Yes, I will study it more. Probably, I need to hack network implementation as 
your suggestions.
Thanks again for your advice.


From: gem5-users-boun...@gem5.org [mailto:gem5-users-boun...@gem5.org] On 
Behalf Of Tushar Krishna
Sent: Wednesday, September 12, 2012 11:30 AM
To: gem5 users mailing list
Subject: Re: [gem5-users] How to specify the latency between L1 and L2 cache

Ah ...
Hmm yeah you'll have to see if it is easier to have 2 separate files then, or 
just have one file with extra states.



On Sep 11, 2012, at 10:58 PM, GE ZHIGUO wrote:


In exclusive cache, each address has a unique cache line, so  a state of the 
state machine only need to
 Memorize the state information for one cache line. While for inclusive cache, 
each address may have two
Cache line(L1 and L2). As a consequence, a state in the state machine need to 
contain the state information
of L1 and L2, and then too many states will be needed due to combination of 
states in L1 and L2.


From: gem5-users-boun...@gem5.org<mailto:gem5-users-boun...@gem5.org> 
[mailto:gem5-users-boun...@gem5.org] On Behalf Of Tushar Krishna
Sent: Wednesday, September 12, 2012 10:33 AM
To: gem5 users mailing list
Subject: Re: [gem5-users] How to specify the latency between L1 and L2 cache

Hi Zhiguo,
I am not very familiar with the protocol files. I think it should be easy to 
model an inclusive L2 using one state machine. Look at how L1 and L2 
communicate currently to implement an exclusive cache. They can communicate via 
a similar manner for inclusive. Since both L1 and L2 are data structures within 
the same file, I don't think you need a "queue" to communicate between them in 
the code.

- Tushar


On Sep 11, 2012, at 10:28 PM, GE ZHIGUO wrote:



Hi, Tushar
      I also thought of using one sm file for L1 and L2 caches like 
MOESI_hammer.sm.
I have two questions.
1. Hammer is exclusive cache which behave like one level cache. And there is 
only one state machine controlling both L1 and L2.
2. If modeling inclusive cache L1 and L2, is it reasonable to model L1 and L2 
using one state machine? How L1 and L2 communicate, using
trigger queue?

  Thanks!

Zhiguo

From: gem5-users-boun...@gem5.org<mailto:gem5-users-boun...@gem5.org> 
[mailto:gem5-users-boun...@gem5.org] On Behalf Of Tushar Krishna
Sent: Wednesday, September 12, 2012 9:52 AM
To: gem5 users mailing list
Subject: Re: [gem5-users] How to specify the latency between L1 and L2 cache

Hi Zhiguo,
To model a Private L2, you need to change the protocol files, not the network. 
MOESI_hammer is an example of a Private L2 protocol in gem5. Here, there is 
only one protocol_cache.sm file instead of separate files protocol_L1cache.sm 
and protocol_L2cache.sm
The cache.sm file contains both L1 and L2, modeling a fixed delay between them. 
Each "cache" is then connected to other "caches" via the network.

cheers,
Tushar


On Sep 11, 2012, at 9:13 PM, GE ZHIGUO wrote:




Hi, Tushar
      Thanks for your advice. I want to model a private L1 cache and a private 
L2 cache connected together.
I am not clear how to get a constant and controllable delay between them. 
Hacking the network
Is a way. But, it will be good to have other simpler ways to do this as private 
L2 architecture is common also.
      Regards,
Zhiguo

From: gem5-users-boun...@gem5.org<mailto:gem5-users-boun...@gem5.org> 
[mailto:gem5-users-boun...@gem5.org] On Behalf Of Tushar Krishna
Sent: Tuesday, September 11, 2012 10:31 PM
To: gem5 users mailing list
Subject: Re: [gem5-users] How to specify the latency between L1 and L2 cache

The point of the interconnect network is to model a realistic and variable 
delay between L1 and L2 (if it is a shared L2 protocol).
If you want a fixed latency between L1 and L2, play around with the topology 
and/or hack into the network to deliver packets with same delay irrespective of 
contention.

- Tushar


On Sep 11, 2012, at 1:10 AM, GE ZHIGUO wrote:





Hi, All
      The L1 and L2 caches are connected via Interconnections. I specify the 
latency of the transitions in SLICC file to 1 cycle when L1 send message to L2.
But still the latency of the message delivery can be many more than 1  cycle.  
My question is how to specify the latency between L1 and L2 cache
Connected via interconnect network.
Thanks!

_______________________________________________
gem5-users mailing list
gem5-users@gem5.org<mailto:gem5-users@gem5.org>
http://m5sim.org/cgi-bin/mailman/listinfo/gem5-users

_______________________________________________
gem5-users mailing list
gem5-users@gem5.org<mailto:gem5-users@gem5.org>
http://m5sim.org/cgi-bin/mailman/listinfo/gem5-users

_______________________________________________
gem5-users mailing list
gem5-users@gem5.org<mailto:gem5-users@gem5.org>
http://m5sim.org/cgi-bin/mailman/listinfo/gem5-users

_______________________________________________
gem5-users mailing list
gem5-users@gem5.org<mailto:gem5-users@gem5.org>
http://m5sim.org/cgi-bin/mailman/listinfo/gem5-users

_______________________________________________
gem5-users mailing list
gem5-users@gem5.org
http://m5sim.org/cgi-bin/mailman/listinfo/gem5-users

Reply via email to