Our __ffs implementation is straightforward and non-optimal,
__builtin_ffsl should be faster.
Signed-off-by: Andrei Vagin <avagin@virtuozzo.com>