The bitops code was optimized to generate test under mask instructions
with the __bitops_byte() helper. However that was many years ago and
in the meantime a lot of new instructions were introduced.

Changing the code so that it always operates on longs nowadays even
generates shorter code (~ -20kb, defconfig, gcc 10, march=zE12).
Signed-off-by: Heiko Carstens <hca@linux.ibm.com>