Function full_multiplication

source/extern/fast_float.h:470–494

compute 64-bit a*b

// compute 64-bit a*b
fastfloat_really_inline FASTFLOAT_CONSTEXPR20 value128
full_multiplication(uint64_t a, uint64_t b) {
  if (cpp20_and_in_constexpr()) {
    value128 answer;
    answer.low = umul128_generic(a, b, &answer.high);
    return answer;
  }
  value128 answer;
#if defined(_M_ARM64) && !defined(__MINGW32__)
  // ARM64 has native support for 64-bit multiplications, no need to emulate
  // But MinGW on ARM64 doesn't have native support for 64-bit multiplications
  answer.high = __umulh(a, b);
  answer.low = a * b;
#elif defined(FASTFLOAT_32BIT) ||                                              \
    (defined(_WIN64) && !defined(__clang__) && !defined(_M_ARM64))
  answer.low = _umul128(a, b, &answer.high); // _umul128 not available on ARM64
#elif defined(FASTFLOAT_64BIT) && defined(__SIZEOF_INT128__)
  __uint128_t r = ((__uint128_t)a) * b;
  answer.low = uint64_t(r);
  answer.high = uint64_t(r >> 64);
#else
  answer.low = umul128_generic(a, b, &answer.high);
#endif
  return answer;
}

struct adjusted_mantissa {
  uint64_t mantissa{0};