How to print __int128 in g++?

I would recommend against overloading operator<< for __int128_t. The reason is that whenever you see cout << x for some integer type, you'd expect that all kinds of manipulators like std::hex or std::setw should also work. The most important guideline when overloading operators is: "do as the ints do".

As an alternative, I would recommend using a decimal_string(__int128_t) function that you can use as cout << decimal_string(x); in your code. For the string conversion, you can use the algorithm from any of the C-related Q&As. This makes it clear that you have special code for your 128-bit ints. Whenever the Standard Library upgrades to 128-bit support, you can drop it (and it's easy to grep for these functions).

If you don't need any of the fancy formatting options, writing your own << operator is trivial. Formally, I suspect that writing one for __int128_t would be considered undefined behavior, but practically, I think it would work, up until the library starts providing actual support for it (at which point, you'd retire your conversion operator).

Anyway, something like the following should work:

std::ostream&
operator<<( std::ostream& dest, __int128_t value )
{
    std::ostream::sentry s( dest );
    if ( s ) {
        __uint128_t tmp = value < 0 ? -value : value;
        char buffer[ 128 ];
        char* d = std::end( buffer );
        do
        {
            -- d;
            *d = "0123456789"[ tmp % 10 ];
            tmp /= 10;
        } while ( tmp != 0 );
        if ( value < 0 ) {
            -- d;
            *d = '-';
        }
        int len = std::end( buffer ) - d;
        if ( dest.rdbuf()->sputn( d, len ) != len ) {
            dest.setstate( std::ios_base::badbit );
        }
    }
    return dest;
}

Note that this is just a quicky, temporary fix, until the time the g++ library supports the type. It counts on 2's complement, wrap around on overflow, for __int128_t, but I'd be very surprised if that wasn't the case (formally, it's undefined behavior). If not, you'll need to fix up the initialization of tmp. And of course, it doesn't handle any of the formatting options; you can add as desired. (Handling padding and the adjustfield correctly can be non-trivial.)