Or

    x / 2 + y / 2 + (x & y & 1)

Edit: This is the same you wrote, but it gets the wrong number for negative values, for negative ints the rounding will go up and not down.

And this is exactly why I like to use higher level programming languages. Let someone smart figure all this out for me, and give me (grug) a generic binary search routine that works on arbitrary collections of arbitrary ordered things.