Use fenv rounding to achieve correct value

Author: mborland

include/boost/decimal/detail/cmath/rint.hpp

@@ -28,20 +28,15 @@ namespace decimal {
 
 namespace detail {
 
-template <BOOST_DECIMAL_INTEGRAL T1, BOOST_DECIMAL_INTEGRAL T2>
-constexpr auto rint_impl(T1& sig, T2 exp, const bool)
+template <BOOST_DECIMAL_DECIMAL_FLOATING_TYPE TargetType, BOOST_DECIMAL_INTEGRAL T1, BOOST_DECIMAL_INTEGRAL T2>
+constexpr auto rint_impl(T1& sig, T2 exp, const bool is_neg)
 {
     const T2 abs_exp { (exp < T2(0)) ? -exp : exp };
 
-    sig /= detail::pow10(static_cast<T1>(abs_exp - 1));
+    const auto res {detail::impl::divmod(sig, detail::pow10(static_cast<T1>(abs_exp - 1)))};
+    sig = res.quotient;
 
-    const auto trailing_num {static_cast<std::uint32_t>(sig % 10U)};
-    sig /= 10U;
-
-    if (trailing_num >= 5U)
-    {
-        ++sig;
-    }
+    detail::fenv_round<TargetType>(sig, is_neg, res.remainder != 0U);
 }
 
 // MSVC 14.1 warns of unary minus being applied to unsigned type from numeric_limits::min
@@ -97,7 +92,7 @@ constexpr auto lrint_impl(const T num) noexcept -> Int
         return 0;
     }
 
-    detail::rint_impl(sig, expptr, is_neg);
+    detail::rint_impl<T>(sig, expptr, is_neg);
 
     auto res {static_cast<Int>(sig)};
     if (is_neg)
@@ -147,7 +142,7 @@ constexpr auto rint(const T num) noexcept
         return is_neg ? -zero : zero;
     }
 
-    detail::rint_impl(sig, expptr, is_neg);
+    detail::rint_impl<T>(sig, expptr, is_neg);
 
     return {sig, 0, is_neg};
 }