Thank you very much for your insight.
Berkin
-----Original Message-----
From: gems-users-bounces@xxxxxxxxxxx [mailto:gems-users-bounces@xxxxxxxxxxx]
On Behalf Of Philip Garcia
Sent: 2009-03-26 19:01
To: Gems Users
Subject: Re: [Gems-users] Opal simulation question
I don't think GEMS 2.0 directly supports doing simulations for a set
number of cycles (although I could be wrong). I just modified opal to
count cycles instead of instructions committed on CPU #0. This can be
done by changing a single line in opal/system/system.C.
In the simulate() method change the while (m_sim_status==SIMSTATUS_OK
&& instrCount>=m_seq[0]->m_stat_committed[0]) loop to be just: while
(m_sim_status==SIMSTATUS_OK && instrCount>=m_global_cycles)
This will just use the instrCount variable to be compared to global
cycles. Personally I prefer using this metric, as the number of
instructions executed on CPU #0 is kind of an arbitrary length of time
(particularly when comparing different benchmarks), whereas a set
number of cycles has at least a baseline involved. It's difficult to
say though which one is definitively a "better" measurement, I'm sure
some people here have a better background in this to make a claim.
Personally, I prefer the set # of cycles measurement, and especially
can't really make sense of what a set number of instructions executed
on cpu #0 means when running multiprocessor workloads.
Of course, when giving number of cache misses, and some of those
statistics it might make more sense to first normalize them, because
giving the number of cache misses for 2 different setups that compute
a different amount of work isn't very meaningful. I think a standard
metric there is miss rate (percent of load/store instructions that
cause a cache miss) or use something like misses per thousand
instructions.
Phil
On Mar 26, 2009, at 5:06 PM, Berkin Ozisikyilmaz wrote:
> Is it possible to stop simulation with opal after a predetermined
> number of
> cycles? I know that opal.sim-step takes the number of instructions to
> simulate. Also would it be fair comparison to compare performance/
> cache
> misses etc. statistics across simulations with the same number of
> cycles?
>
> Thanks
> Berkin
>
>
> _______________________________________________
> 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.
>
_______________________________________________
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.
|