Hello,
in boost::math::cstdfloat there are several inconsistencies with __float128/_Quad. Generally these are only available with gcc and intel. Since I don't have an intel compiler myself, you might have to modify this path; for now it should behave like the previous implementations.
hypot(x, y, z) correct result even with large values; slower than previous naive implementation
isfinite(x) branch-free
limits <boost/math/cstdfloat/cstdfloat_limits.hpp>
All functions can now be provided constexpr noexcept.
future
branch-free isgreater, isgreaterequal, ... I don't understand the BOOST_PREVENT_MACRO_SUBSTITUTION inside the functions, otherwise I could have adjusted that right away.
With Matt's boost::charconv the IO functions can be provided more efficiently (operator<<, operator>>, to_string, ...).
It might be useful to provide quadmath as a separate sublibrary in boost::math, e.g. to avoid the #define/#undef orgies in <boost/math/cstdfloat/cstdfloat_cmath.hpp>.
Hello, in boost::math::cstdfloat there are several inconsistencies with __float128/_Quad. Generally these are only available with gcc and intel. Since I don't have an intel compiler myself, you might have to modify this path; for now it should behave like the previous implementations.
types <boost/math/cstdfloat/cstdfloat_types.hpp> intel https://community.intel.com/t5/Intel-C-Compiler/C-Compiler-128-bit-floating-point/m-p/1216949
defines/constants <boost/math/cstdfloat/cstdfloat_types.hpp> gcc
functions <boost/math/cstdfloat/cstdfloat_cmath.hpp> modified:
limits <boost/math/cstdfloat/cstdfloat_limits.hpp> All functions can now be provided constexpr noexcept.
future
cstdfloat_types.hpp.txt cstdfloat_cmath.hpp.txt cstdfloat_limits.hpp.txt
thx Gero