_umul128 was not declared error when building Boost library

[vejbomar]

When building Boost library with either x86_64-pc-linux-gnu -> aarch64-w64-mingw32 or aarch64-pc-linux-gnu -> aarch64-w64-mingw32 toolchain I got the following errors about missing _umul128 from json and charconv modules.

json module

In file included from ./boost/json/detail/charconv/detail/fast_float/fast_float.hpp:11,
                 from ./boost/json/detail/charconv/impl/from_chars.ipp:14,
                 from ./boost/json/src.hpp:58,
                 from /home/vejby/boost-build-19-full/libs/json/src/src.cpp:10:
./boost/json/detail/charconv/detail/fast_float/float_common.hpp: In function ‘boost::json::detail::charconv::detail::fast_float::value128 boost::json::detail::charconv::detail::fast_float::full_multiplication(uint64_t, uint64_t)’:
./boost/json/detail/charconv/detail/fast_float/float_common.hpp:212:16: error: ‘_umul128’ was not declared in this scope
  212 |   answer.low = _umul128(a, b, &high); // _umul128 not available on ARM64
      |                ^~~~~~~~

charconv module

In file included from ./boost/charconv/detail/fast_float/fast_float.hpp:11,
                 from libs/charconv/build/../src/from_chars.cpp:16:
./boost/charconv/detail/fast_float/float_common.hpp: In function ‘boost::charconv::detail::fast_float::value128 boost::charconv::detail::fast_float::full_multiplication(uint64_t, uint64_t)’:
./boost/charconv/detail/fast_float/float_common.hpp:254:16: error: ‘_umul128’ was not declared in this scope; did you mean ‘umul128’?
  254 |   answer.low = _umul128(a, b, &high); // _umul128 not available on ARM64
      |                ^~~~~~~~
      |                umul128

The code seems to come originally from fast_float library https://github.com/fastfloat/fast_float/blob/4266c34620b3120718d8d631c354ba22e9607b36/include/fast_float/float_common.h#L348 and there're multiple versions of the function, so I think the problem is just that the wrong version is selected for arm64 + win + mingw combination.

[eukarpov]

please apply a fix in a similar way to how it was done here
fastfloat/fast_float#269

[vejbomar]

PR with fix accepted in Boost.charconv: boostorg/charconv#230

Created similar PR in Boost.JSON: boostorg/json#1057

[vejbomar]

Both PRs were accepted and merged into develop branches of charconv and JSON modules, so the fixes should be available in the next Boost release.