This is the whole point of full-system simulation instead of just feeding
traces to a memory simulator. There is feedback between the memory system
and the execution of a program. Locks may be acquired differently,
processors may spin longer, etc, etc, etc
--Mike
> I would like to ask whether the access requests from Simics can vary
> among
> different protocols. At first sight this does not seem possible to me.
> But maybe I'm missign something. Here are some details of my
> simulations:
>
> 0) I'm doing two simulations to compare two different memory
> architecture
> In one simulation I'm using MOESI_SMP_directory and in the other I'm
> using MSI_MOSI_CMP_directory. The architecture using
> MOESI_SMP_directory
> is meant to be a CMP with private L2 caches, while the other is used
> for a shared L2 cached CMP. I'm using a system with 4 processors.
> 1) The simulations are started from the same checkpoint.
> 2) The same programs are run from the simulated machine console.
> 3) When I print the memory address requests sent from Simics
> (I do the printf's in the makeRequest function at SimicsProcessor.C
> file),
> I see that for the same processor the address requests may differ.
> Generally the address request match for a good bunch of addresses but
> later a stream of different addresses are seen at the same processor
> for different simulations. After that stream, again the addresses
> match
> for a while.
> Also for some processors, I have even seen that in one simulation the
> processor is asked some addresses while in the other simulation these
> addresses were not present at all in the processor request trace.
>
> What am I missing? Why are the traces show differences? (I do not think
> different protocols create different exceptions such that the operating
> system reacts differently because exceptions related to memory system
> are
> raised in the MMU as far as I know,so I was confused).
>
> Any help will really be appreciated.
>
> Best Regards,
>
>
> Derin Harmanci.
>
>
>
> _______________________________________________
> 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.
|