Sprout/sprout/math/tgamma.hpp
2019-01-07 17:47:17 +09:00

166 lines
6.2 KiB
C++
Raw Blame History

/*=============================================================================
Copyright (c) 2011-2019 Bolero MURAKAMI
https://github.com/bolero-MURAKAMI/Sprout
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 SPROUT_MATH_TGAMMA_HPP
#define SPROUT_MATH_TGAMMA_HPP
#include <type_traits>
#include <sprout/config.hpp>
#include <sprout/workaround/std/cstddef.hpp>
#include <sprout/limits.hpp>
#include <sprout/math/detail/config.hpp>
#include <sprout/math/detail/float_compute.hpp>
#include <sprout/math/isnan.hpp>
#include <sprout/math/copysign.hpp>
#include <sprout/math/factorial.hpp>
#include <sprout/math/iround.hpp>
#include <sprout/math/is_integer.hpp>
#include <sprout/algorithm/clamp.hpp>
#include <sprout/type_traits/enabler_if.hpp>
#include HDR_ALGORITHM_MIN_MAX_SSCRISK_CEL_OR_SPROUT
namespace sprout {
namespace math {
namespace detail {
#if SPROUT_USE_BUILTIN_CMATH_FUNCTION
inline SPROUT_CONSTEXPR float
builtin_tgamma(float x) {
return __builtin_tgammaf(x);
}
inline SPROUT_CONSTEXPR double
builtin_tgamma(double x) {
return __builtin_tgamma(x);
}
inline SPROUT_CONSTEXPR long double
builtin_tgamma(long double x) {
return __builtin_tgammal(x);
}
#endif
template<typename T>
inline SPROUT_CONSTEXPR T
tgamma_impl_y(T w) {
return ((((((((((((
-1.99542863674e-7 * w + 1.337767384067e-6) * w
- 2.591225267689e-6) * w - 1.7545539395205e-5) * w
+ 1.45596568617526e-4) * w - 3.60837876648255e-4) * w
- 8.04329819255744e-4) * w + 0.008023273027855346) * w
- 0.017645244547851414) * w - 0.024552490005641278) * w
+ 0.19109110138763841) * w - 0.233093736421782878) * w
- 0.422784335098466784) * w + 0.99999999999999999
;
}
template<typename T>
inline SPROUT_CONSTEXPR T
tgamma_impl_t_pos(T x, std::size_t n, std::size_t last) {
return last - n == 1
? x - static_cast<T>(n)
: sprout::math::detail::tgamma_impl_t_pos(x, n, n + (last - n) / 2)
* sprout::math::detail::tgamma_impl_t_pos(x, n + (last - n) / 2, last)
;
}
template<typename T>
inline SPROUT_CONSTEXPR T
tgamma_impl_t_neg(T x, std::size_t n, std::size_t last) {
return last - n == 1
? x + static_cast<T>(n)
: sprout::math::detail::tgamma_impl_t_neg(x, n, n + (last - n) / 2)
* sprout::math::detail::tgamma_impl_t_neg(x, n + (last - n) / 2, last)
;
}
template<typename T>
inline SPROUT_CONSTEXPR T
tgamma_impl_t_pos_rec(T x, std::size_t n, std::size_t last) {
return last - n == 1
? T(1) / (x - static_cast<T>(n))
: sprout::math::detail::tgamma_impl_t_pos_rec(x, n, n + (last - n) / 2)
* sprout::math::detail::tgamma_impl_t_pos_rec(x, n + (last - n) / 2, last)
;
}
template<typename T>
inline SPROUT_CONSTEXPR T
tgamma_impl_t_neg_rec(T x, std::size_t n, std::size_t last) {
return last - n == 1
? T(1) / (x + static_cast<T>(n))
: sprout::math::detail::tgamma_impl_t_neg_rec(x, n, n + (last - n) / 2)
* sprout::math::detail::tgamma_impl_t_neg_rec(x, n + (last - n) / 2, last)
;
}
template<typename T>
inline SPROUT_CONSTEXPR T
tgamma_impl_2_pos_rec(T x, T y, T t) {
return t == 0 ? sprout::numeric_limits<T>::infinity()
: (x - T(1)) / y / t
;
}
template<typename T>
inline SPROUT_CONSTEXPR T
tgamma_impl_1(T x, int n) {
return n == 1 ? (x - T(1)) / sprout::math::detail::tgamma_impl_y(x - (n + 2))
: n == 0 ? T(1) / sprout::math::detail::tgamma_impl_y(x - (n + 2))
: n == static_cast<int>(sprout::math::factorial_limit<T>())
? sprout::math::detail::tgamma_impl_2_pos_rec(x, sprout::math::detail::tgamma_impl_y(x - (n + 2)), sprout::math::detail::tgamma_impl_t_pos_rec(x, 2, n + 1))
: n == -static_cast<int>(sprout::math::factorial_limit<T>())
? T(1) / sprout::math::detail::tgamma_impl_y(x - (n + 2)) * sprout::math::detail::tgamma_impl_t_neg_rec(x, 0, -n)
: n > 1 ? (x - T(1)) / sprout::math::detail::tgamma_impl_y(x - (n + 2)) * sprout::math::detail::tgamma_impl_t_pos(x, 2, n + 1)
: T(1) / sprout::math::detail::tgamma_impl_y(x - (n + 2)) / sprout::math::detail::tgamma_impl_t_neg(x, 0, -n)
;
}
template<typename T>
inline SPROUT_CONSTEXPR T
tgamma_impl(T x) {
return sprout::math::detail::tgamma_impl_1(
x,
sprout::clamp(
sprout::math::iround(x - T(2)),
-static_cast<int>(sprout::math::factorial_limit<T>()),
static_cast<int>(sprout::math::factorial_limit<T>())
)
);
}
} // namespace detail
//
// tgamma
//
// issue:
// [ !(SPROUT_USE_BUILTIN_CMATH_FUNCTION || SPROUT_USE_BUILTIN_COPYSIGN_FUNCTION) ]
// tgamma(-0) returns -<2D><> .
// # returns +<2B><> . ( same as tgamma(+0) )
//
template<
typename FloatType,
typename sprout::enabler_if<std::is_floating_point<FloatType>::value>::type = sprout::enabler
>
inline SPROUT_CONSTEXPR FloatType
tgamma(FloatType x) {
return sprout::math::isnan(x) ? x
: x == 0 ? sprout::math::copysign(sprout::numeric_limits<FloatType>::infinity(), x)
: x == -sprout::numeric_limits<FloatType>::infinity() ? -sprout::numeric_limits<FloatType>::quiet_NaN()
: x == sprout::numeric_limits<FloatType>::infinity() ? sprout::numeric_limits<FloatType>::infinity()
: x < 0 && sprout::math::is_integer(x) ? sprout::numeric_limits<FloatType>::quiet_NaN()
#if SPROUT_USE_BUILTIN_CMATH_FUNCTION
: sprout::math::detail::builtin_tgamma(x)
#else
: static_cast<FloatType>(sprout::math::detail::tgamma_impl(static_cast<typename sprout::math::detail::float_compute<FloatType>::type>(x)))
#endif
;
}
template<
typename IntType,
typename sprout::enabler_if<std::is_integral<IntType>::value>::type = sprout::enabler
>
inline SPROUT_CONSTEXPR double
tgamma(IntType x) {
return sprout::math::tgamma(static_cast<double>(x));
}
} // namespace math
using sprout::math::tgamma;
} // namespace sprout
#endif // #ifndef SPROUT_MATH_TGAMMA_HPP