nathan binkert wrote:
> Well, it's a question of which objects can have references in multiple
> threads simultaneously.  We could also be more careful to not just
> blindly use the reference counting pointer unless we really are
> hanging onto a reference.
>   
If you want a system where concurrently running partitions of SimObjects
only communicate through events and you demand that those events only
affect one target object (so it can't have any references other then the
one to the target SimObject) then this is not a problem. This is one of
the restrictions I would like in a mandatory interface between
concurrently running objects. If you want a 100% safe execution of a
certain configuration script then this restriction should be checked and
enforced for every configuration script. In other words, concurrently
running partitions of SimObjects should only communicate through such an
interface.

Ali Saidi wrote:
> At some point we're going to have to discuss what systems we're willing to
> support multi-threaded M5 on and what we're not. 
>   
I was thinking of using OpenMP to start up the threads. This is pretty
portable.

Concerning the CMPXCHG16B instruction. You should know I don't have the
time to focus on such fine grained optimizations but feel free to discus
them. At first I'm just going to lock every critical section.
_______________________________________________
m5-dev mailing list
[email protected]
http://m5sim.org/mailman/listinfo/m5-dev

Reply via email to