[RFC] libstdc++ atomicity, x86 agnostic version

[Benjamin Kosnik]

>I have just finished running some more tests and I can confirm that the
>numbers
>are basically the same that I have reported this morning (italian
>time), which I had obtained just removing the inline keyword from the
>existing code.

Ok. I think this is something we can live with, pe.

>Therefore, all in all, you have my vote, at least for 3.4, and... let's
>work on disabling ref-counting on request (at build time, of course)!
>;)

Sounds like a plan. 

I've added a bugzilla enhancement request for the builtin atomic
operations. It's libstdc++/14311.

I'll check this in to mainline, let is spin 24 hrs, and then request for
it to go into gcc-3_4-branch.

best,
benjamin