Re: [Gems-users] counting number of hits and misses

[Nauman Rafique <nrafique@xxxxxxxxxx>]

I thought the per processor L2 misses are already printed, at least for the 
protocols that I am using.
"misses_per_thousand_instructions: " has a total number, as well as, per 
processor number.

Here is the part from MSI_MOSI_CMP_directory-L2cache.sm that is profiling.
It passes processor number as the node id.
------------------
      profile_L2Cache_miss(convertToGenericType(in_msg.Type), 
in_msg.AccessMode, MessageSizeTypeToInt(in_msg.MessageSize), i\
n_msg.Prefetch, L1CacheMachIDToProcessorNum(in_msg.RequestorMachId));
------------------

Please correct me if I am missing something.

Thanks.
-- 
Nauman


Quoting "Min Xu (Hsu)" <xu@xxxxxxxxxxx>:

> Yes, Mike is right, thanks for correcting me. :-)
> 
> On Sun, 12 Mar 2006 Mike Marty wrote :
> > > I think it is in the CacheMemory.h, but not 100%
> > > Try search the string that is printed when you call ruby0.dumpstats
> > > (or it could be ruby0.dump_stats).
> > >
> > 
> > To profile L2 misses in a CMP protocol, the Slicc controller makes calls
> > to the profiler (ruby/profiler/Profiler.C).
> > 
> > For example, in the MSI_MOSI_CMP_directory-L2cache.sm file, the
> > uu_profileMiss action is called on every L2 miss.  This in turn will
> > invoke an indirection through
> > ruby/slicc_interface/RubySlicc_Profiler_interface.C to call the
> > Profiler::addSecondaryStatSample() function.  Here you can add the
> > requestor ID.
> > 
> > I don't believe the other CMP protocols currently call these functions on
> > a miss, but it would be straightforward to add (and should have been there
> > in the first place if missing).
> > 
> > --Mike
> > 
> > _______________________________________________
> > Gems-users mailing list
> > Gems-users@xxxxxxxxxxx
> > https://lists.cs.wisc.edu/mailman/listinfo/gems-users
> _______________________________________________
> Gems-users mailing list
> Gems-users@xxxxxxxxxxx
> https://lists.cs.wisc.edu/mailman/listinfo/gems-users
>