resolve libstdc++/5198

[Roman Zippel]

Hi,

Richard Henderson wrote:

 > Also, I don't see anything in the moto docs that require addl with a
 > memory destination to be atomic.  It may well be that it is on
 > certain implementations, but I don't know which for sure. So make
 > it use the known atomic __exchange_and_add and let the compiler
 > throw away the result.


add.l is only atomic local to the cpu, so that would be only a problem
for m68k smp systems. I heard of rumors that such machines would exist,
but does gcc support them?

bye, Roman