123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248 |
- ///////////////////////////////////////////////////////////////////////////////
- // Copyright 2018 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_MP_EIGEN_HPP
- #define BOOST_MP_EIGEN_HPP
- #include <boost/multiprecision/number.hpp>
- #include <Eigen/Core>
- //
- // Generic Eigen support code:
- //
- namespace Eigen {
- template <class B1, class B2>
- struct NumTraitsImp;
- template <class B1>
- struct NumTraitsImp<B1, B1>
- {
- using self_type = B1;
- using Real = typename boost::multiprecision::scalar_result_from_possible_complex<self_type>::type;
- using NonInteger = self_type; // Not correct but we can't do much better??
- using Literal = double;
- using Nested = self_type;
- enum
- {
- IsComplex = boost::multiprecision::number_category<self_type>::value == boost::multiprecision::number_kind_complex,
- IsInteger = boost::multiprecision::number_category<self_type>::value == boost::multiprecision::number_kind_integer,
- ReadCost = 1,
- AddCost = 4,
- MulCost = 8,
- IsSigned = std::numeric_limits<self_type>::is_specialized ? std::numeric_limits<self_type>::is_signed : true,
- RequireInitialization = 1,
- };
- static Real epsilon()
- {
- static_assert(std::numeric_limits<Real>::is_specialized, "Eigen's NumTraits instantiated on a type with no numeric_limits support. Are you using a variable precision type?");
- return std::numeric_limits<Real>::epsilon();
- }
- static Real dummy_precision()
- {
- return 1000 * epsilon();
- }
- static Real highest()
- {
- static_assert(std::numeric_limits<Real>::is_specialized, "Eigen's NumTraits instantiated on a type with no numeric_limits support. Are you using a variable precision type?");
- return (std::numeric_limits<Real>::max)();
- }
- static Real lowest()
- {
- static_assert(std::numeric_limits<Real>::is_specialized, "Eigen's NumTraits instantiated on a type with no numeric_limits support. Are you using a variable precision type?");
- return (std::numeric_limits<Real>::min)();
- }
- static int digits10_imp(const std::integral_constant<bool, true>&)
- {
- static_assert(std::numeric_limits<Real>::is_specialized, "Eigen's NumTraits instantiated on a type with no numeric_limits support. Are you using a variable precision type?");
- return std::numeric_limits<Real>::digits10;
- }
- template <bool B>
- static int digits10_imp(const std::integral_constant<bool, B>&)
- {
- return Real::thread_default_precision();
- }
- static int digits10()
- {
- return digits10_imp(std::integral_constant < bool, std::numeric_limits<Real>::digits10 && (std::numeric_limits<Real>::digits10 != INT_MAX) ? true : false > ());
- }
- static int digits()
- {
- // return the number of digits in the component type in case Real is complex
- // and we have no numeric_limits specialization.
- static_assert(std::numeric_limits<Real>::is_specialized, "Eigen's NumTraits instantiated on a type with no numeric_limits support. Are you using a variable precision type?");
- return std::numeric_limits<Real>::digits;
- }
- static int min_exponent()
- {
- static_assert(std::numeric_limits<Real>::is_specialized, "Eigen's NumTraits instantiated on a type with no numeric_limits support. Are you using a variable precision type?");
- return std::numeric_limits<Real>::min_exponent;
- }
- static int max_exponent()
- {
- static_assert(std::numeric_limits<Real>::is_specialized, "Eigen's NumTraits instantiated on a type with no numeric_limits support. Are you using a variable precision type?");
- return std::numeric_limits<Real>::max_exponent;
- }
- static Real infinity()
- {
- static_assert(std::numeric_limits<Real>::is_specialized, "Eigen's NumTraits instantiated on a type with no numeric_limits support. Are you using a variable precision type?");
- return std::numeric_limits<Real>::infinity();
- }
- static Real quiet_NaN()
- {
- static_assert(std::numeric_limits<Real>::is_specialized, "Eigen's NumTraits instantiated on a type with no numeric_limits support. Are you using a variable precision type?");
- return std::numeric_limits<Real>::quiet_NaN();
- }
- };
-
- template <class B1, class B2>
- struct NumTraitsImp : public NumTraitsImp<B2, B2>
- {
- //
- // This version is instantiated when B1 and B2 are different types, this happens for rational/complex/interval
- // types, in which case many methods defer to those of the "component type" B2.
- //
- using self_type = B1;
- using Real = typename boost::multiprecision::scalar_result_from_possible_complex<self_type>::type;
- using NonInteger = self_type; // Not correct but we can't do much better??
- using Literal = double;
- using Nested = self_type;
- enum
- {
- IsComplex = boost::multiprecision::number_category<self_type>::value == boost::multiprecision::number_kind_complex,
- IsInteger = boost::multiprecision::number_category<self_type>::value == boost::multiprecision::number_kind_integer,
- ReadCost = 1,
- AddCost = 4,
- MulCost = 8,
- IsSigned = std::numeric_limits<self_type>::is_specialized ? std::numeric_limits<self_type>::is_signed : true,
- RequireInitialization = 1,
- };
- static B2 epsilon()
- {
- return NumTraitsImp<B2, B2>::epsilon();
- }
- static B2 dummy_precision()
- {
- return 1000 * epsilon();
- }
- static B2 highest()
- {
- return NumTraitsImp<B2, B2>::highest();
- }
- static B2 lowest()
- {
- return NumTraitsImp<B2, B2>::lowest();
- }
- static int digits10()
- {
- return NumTraitsImp<B2, B2>::digits10();
- }
- static int digits()
- {
- return NumTraitsImp<B2, B2>::digits();
- }
- static int min_exponent()
- {
- return NumTraitsImp<B2, B2>::min_exponent();
- }
- static int max_exponent()
- {
- return NumTraitsImp<B2, B2>::max_exponent();
- }
- static B2 infinity()
- {
- return NumTraitsImp<B2, B2>::infinity();
- }
- static B2 quiet_NaN()
- {
- return NumTraitsImp<B2, B2>::quiet_NaN();
- }
- };
- template <class Backend, boost::multiprecision::expression_template_option ExpressionTemplates>
- struct NumTraits<boost::multiprecision::number<Backend, ExpressionTemplates> > : public NumTraitsImp<boost::multiprecision::number<Backend, ExpressionTemplates>, typename boost::multiprecision::number<Backend, ExpressionTemplates>::value_type>
- {};
- template <class tag, class Arg1, class Arg2, class Arg3, class Arg4>
- struct NumTraits<boost::multiprecision::detail::expression<tag, Arg1, Arg2, Arg3, Arg4> > : public NumTraits<typename boost::multiprecision::detail::expression<tag, Arg1, Arg2, Arg3, Arg4>::result_type>
- {};
- #define BOOST_MP_EIGEN_SCALAR_TRAITS_DECL(A) \
- template <class Backend, boost::multiprecision::expression_template_option ExpressionTemplates, typename BinaryOp> \
- struct ScalarBinaryOpTraits<boost::multiprecision::number<Backend, ExpressionTemplates>, A, BinaryOp> \
- { \
- /*static_assert(boost::multiprecision::is_compatible_arithmetic_type<A, boost::multiprecision::number<Backend, ExpressionTemplates> >::value, "Interoperability with this arithmetic type is not supported.");*/ \
- using ReturnType = boost::multiprecision::number<Backend, ExpressionTemplates>; \
- }; \
- template <class Backend, boost::multiprecision::expression_template_option ExpressionTemplates, typename BinaryOp> \
- struct ScalarBinaryOpTraits<A, boost::multiprecision::number<Backend, ExpressionTemplates>, BinaryOp> \
- { \
- /*static_assert(boost::multiprecision::is_compatible_arithmetic_type<A, boost::multiprecision::number<Backend, ExpressionTemplates> >::value, "Interoperability with this arithmetic type is not supported.");*/ \
- using ReturnType = boost::multiprecision::number<Backend, ExpressionTemplates>; \
- };
- BOOST_MP_EIGEN_SCALAR_TRAITS_DECL(float)
- BOOST_MP_EIGEN_SCALAR_TRAITS_DECL(double)
- BOOST_MP_EIGEN_SCALAR_TRAITS_DECL(long double)
- BOOST_MP_EIGEN_SCALAR_TRAITS_DECL(char)
- BOOST_MP_EIGEN_SCALAR_TRAITS_DECL(unsigned char)
- BOOST_MP_EIGEN_SCALAR_TRAITS_DECL(signed char)
- BOOST_MP_EIGEN_SCALAR_TRAITS_DECL(short)
- BOOST_MP_EIGEN_SCALAR_TRAITS_DECL(unsigned short)
- BOOST_MP_EIGEN_SCALAR_TRAITS_DECL(int)
- BOOST_MP_EIGEN_SCALAR_TRAITS_DECL(unsigned int)
- BOOST_MP_EIGEN_SCALAR_TRAITS_DECL(long)
- BOOST_MP_EIGEN_SCALAR_TRAITS_DECL(unsigned long)
- #if 0
- template<class Backend, boost::multiprecision::expression_template_option ExpressionTemplates, class Backend2, boost::multiprecision::expression_template_option ExpressionTemplates2, typename BinaryOp>
- struct ScalarBinaryOpTraits<boost::multiprecision::number<Backend, ExpressionTemplates>, boost::multiprecision::number<Backend2, ExpressionTemplates2>, BinaryOp>
- {
- static_assert(
- boost::multiprecision::is_compatible_arithmetic_type<boost::multiprecision::number<Backend2, ExpressionTemplates2>, boost::multiprecision::number<Backend, ExpressionTemplates> >::value
- || boost::multiprecision::is_compatible_arithmetic_type<boost::multiprecision::number<Backend, ExpressionTemplates>, boost::multiprecision::number<Backend2, ExpressionTemplates2> >::value, "Interoperability with this arithmetic type is not supported.");
- using ReturnType = typename std::conditional<std::is_convertible<boost::multiprecision::number<Backend2, ExpressionTemplates2>, boost::multiprecision::number<Backend, ExpressionTemplates> >::value,
- boost::multiprecision::number<Backend, ExpressionTemplates>, boost::multiprecision::number<Backend2, ExpressionTemplates2> >::type;
- };
- template<unsigned D, typename BinaryOp>
- struct ScalarBinaryOpTraits<boost::multiprecision::number<boost::multiprecision::backends::mpc_complex_backend<D>, boost::multiprecision::et_on>, boost::multiprecision::mpfr_float, BinaryOp>
- {
- using ReturnType = boost::multiprecision::number<boost::multiprecision::backends::mpc_complex_backend<D>, boost::multiprecision::et_on>;
- };
- template<typename BinaryOp>
- struct ScalarBinaryOpTraits<boost::multiprecision::mpfr_float, boost::multiprecision::mpc_complex, BinaryOp>
- {
- using ReturnType = boost::multiprecision::number<boost::multiprecision::backends::mpc_complex_backend<0>, boost::multiprecision::et_on>;
- };
- template<class Backend, boost::multiprecision::expression_template_option ExpressionTemplates, typename BinaryOp>
- struct ScalarBinaryOpTraits<boost::multiprecision::number<Backend, ExpressionTemplates>, boost::multiprecision::number<Backend, ExpressionTemplates>, BinaryOp>
- {
- using ReturnType = boost::multiprecision::number<Backend, ExpressionTemplates>;
- };
- #endif
- template <class Backend, boost::multiprecision::expression_template_option ExpressionTemplates, class tag, class Arg1, class Arg2, class Arg3, class Arg4, typename BinaryOp>
- struct ScalarBinaryOpTraits<boost::multiprecision::number<Backend, ExpressionTemplates>, boost::multiprecision::detail::expression<tag, Arg1, Arg2, Arg3, Arg4>, BinaryOp>
- {
- static_assert(std::is_convertible<typename boost::multiprecision::detail::expression<tag, Arg1, Arg2, Arg3, Arg4>::result_type, boost::multiprecision::number<Backend, ExpressionTemplates> >::value, "Interoperability with this arithmetic type is not supported.");
- using ReturnType = boost::multiprecision::number<Backend, ExpressionTemplates>;
- };
- template <class tag, class Arg1, class Arg2, class Arg3, class Arg4, class Backend, boost::multiprecision::expression_template_option ExpressionTemplates, typename BinaryOp>
- struct ScalarBinaryOpTraits<boost::multiprecision::detail::expression<tag, Arg1, Arg2, Arg3, Arg4>, boost::multiprecision::number<Backend, ExpressionTemplates>, BinaryOp>
- {
- static_assert(std::is_convertible<typename boost::multiprecision::detail::expression<tag, Arg1, Arg2, Arg3, Arg4>::result_type, boost::multiprecision::number<Backend, ExpressionTemplates> >::value, "Interoperability with this arithmetic type is not supported.");
- using ReturnType = boost::multiprecision::number<Backend, ExpressionTemplates>;
- };
- namespace numext {
- using boost::multiprecision::conj;
- }
- } // namespace Eigen
- #endif
|