For 8/16, use the 32 bit variant and properly subtract the added
leading zero bits.
Reviewed-by: NRichard Henderson <richard.henderson@linaro.org>
Signed-off-by: NDavid Hildenbrand <david@redhat.com>