Use adx instruction when available

Author: mborland

include/boost/decimal/detail/emulated128.hpp

@@ -1328,21 +1328,35 @@ constexpr auto int128::operator>(int rhs) const noexcept -> bool
 
 constexpr auto operator+(const int128& lhs, const int128& rhs) noexcept -> int128
 {
-    #ifdef BOOST_DECIMAL_HAS_INT128
+    #if (defined(BOOST_DECIMAL_HAS_X64_INTRINSICS) || defined(BOOST_DECIMAL_HAS_MSVC_64BIT_INTRINSICS)) && !defined(BOOST_DECIMAL_NO_CONSTEVAL_DETECTION)
+    if (!BOOST_DECIMAL_IS_CONSTANT_EVALUATED(lhs.low))
+    {
+        unsigned long long low {};
+        unsigned long long high {};
 
-    const auto lhs_full {(static_cast<__uint128_t>(lhs.high) << 64) | lhs.low};
-    const auto rhs_full {(static_cast<__uint128_t>(rhs.high) << 64) | rhs.low};
-    const auto result {lhs_full + rhs_full};
+        const auto carry {BOOST_DECIMAL_ADD_CARRY(0, lhs.low, rhs.low, &low)};
+        BOOST_DECIMAL_ADD_CARRY(carry, lhs.high, rhs.high, &high);
 
-    return {static_cast<std::int64_t>(result >> 64), static_cast<std::uint64_t>(result)};
+        return {static_cast<std::int64_t>(high), low};
+    }
+    #endif
+    {
+        #ifdef BOOST_DECIMAL_HAS_INT128
 
-    #else
+        const auto lhs_full {(static_cast<__uint128_t>(lhs.high) << 64) | lhs.low};
+        const auto rhs_full {(static_cast<__uint128_t>(rhs.high) << 64) | rhs.low};
+        const auto result {lhs_full + rhs_full};
 
-    const auto new_low {lhs.low + rhs.low};
-    const auto new_high {lhs.high + rhs.high + static_cast<std::int64_t>(new_low < lhs.low)};
-    return int128{new_high, new_low};
+        return {static_cast<std::int64_t>(result >> 64), static_cast<std::uint64_t>(result)};
 
-    #endif
+        #else
+
+        const auto new_low {lhs.low + rhs.low};
+        const auto new_high {lhs.high + rhs.high + static_cast<std::int64_t>(new_low < lhs.low)};
+        return int128{new_high, new_low};
+
+        #endif
+    }
 }
 
 constexpr auto operator-(const int128& lhs, int128& rhs) noexcept -> int128