Then, we must convince ourselves "by eyeball" that out candidate INV is indeed invariant.
Check its validity just prior to loop execution for the first time: here, q = 0, r = x, so INV reduces to 0 <= x and x = x, which is true.
Now check what one loop execution does to the INV... r get smaller by y, q goes up by 1, so the effect on INV is x = (q+1)*y + (r-y) = q*y + r
So things look good... lets formalize this thinking.