Re: [Gems-users] MOESI_SMP_hammer question

[Mike Marty <mikem@xxxxxxxxxxx>]

Mike Marty wrote:

> We added an optimization to the AMD Hammer protocol so that data is not 
> returned twice on any sharing miss. This requires meta-state to the 
> memory controller. It isn't really a directory.
>  
If I wasn't clear, the memory controller in the AMD Hammer protocol 
_always_ responds with Data even if the data is stale.  If the data is 
dirty in a cache, then the requestor will eventually get the dirty data 
and throw away the stale data.  Thus a 2nd data response only happens 
for dirty sharing misses.
You should be able to easily modify the protocol to match what is 
actually done.