boost::multiprecision::gmp_float<0>, ExpressionTemplates> >::is_iec559;
|
|
template <boost::multiprecision::expression_template_option ExpressionTemplates>
|
|
BOOST_CONSTEXPR_OR_CONST bool numeric_limits<boost::multiprecision::number<boost::multiprecision::gmp_float<0>, ExpressionTemplates> >::is_bounded;
|
|
template <boost::multiprecision::expression_template_option ExpressionTemplates>
|
|
BOOST_CONSTEXPR_OR_CONST bool numeric_limits<boost::multiprecision::number<boost::multiprecision::gmp_float<0>, ExpressionTemplates> >::is_modulo;
|
|
template <boost::multiprecision::expression_template_option ExpressionTemplates>
|
|
BOOST_CONSTEXPR_OR_CONST bool numeric_limits<boost::multiprecision::number<boost::multiprecision::gmp_float<0>, ExpressionTemplates> >::traps;
|
|
template <boost::multiprecision::expression_template_option ExpressionTemplates>
|
|
BOOST_CONSTEXPR_OR_CONST bool numeric_limits<boost::multiprecision::number<boost::multiprecision::gmp_float<0>, ExpressionTemplates> >::tinyness_before;
|
|
template <boost::multiprecision::expression_template_option ExpressionTemplates>
|
|
BOOST_CONSTEXPR_OR_CONST float_round_style numeric_limits<boost::multiprecision::number<boost::multiprecision::gmp_float<0>, ExpressionTemplates> >::round_style;
|
|
|
|
#endif
|
|
|
|
template<boost::multiprecision::expression_template_option ExpressionTemplates>
|
|
class numeric_limits<boost::multiprecision::number<boost::multiprecision::gmp_int, ExpressionTemplates> >
|
|
{
|
|
typedef boost::multiprecision::number<boost::multiprecision::gmp_int, ExpressionTemplates> number_type;
|
|
public:
|
|
BOOST_STATIC_CONSTEXPR bool is_specialized = true;
|
|
//
|
|
// Largest and smallest numbers are bounded only by available memory, set
|
|
// to zero:
|
|
//
|
|
static number_type (min)()
|
|
{
|
|
return number_type();
|
|
}
|
|
static number_type (max)()
|
|
{
|
|
return number_type();
|
|
}
|
|
static number_type lowest() { return (min)(); }
|
|
BOOST_STATIC_CONSTEXPR int digits = INT_MAX;
|
|
BOOST_STATIC_CONSTEXPR int digits10 = (INT_MAX / 1000) * 301L;
|
|
BOOST_STATIC_CONSTEXPR int max_digits10 = digits10 + 3;
|
|
BOOST_STATIC_CONSTEXPR bool is_signed = true;
|
|
BOOST_STATIC_CONSTEXPR bool is_integer = true;
|
|
BOOST_STATIC_CONSTEXPR bool is_exact = true;
|
|
BOOST_STATIC_CONSTEXPR int radix = 2;
|
|
static number_type epsilon() { return number_type(); }
|
|
static number_type round_error() { return number_type(); }
|
|
BOOST_STATIC_CONSTEXPR int min_exponent = 0;
|
|
BOOST_STATIC_CONSTEXPR int min_exponent10 = 0;
|
|
BOOST_STATIC_CONSTEXPR int max_exponent = 0;
|
|
BOOST_STATIC_CONSTEXPR int max_exponent10 = 0;
|
|
BOOST_STATIC_CONSTEXPR bool has_infinity = false;
|
|
BOOST_STATIC_CONSTEXPR bool has_quiet_NaN = false;
|
|
BOOST_STATIC_CONSTEXPR bool has_signaling_NaN = false;
|
|
BOOST_STATIC_CONSTEXPR float_denorm_style has_denorm = denorm_absent;
|
|
BOOST_STATIC_CONSTEXPR bool has_denorm_loss = false;
|
|
static number_type infinity() { return number_type(); }
|
|
static number_type quiet_NaN() { return number_type(); }
|
|
static number_type signaling_NaN() { return number_type(); }
|
|
static number_type denorm_min() { return number_type(); }
|
|
BOOST_STATIC_CONSTEXPR bool is_iec559 = false;
|
|
BOOST_STATIC_CONSTEXPR bool is_bounded = false;
|
|
BOOST_STATIC_CONSTEXPR bool is_modulo = false;
|
|
BOOST_STATIC_CONSTEXPR bool traps = false;
|
|
BOOST_STATIC_CONSTEXPR bool tinyness_before = false;
|
|
BOOST_STATIC_CONSTEXPR float_round_style round_style = round_toward_zero;
|
|
};
|
|
|
|
#ifndef BOOST_NO_INCLASS_MEMBER_INITIALIZATION
|
|
|
|
template <boost::multiprecision::expression_template_option ExpressionTemplates>
|
|
BOOST_CONSTEXPR_OR_CONST int numeric_limits<boost::multiprecision::number<boost::multiprecision::gmp_int, ExpressionTemplates> >::digits;
|
|
template <boost::multiprecision::expression_template_option ExpressionTemplates>
|
|
BOOST_CONSTEXPR_OR_CONST int numeric_limits<boost::multiprecision::number<boost::multiprecision::gmp_int, ExpressionTemplates> >::digits10;
|
|
template <boost::multiprecision::expression_template_option ExpressionTemplates>
|
|
BOOST_CONSTEXPR_OR_CONST int numeric_limits<boost::multiprecision::number<boost::multiprecision::gmp_int, ExpressionTemplates> >::max_digits10;
|
|
template <boost::multiprecision::expression_template_option ExpressionTemplates>
|
|
BOOST_CONSTEXPR_OR_CONST bool numeric_limits<boost::multiprecision::number<boost::multiprecision::gmp_int, ExpressionTemplates> >::is_signed;
|
|
template <boost::multiprecision::expression_template_option ExpressionTemplates>
|
|
BOOST_CONSTEXPR_OR_CONST bool numeric_limits<boost::multiprecision::number<boost::multiprecision::gmp_int, ExpressionTemplates> >::is_integer;
|
|
template <boost::multiprecision::expression_template_option ExpressionTemplates>
|
|
BOOST_CONSTEXPR_OR_CONST bool numeric_limits<boost::multiprecision::number<boost::multiprecision::gmp_int, ExpressionTemplates> >::is_exact;
|
|
template <boost::multiprecision::expression_template_option ExpressionTemplates>
|
|
BOOST_CONSTEXPR_OR_CONST int numeric_limits<boost::multiprecision::number<boost::multiprecision::gmp_int, ExpressionTemplates> >::radix;
|
|
template <boost::multiprecision::expression_template_option ExpressionTemplates>
|
|
BOOST_CONSTEXPR_OR_CONST int numeric_limits<boost::multiprecision::number<boost::multiprecision::gmp_int, ExpressionTemplates> >::min_exponent;
|
|
template <boost::multiprecision::expression_template_option ExpressionTemplates>
|
|
BOOST_CONSTEXPR_OR_CONST int numeric_limits<boost::multiprecision::number<boost::multiprecision::gmp_int, ExpressionTemplates> >::min_exponent10;
|
|
template <boost::multiprecision::expression_template_option ExpressionTemplates>
|
|
BOOST_CONSTEXPR_OR_CONST int numeric_limits<boost::multiprecision::number<boost::multiprecision::gmp_int, ExpressionTemplates> >::max_exponent;
|
|
template <boost::multiprecision::expression_template_option ExpressionTemplates>
|
|
BOOST_CONSTEXPR_OR_CONST int numeric_limits<boost::multiprecision::number<boost::multiprecision::gmp_int, ExpressionTemplates> >::max_exponent10;
|
|
template <boost::multiprecision::expression_template_option ExpressionTemplates>
|
|
BOOST_CONSTEXPR_OR_CONST bool numeric_limits<boost::multiprecision::number<boost::multiprecision::gmp_int, ExpressionTemplates> >::has_infinity;
|
|
template <boost::multiprecision::expression_template_option ExpressionTemplates>
|
|
BOOST_CONSTEXPR_OR_CONST bool numeric_limits<boost::multiprecision::number<boost::multiprecision::gmp_int, ExpressionTemplates> >::has_quiet_NaN;
|
|
template <boost::multiprecision::expression_template_option ExpressionTemplates>
|
|
BOOST_CONSTEXPR_OR_CONST bool numeric_limits<boost::multiprecision::number<boost::multiprecision::gmp_int, ExpressionTemplates> >::has_signaling_NaN;
|
|
template <boost::multiprecision::expression_template_option ExpressionTemplates>
|
|
BOOST_CONSTEXPR_OR_CONST float_denorm_style numeric_limits<boost::multiprecision::number<boost::multiprecision::gmp_int, ExpressionTemplates> >::has_denorm;
|
|
template <boost::multiprecision::expression_template_option ExpressionTemplates>
|
|
BOOST_CONSTEXPR_OR_CONST bool numeric_limits<boost::multiprecision::number<boost::multiprecision::gmp_int, ExpressionTemplates> >::has_denorm_loss;
|
|
template <boost::multiprecision::expression_template_option ExpressionTemplates>
|
|
BOOST_CONSTEXPR_OR_CONST bool numeric_limits<boost::multiprecision::number<boost::multiprecision::gmp_int, ExpressionTemplates> >::is_iec559;
|
|
template <boost::multiprecision::expression_template_option ExpressionTemplates>
|
|
BOOST_CONSTEXPR_OR_CONST bool numeric_limits<boost::multiprecision::number<boost::multiprecision::gmp_int, ExpressionTemplates> >::is_bounded;
|
|
template <boost::multiprecision::expression_template_option ExpressionTemplates>
|
|
BOOST_CONSTEXPR_OR_CONST bool numeric_limits<boost::multiprecision::number<boost::multiprecision::gmp_int, ExpressionTemplates> >::is_modulo;
|
|
template <boost::multiprecision::expression_template_option ExpressionTemplates>
|
|
BOOST_CONSTEXPR_OR_CONST bool numeric_limits<boost::multiprecision::number<boost::multiprecision::gmp_int, ExpressionTemplates> >::traps;
|
|
template <boost::multiprecision::expression_template_option ExpressionTemplates>
|
|
BOOST_CONSTEXPR_OR_CONST bool numeric_limits<boost::multiprecision::number<boost::multiprecision::gmp_int, ExpressionTemplates> >::tinyness_before;
|
|
template <boost::multiprecision::expression_template_option ExpressionTemplates>
|
|
BOOST_CONSTEXPR_OR_CONST float_round_style numeric_limits<boost::multiprecision::number<boost::multiprecision::gmp_int, ExpressionTemplates> >::round_style;
|
|
|
|
#endif
|
|
|
|
template<boost::multiprecision::expression_template_option ExpressionTemplates>
|
|
class numeric_limits<boost::multiprecision::number<boost::multiprecision::gmp_rational, ExpressionTemplates> >
|
|
{
|
|
typedef boost::multiprecision::number<boost::multiprecision::gmp_rational, ExpressionTemplates> number_type;
|
|
public:
|
|
BOOST_STATIC_CONSTEXPR bool is_specialized = true;
|
|
//
|
|
// Largest and smallest numbers are bounded only by available memory, set
|
|
// to zero:
|
|
//
|
|
static number_type (min)()
|
|
{
|
|
return number_type();
|
|
}
|
|
static number_type (max)()
|
|
{
|
|
return number_type();
|
|
}
|
|
static number_type lowest() { return (min)(); }
|
|
// Digits are unbounded, use zero for now:
|
|
BOOST_STATIC_CONSTEXPR int digits = INT_MAX;
|
|
BOOST_STATIC_CONSTEXPR int digits10 = (INT_MAX / 1000) * 301L;
|
|
BOOST_STATIC_CONSTEXPR int max_digits10 = digits10 + 3;
|
|
BOOST_STATIC_CONSTEXPR bool is_signed = true;
|
|
BOOST_STATIC_CONSTEXPR bool is_integer = false;
|
|
BOOST_STATIC_CONSTEXPR bool is_exact = true;
|
|
BOOST_STATIC_CONSTEXPR int radix = 2;
|
|
static number_type epsilon() { return number_type(); }
|
|
static number_type round_error() { return number_type(); }
|
|
BOOST_STATIC_CONSTEXPR int min_exponent = 0;
|
|
BOOST_STATIC_CONSTEXPR int min_exponent10 = 0;
|
|
BOOST_STATIC_CONSTEXPR int max_exponent = 0;
|
|
BOOST_STATIC_CONSTEXPR int max_exponent10 = 0;
|
|
BOOST_STATIC_CONSTEXPR bool has_infinity = false;
|
|
BOOST_STATIC_CONSTEXPR bool has_quiet_NaN = false;
|
|
BOOST_STATIC_CONSTEXPR bool has_signaling_NaN = false;
|
|
BOOST_STATIC_CONSTEXPR float_denorm_style has_denorm = denorm_absent;
|
|
BOOST_STATIC_CONSTEXPR bool has_denorm_loss = false;
|
|
static number_type infinity() { return number_type(); }
|
|
static number_type quiet_NaN() { return number_type(); }
|
|
static number_type signaling_NaN() { return number_type(); }
|
|
static number_type denorm_min() { return number_type(); }
|
|
BOOST_STATIC_CONSTEXPR bool is_iec559 = false;
|
|
BOOST_STATIC_CONSTEXPR bool is_bounded = false;
|
|
BOOST_STATIC_CONSTEXPR bool is_modulo = false;
|
|
BOOST_STATIC_CONSTEXPR bool traps = false;
|
|
BOOST_STATIC_CONSTEXPR bool tinyness_before = false;
|
|
BOOST_STATIC_CONSTEXPR float_round_style round_style = round_toward_zero;
|
|
};
|
|
|
|
#ifndef BOOST_NO_INCLASS_MEMBER_INITIALIZATION
|
|
|
|
template <boost::multiprecision::expression_template_option ExpressionTemplates>
|
|
BOOST_CONSTEXPR_OR_CONST int numeric_limits<boost::multiprecision::number<boost::multiprecision::gmp_rational, ExpressionTemplates> >::digits;
|
|
template <boost::multiprecision::expression_template_option ExpressionTemplates>
|
|
BOOST_CONSTEXPR_OR_CONST int numeric_limits<boost::multiprecision::number<boost::multiprecision::gmp_rational, ExpressionTemplates> >::digits10;
|
|
template <boost::multiprecision::expression_template_option ExpressionTemplates>
|
|
BOOST_CONSTEXPR_OR_CONST int numeric_limits<boost::multiprecision::number<boost::multiprecision::gmp_rational, ExpressionTemplates> >::max_digits10;
|
|
template <boost::multiprecision::expression_template_option ExpressionTemplates>
|
|
BOOST_CONSTEXPR_OR_CONST bool numeric_limits<boost::multiprecision::number<boost::multiprecision::gmp_rational, ExpressionTemplates> >::is_signed;
|
|
template <boost::multiprecision::expression_template_option ExpressionTemplates>
|
|
BOOST_CONSTEXPR_OR_CONST bool numeric_limits<boost::multiprecision::number<boost::multiprecision::gmp_rational, ExpressionTemplates> >::is_integer;
|
|
template <boost::multiprecision::expression_template_option ExpressionTemplates>
|
|
BOOST_CONSTEXPR_OR_CONST bool numeric_limits<boost::multiprecision::number<boost::multiprecision::gmp_rational, ExpressionTemplates> >::is_exact;
|
|
template <boost::multiprecision::expression_template_option ExpressionTemplates>
|
|
BOOST_CONSTEXPR_OR_CONST int numeric_limits<boost::multiprecision::number<boost::multiprecision::gmp_rational, ExpressionTemplates> >::radix;
|
|
template <boost::multiprecision::expression_template_option ExpressionTemplates>
|
|
BOOST_CONSTEXPR_OR_CONST int numeric_limits<boost::multiprecision::number<boost::multiprecision::gmp_rational, ExpressionTemplates> >::min_exponent;
|
|
template <boost::multiprecision::expression_template_option ExpressionTemplates>
|
|
BOOST_CONSTEXPR_OR_CONST int numeric_limits<boost::multiprecision::number<boost::multiprecision::gmp_rational, ExpressionTemplates> >::min_exponent10;
|
|
template <boost::multiprecision::expression_template_option ExpressionTemplates>
|
|
BOOST_CONSTEXPR_OR_CONST int numeric_limits<boost::multiprecision::number<boost::multiprecision::gmp_rational, ExpressionTemplates> >::max_exponent;
|
|
template <boost::multiprecision::expression_template_option ExpressionTemplates>
|
|
BOOST_CONSTEXPR_OR_CONST int numeric_limits<boost::multiprecision::number<boost::multiprecision::gmp_rational, ExpressionTemplates> >::max_exponent10;
|
|
template <boost::multiprecision::expression_template_option ExpressionTemplates>
|
|
BOOST_CONSTEXPR_OR_CONST bool numeric_limits<boost::multiprecision::number<boost::multiprecision::gmp_rational, ExpressionTemplates> >::has_infinity;
|
|
template <boost::multiprecision::expression_template_option ExpressionTemplates>
|
|
BOOST_CONSTEXPR_OR_CONST bool numeric_limits<boost::multiprecision::number<boost::multiprecision::gmp_rational, ExpressionTemplates> >::has_quiet_NaN;
|
|
template <boost::multiprecision::expression_template_option ExpressionTemplates>
|
|
BOOST_CONSTEXPR_OR_CONST bool numeric_limits<boost::multiprecision::number<boost::multiprecision::gmp_rational, ExpressionTemplates> >::has_signaling_NaN;
|
|
template <boost::multiprecision::expression_template_option ExpressionTemplates>
|
|
BOOST_CONSTEXPR_OR_CONST float_denorm_style numeric_limits<boost::multiprecision::number<boost::multiprecision::gmp_rational, ExpressionTemplates> >::has_denorm;
|
|
template <boost::multiprecision::expression_template_option ExpressionTemplates>
|
|
BOOST_CONSTEXPR_OR_CONST bool numeric_limits<boost::multiprecision::number<boost::multiprecision::gmp_rational, ExpressionTemplates> >::has_denorm_loss;
|
|
template <boost::multiprecision::expression_template_option ExpressionTemplates>
|
|
BOOST_CONSTEXPR_OR_CONST bool numeric_limits<boost::multiprecision::number<boost::multiprecision::gmp_rational, ExpressionTemplates> >::is_iec559;
|
|
template <boost::multiprecision::expression_template_option ExpressionTemplates>
|
|
BOOST_CONSTEXPR_OR_CONST bool numeric_limits<boost::multiprecision::number<boost::multiprecision::gmp_rational, ExpressionTemplates> >::is_bounded;
|
|
template <boost::multiprecision::expression_template_option ExpressionTemplates>
|
|
BOOST_CONSTEXPR_OR_CONST bool numeric_limits<boost::multiprecision::number<boost::multiprecision::gmp_rational, ExpressionTemplates> >::is_modulo;
|
|
template <boost::multiprecision::expression_template_option ExpressionTemplates>
|
|
BOOST_CONSTEXPR_OR_CONST bool numeric_limits<boost::multiprecision::number<boost::multiprecision::gmp_rational, ExpressionTemplates> >::traps;
|
|
template <boost::multiprecision::expression_template_option ExpressionTemplates>
|
|
BOOST_CONSTEXPR_OR_CONST bool numeric_limits<boost::multiprecision::number<boost::multiprecision::gmp_rational, ExpressionTemplates> >::tinyness_before;
|
|
template <boost::multiprecision::expression_template_option ExpressionTemplates>
|
|
BOOST_CONSTEXPR_OR_CONST float_round_style numeric_limits<boost::multiprecision::number<boost::multiprecision::gmp_rational, ExpressionTemplates> >::round_style;
|
|
|
|
#endif
|
|
|
|
#ifdef BOOST_MSVC
|
|
#pragma warning(pop)
|
|
#endif
|
|
|
|
} // namespace std
|
|
|
|
#endif
|