Because the L1 cache is uniform, but L2 is non-uniform. I want to create a
new function to calculate hitLatency in the cache code, and hope the
function code only affect L2 cache, rather than affect both caches.In this
case, I want to add a flag in the function like "uniformFlag". Then I set
the value of flag in python configuration file. For example, in my fs.py, I
set:
test_sys.cpu.addPrivateSplitL1Caches(L1Cache(size = '32kB', uniformFlag =
1), L2Cache(size = '64kB', uniformFlag = 1))
test_sys.l2=L2Cache(size = '2MB', uniformFlag = 0)
In the c code, I use a new function CalhitLatency (PacketPtr pkt, int
uniformFlag) to replace all "hitLatency" variables.
int CalhitLatency (PachetPtr pkt, int uniformFlag)
{
int lat;
if (uniformFlag) { lat = hitLatency; }
else {
addr = pkt->getAddr();
/* here add code to assign lat depending on the address */
}
return lat;
}
Is it possible to get what I want in this case?
BTW, When we edit the c code, we need to recompile m5 before running it. Is
it right?
Thanks
Gary
On Tue, Jul 15, 2008 at 12:46 PM, nathan binkert <[EMAIL PROTECTED]> wrote:
> You should look into the cache access code. Actually, you should in
> general look for any case where you see hitLatency in the cache code.
> You'll want to change that to some sort of function.
>
> Nate
>
> On Tue, Jul 15, 2008 at 9:26 AM, Gary Chai <[EMAIL PROTECTED]> wrote:
> > Thank you for your reply.
> > It is like nonuniform cache. When the L2 cache receives a different
> address
> > request, It assigns a different latency. Which codes describe such kind
> of
> > process?
> > Thanks
> > Gary
> >
> > On Mon, Jul 14, 2008 at 3:28 PM, nathan binkert <[EMAIL PROTECTED]>
> wrote:
> >>
> >> > If I want to let the latency of accessing different parts in L2 cache
> is
> >> > different, so the latency is not uniform. Is it possible in m5? how
> can
> >> > I
> >> > get it and where can I edit the code?
> >> > Any suggestion is really appreciated.
> >>
> >> I.e. you want to implement NUCA of some sort? I don't anyone has
> >> modeled this yet, so you'd have to hack on the cache model a bit. I
> >> honestly don't think this should be all that difficult to do depending
> >> on how complicated your NUCA model is.
> >>
> >> Anyone like to correct me?
> >>
> >> Nate
> >> _______________________________________________
> >> m5-users mailing list
> >> [email protected]
> >> http://m5sim.org/cgi-bin/mailman/listinfo/m5-users
> >
> >
> > _______________________________________________
> > m5-users mailing list
> > [email protected]
> > http://m5sim.org/cgi-bin/mailman/listinfo/m5-users
> >
> _______________________________________________
> m5-users mailing list
> [email protected]
> http://m5sim.org/cgi-bin/mailman/listinfo/m5-users
>
_______________________________________________
m5-users mailing list
[email protected]
http://m5sim.org/cgi-bin/mailman/listinfo/m5-users
