[rescue] Screw (software) emulators!

[rescue at sunhelp.org]

On Mon, 6 Aug 2001, joshua d boyd wrote:

> On Mon, Aug 06, 2001 at 06:20:08PM -0600, james at foonly.com wrote:
> > Since the RAM can be clocked much higher than 8MHz, a shared-memory SMP
> > PDP-8 could be built fairly easily by clocking the RAM accesses faster and
> > interleaving requests from each CPU.
>
> How would an atomic memory test and check be implemented if one were to
> persue such a chore?

You don't need one.  It makes semaphore synchronization simpler and
faster but it's hardly necessary.  Without T/S you have to do 2-stage
commits.

On the other hand, if you're reimplementing the CPU from scratch a T/S
instruction could certainly be added.

-James