On Wed, 9 Jan 2013, Andreas Hansson wrote:
On Jan. 8, 2013, 7:01 p.m., Nilay Vaish wrote:
I am not opposed to committing this patch. But I am not convinced
if all the controllers should be under the object ruby_system.
If we ever move to a system in which different controllers have
different frequencies of operation, it seems that all the cache
controllers private to a core would be more tightly bound to
core, than being bound to memory system.
I support that observation, and it would be good to have a separation of L1
controllers, L2 controllers, interconnect and memory in terms of clock domains.
That said, I think the current change is a good step in not using
system_ptr->getClock() at any point.
Could we perhaps start with the current patch, remove the getClock()
from the RubySystem all together, and then follow up by shifting the
bits and pieces around until they are all in the right clock domain?
As I said, I not opposed to patch. So, we can commit it to the repo.
--
Nilay
_______________________________________________
gem5-dev mailing list
[email protected]
http://m5sim.org/mailman/listinfo/gem5-dev
