Hi Armel,

The classic caches do not support strictly exclusive/inclusive.

For "non-inclusive", either one will work. It just matters if you want the
blocks inserted at the lower level when there is a miss or not. Similarly,
you need to set writeback clean to "true" for the higher level cache if you
want to implement a mostly exclusive cache.

I suggest diving into the code to understand these things better.

Jason

On Thu, Jul 11, 2019 at 4:05 AM Armel Jeatsa <[email protected]> wrote:

> Hello,
> I started designing my system but since there is only* 'mostly_incl'* and
> *'mostly_excl'* for *clusivity*, how can i implement a cache with a
> *non-inclusive* policy? And how to implement *strictly
> inclusive/exclusive *policy? Are '*mostly_incl*' and '*mostly_excl*'
> sufficient for this?
>
> kind regards,
> Armel
> --
> *JEATSA TOULEPI Armel*
> Étudiant En Génie Informatique
> Tel: +237 650771894 / 655296800
>
>
> Le mer. 10 juil. 2019 à 18:35, Armel Jeatsa <[email protected]> a
> écrit :
>
>> Thank you very much for your replies.
>> I will start using classic caches since i do not have to modify coherance
>> protocol for my work.
>>
>> kind regards,
>> Armel
>> --
>> *JEATSA TOULEPI Armel*
>> Étudiant En Génie Informatique
>> Tel: +237 650771894 / 655296800
>>
>>
>> Le mer. 10 juil. 2019 à 18:11, Jason Lowe-Power <[email protected]> a
>> écrit :
>>
>>> Hello, and welcome to the gem5 mailing list :).
>>>
>>> I suggest, at least as a first pass, using the classic caches, not Ruby.
>>> You can check out the Learning gem5 book (learning.gem5.org) to see
>>> examples of how to create different kinds of systems that don't necessarily
>>> match what's in se.py. The classic caches have an option to be "mostly
>>> inclusive" and "mostly exclusive" which might do what you need.
>>>
>>> I don't know if anyone has tried to have separate L2 instruction and
>>> data caches. However, this might "just work" with the classic caches
>>> assuming you have separate instruction and data busses (this should be
>>> possible). I suggest drawing the system you want to simulate on paper, then
>>> trying to create the configuration in gem5.
>>>
>>> If you want to investigate modifying the cache coherence protocol, or
>>> your investigation depends on high-fidelity modeling of the on chip
>>> network, then you should use Ruby. In this case, you'll probably have to
>>> write a new coherence protocol. You might be able to use other coherence
>>> protocols (e.g., MESI_Two_Level) as a starting point, or you might have to
>>> develop a new one from scratch (See Part 3 of Learning gem5 for details).
>>>
>>> Cheers,
>>> Jason
>>>
>>>
>>>
>>> On Wed, Jul 10, 2019 at 9:41 AM Armel Jeatsa <[email protected]>
>>> wrote:
>>>
>>>> Hello,
>>>> I am relatively new in gem5 simulator, i have to design a
>>>> multiprocessor architecture with two levels of caches composed of:
>>>>
>>>>    - At level 1: the usual L1i and L1d caches
>>>>    - At level 2 (shared by all processors): L2i and L2d caches, L2i
>>>>    inclusive with L1i, and L2d non-inclusive with L1d.
>>>>
>>>> What is the best solution? Use *Ruby *and* Slicc*, or just *Python*
>>>> and *C++*. How should I proceed in this case?
>>>> I ask this question because the *System* object in gem5 seems to only
>>>> accept an L2 bus.
>>>> Thank you in advance.
>>>> --
>>>> *JEATSA TOULEPI Armel*
>>>> Étudiant En Génie Informatique
>>>> Tel: +237 650771894 / 655296800
>>>> _______________________________________________
>>>> gem5-users mailing list
>>>> [email protected]
>>>> http://m5sim.org/cgi-bin/mailman/listinfo/gem5-users
>>>
>>> _______________________________________________
>>> gem5-users mailing list
>>> [email protected]
>>> http://m5sim.org/cgi-bin/mailman/listinfo/gem5-users
>>
>> _______________________________________________
> gem5-users mailing list
> [email protected]
> http://m5sim.org/cgi-bin/mailman/listinfo/gem5-users
_______________________________________________
gem5-users mailing list
[email protected]
http://m5sim.org/cgi-bin/mailman/listinfo/gem5-users

Reply via email to