We usually try to compile for size, not for speed. Unfortunately with the
more powerful GOP infrastructure to handle all sorts of GOP operations, we
end up slowing down our copying hot path quite a lot.
So this patch moves the 4 possible GOP operation modes into separate
functions which call a common function again. The end result of that is
more optimized code that can properly do constant propagation throughout
its switch() statements and thus removes compares in the hot path.