/////////////////////////////////////////////////////////////////////////////// // Copyright Christopher Kormanyos 2014. // Copyright John Maddock 2014. // Copyright Paul Bristow 2014. // 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) // // Implement quadruple-precision std::numeric_limits<> support. #ifndef _BOOST_CSTDFLOAT_LIMITS_2014_01_09_HPP_ #define _BOOST_CSTDFLOAT_LIMITS_2014_01_09_HPP_ #include #if defined(BOOST_CSTDFLOAT_HAS_INTERNAL_FLOAT128_T) && defined(BOOST_MATH_USE_FLOAT128) && !defined(BOOST_CSTDFLOAT_NO_LIBQUADMATH_SUPPORT) #include // Define the name of the global quadruple-precision function to be used for // calculating quiet_NaN() in the specialization of std::numeric_limits<>. #if defined(BOOST_INTEL) #define BOOST_CSTDFLOAT_FLOAT128_SQRT __sqrtq #elif defined(__GNUC__) #define BOOST_CSTDFLOAT_FLOAT128_SQRT sqrtq #endif // Forward declaration of the quadruple-precision square root function. extern "C" boost::math::cstdfloat::detail::float_internal128_t BOOST_CSTDFLOAT_FLOAT128_SQRT(boost::math::cstdfloat::detail::float_internal128_t) throw(); namespace std { template<> class numeric_limits { public: BOOST_STATIC_CONSTEXPR bool is_specialized = true; static boost::math::cstdfloat::detail::float_internal128_t (min) () BOOST_NOEXCEPT { return BOOST_CSTDFLOAT_FLOAT128_MIN; } static boost::math::cstdfloat::detail::float_internal128_t (max) () BOOST_NOEXCEPT { return BOOST_CSTDFLOAT_FLOAT128_MAX; } static boost::math::cstdfloat::detail::float_internal128_t lowest() BOOST_NOEXCEPT { return -(max)(); } BOOST_STATIC_CONSTEXPR int digits = 113; BOOST_STATIC_CONSTEXPR int digits10 = 34; BOOST_STATIC_CONSTEXPR int max_digits10 = 36; 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 boost::math::cstdfloat::detail::float_internal128_t epsilon () { return BOOST_CSTDFLOAT_FLOAT128_EPS; } static boost::math::cstdfloat::detail::float_internal128_t round_error() { return BOOST_FLOAT128_C(0.5); } BOOST_STATIC_CONSTEXPR int min_exponent = -16381; BOOST_STATIC_CONSTEXPR int min_exponent10 = static_cast((min_exponent * 301L) / 1000L); BOOST_STATIC_CONSTEXPR int max_exponent = +16384; BOOST_STATIC_CONSTEXPR int max_exponent10 = static_cast((max_exponent * 301L) / 1000L); BOOST_STATIC_CONSTEXPR bool has_infinity = true; BOOST_STATIC_CONSTEXPR bool has_quiet_NaN = true; BOOST_STATIC_CONSTEXPR bool has_signaling_NaN = false; BOOST_STATIC_CONSTEXPR float_denorm_style has_denorm = denorm_absent; BOOST_STATIC_CONSTEXPR bool has_denorm_loss = false; static boost::math::cstdfloat::detail::float_internal128_t infinity () { return BOOST_FLOAT128_C(1.0) / BOOST_FLOAT128_C(0.0); } static boost::math::cstdfloat::detail::float_internal128_t quiet_NaN () { return ::BOOST_CSTDFLOAT_FLOAT128_SQRT(BOOST_FLOAT128_C(-1.0)); } static boost::math::cstdfloat::detail::float_internal128_t signaling_NaN() { return BOOST_FLOAT128_C(0.0); } static boost::math::cstdfloat::detail::float_internal128_t denorm_min () { return BOOST_FLOAT128_C(0.0); } BOOST_STATIC_CONSTEXPR bool is_iec559 = true; 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_to_nearest; }; } // namespace std #endif // Not BOOST_CSTDFLOAT_NO_LIBQUADMATH_SUPPORT (i.e., the user would like to have libquadmath support) #endif // _BOOST_CSTDFLOAT_LIMITS_2014_01_09_HPP_