percpu, x86: Generic inc / dec percpu instructions

Optimize code generated for percpu access by checking for increment and
decrements.

tj: fix incorrect usage of __builtin_constant_p() and restructure
    percpu_add_op() macro.

Signed-off-by: Christoph Lameter <cl@linux-foundation.org>
Signed-off-by: Tejun Heo <tj@kernel.org>