[Patch] MIPS: Emit optimized sync instructions for Octeon CPUs.

[Richard Henderson]

On 08/17/2009 11:22 AM, Paolo Bonzini wrote:
> On 08/17/2009 06:51 PM, David Daney wrote:
>>
>> What do you think about adding an additional optional parameter to
>> __sync_synchronize that would weaken it to just a write barrier?
>
> I'd rather have __sync_rmb and __sync_wmb (or maybe
> __sync_acquire_synchronize and __sync_release_synchronize?).

I'd like to have RMB and WMB, since these get used (via asm) in
libgomp and such.  However, instead of extending the __sync functions,
perhaps we should work more toward the c++0x atomics thing that
was discussed here recently.


r~