[PATCH] atomic.h fixes, 2003-03-26

[Roland McGrath]

Hey, no fair posting from the future!

I put your changes in.  It's unfortunate the powerpc atomic_add change is
unavoidable.  I have no clue about ppc really, but it seems like loading a
register with a constant instead of using the addi form will be a (micro)
performance hit and at least for atomic_increment/decrement{,_and_test}
uses (of course those could be done explicitly in powerpc/bits/atomic.h).


Thanks,
Roland