On Wed, Feb 23, 2011 at 7:40 AM, Timothy Hayes <[email protected]> wrote:
> Hello, > > I have some general questions regarding MARSS. > > I've been trying to find at the number of loads that failed to issue > due to bank conflicts. Going through the code, I can see that PTLsim's > memory implementation has been modified somewhat but some of the > functions/statistics still remain in MARSS. For example, in > ReorderBufferEntry:issueload(), 'dcache.load.issue.miss' is > incremented whether or not there is a l1d hit or not. I'm curious as > to which statistics are deprecated/unreliable in MARSS' current > implementation. Is all of the /ooocore/dcache unreliable? > > Yes. When we code new memory model, we didn't remove the old statistics, so they are deprecated and should not be used. In future 0.2 release we have a new statistics model, which is more flexible and more user friendly, and we will remove these deprecated stats in that release. Just as an aside, does the bank conflict code represent any real x86 > implementation? I see that it compares the least significant number of > bits of two addresses which issue on the same cycle; this would > suggest that a cache line is divided and distributed amongst the > banks. I was under the impression that most banked cache > implementations will contain adjacent lines in different banks; this > way the tag structure doesn't need to be multiported. It's simple to > change the code to reflect this, I was just curious if the code in the > current incarnation is reflective of something specific. :-) > > This implementation is not accurate. This code was aimed to get a working simulator that can be improved over the time to be more accurate to current x86 systems. I would really appreciate if you can change the code to be more like real systems. Let me know if you need any help with this code. Thanks, Avadh Kind regards > Tim > > _______________________________________________ > 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
