If you want a Multiple-CMP with a shared cache distributed with the
cores, then use the MOESI_CMP_directory protocol. You will want to
specify a network topology so the banks are distributed with the cores.
See $GEMS/ruby/network/simple/Network_Files/TILED_* for an example of this.
You can partition the address space by changing the mapping function
from L1 requests to L2 cache banks.
--Mike
Niket Agarwal wrote:
Hey Mike.
Thanks for the info.
Please address a few issues:
1. In the Multi-CMP design, if I use a SMP design with 1 processor/chip,
and instantiate many chips, it would be a proxy for a many core CMP.
But, , the L2 cache will become private then. What I really want is that
the L2 cache be shared and distributed with the cores.
2. The distributed L2 issue is addressed in the NUCA design with using
multiple banks and partitioning the address space. But, will this work
for 32+ processors ?
Thanks for the help,
Niket.
_______________________________________________
Gems-users mailing list
Gems-users@xxxxxxxxxxx
https://lists.cs.wisc.edu/mailman/listinfo/gems-users
Use Google to search the GEMS Users mailing list by adding "site:https://lists.cs.wisc.edu/archive/gems-users/" to your search.
|