PATCH: v2 of Joel's idea to reenabled non-generic on i386

[Martin v. Löwis]

Loren James Rittle <rittle@latour.rsch.comm.mot.com> writes:

> 3. Reassess the i486 special case.  I wonder if breaking the effective
> ABI between i386 and i486 is a good idea based on all the people
> coming out of the wood work at the last minute on this issue. ;-)

I'm not sure you can back out of this. gcc 3.2 is in use already,
and people may expect compatibility of gcc 3.3 with gcc 3.2.

[Of course, if this compatibility is not there anyway (I don't know),
you may consider providing a uniform ABI for all of i386+, perhaps
after timing the various approaches to understand the penalty of a
spinlock]

> +   do {
> +     __asm__ __volatile__ ("xchgl %0,%1"
> + 			  : "+m" (__Atomicity_lock<0>::_S_atomicity_lock),
> + 			    "+r" (__tmp));
> +   } while (__tmp);

I think this is correct. You may want to make use of the
[symbolic_name] feature, just to educate users it is there.

Regards,
Martin