Add R2 optimized variants of clear_bit, set_bit and test_and_clear_bit.
With gcc 4.1.1 this saves 1592 bytes on a defconfig (minus IPv6) kernel.
Turns out that R2 bitop instructions are no gain for the other bitop
functions.
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>