I appreciate your detailed reply. I am only a beginner in multi-core
programming and am only learning about coherence. By reading different
replies on the list, I am trying to understand the code.

I have one question also. I checked out marss.dramsim recently. The Marss
webpage has a page on YAML, which mentions about the default.conf file and
config folder. However, I could not find these in the downloaded code.
Please let me know, if I need to checkout something else also.

Thanks and Regards
Sparsh Mittal




On Tue, Aug 23, 2011 at 12:05 AM, DRAM Ninjas <[email protected]> wrote:

> So are you asking whether the L2 should be WT or MESI?
>
> On Mon, Aug 22, 2011 at 5:22 PM, sparsh1 mittal1 
> <[email protected]>wrote:
>
>> Hello
>>
>> Thanks for timely response.
>>
>> I saw the code in memoryHierarchy.cpp:
>> void MemoryHierarchy::shared_L2_configuration()
>> {
>>         memdebug("Setting up shared L2 Configuration\n");
>> .....
>>         //using namespace Memory::SimpleWTCache;
>>         using namespace Memory::MESICache;
>> .....
>> }
>>
>
> This must be new code, I thought previously it was a SimpleWTCache ....
>
>
>> Also, one user posted this.
>>
>> *well, i think the main reason that the last level (shared) cache is a wt
>> cache is because you don't need to enforce coherence on a shared cache. the
>> bus that connects the private caches to the last level shared cache provides
>> an ordering on requests.
>>
>> i can't really speak to the inclusivity issues, but i think that's the
>> rationale for why the llc is a wt cache instead of a mesi cache .*..
>>
>
> This was me. Let me try to clarify -- it's not so much that you don't have
> to enforce coherence, but it's more that the bus provides an ordering on all
> requests headed into the shared L2 cache, so a writethrough or a writeback
> cache at the last level should be sufficient to maintain coherence (because
> once you have an ordering, all processors will see the same value in L2 and
> so coherence is maintained). Now, granted, I'm not a cache expert so I could
> be completely off base here.
>
> Just a related note -- I believe this configuration mechanism (and the
> shared_L2_configuration() function are all going away soon because of
> Avadh's new configuration mechanism and decoupled cache/coherence code), so
> it might be worthwhile to see what the code in the core-models branch looks
> like.
>
>
>>
>> This code and the  answer seem to be contradictory or something else I
>> don't know of.  Please clarify whether L2 should be WT cache or MESI cache.
>> Thanks and Regards
>> Sparsh Mittal
>>
>>
>>
>> _______________________________________________
>> http://www.marss86.org
>> Marss86-Devel mailing list
>> [email protected]
>> https://www.cs.binghamton.edu/mailman/listinfo/marss86-devel
>>
>>
>
_______________________________________________
http://www.marss86.org
Marss86-Devel mailing list
[email protected]
https://www.cs.binghamton.edu/mailman/listinfo/marss86-devel

Reply via email to