If you look through src/mem/cache/cache_impl.hh, you'll see that all of the places that once had pkt->req->threadId() as an index for per-thread stats are hardwired to 0 now. To be honest I don't remember the reason for that (it was almost three years ago that that happened, according to mercurial), but it probably had something to do with dealing with requests that don't have a thread ID.
If you really need per-thread cache stats, you can try changing all those indexes back and see if that works. If someone else has already done this, please speak up. If someone gets it to work robustly, it would be great to fix that in the repository. Steve On Mon, Aug 10, 2009 at 4:21 AM, Ashutosh Jain<[email protected]> wrote: > Hi > I am running multi-core multi-thread (2 core X 4 threads per core) in SE > mode. I do get results. > Most of the results like > system.cpu0.cpi_0 2.683886 > # CPI: Cycles Per Instruction > system.cpu0.cpi_1 2.509380 > # CPI: Cycles Per Instruction > system.cpu0.cpi_2 5.760678 > # CPI: Cycles Per Instruction > system.cpu0.cpi_3 4.374111 > # CPI: Cycles Per Instruction > system.cpu0.cpi_total 0.852292 > # CPI: Total CPI of All Threads > depicts the usage of multi threads > But some results like > system.cpu0.dcache.overall_miss_rate <err: div-0> > # miss rate for overall accesses > system.cpu0.dcache.overall_miss_rate_0 0.084201 > # miss rate for overall accesses > system.cpu0.dcache.overall_miss_rate_1 <err: div-0> > # miss rate for overall accesses > system.cpu0.dcache.overall_miss_rate_2 <err: div-0> > # miss rate for overall accesses > system.cpu0.dcache.overall_miss_rate_3 <err: div-0> > # miss rate for overall accesses > system.l2a.overall_miss_rate <err: div-0> > # miss rate for overall accesses > system.l2a.overall_miss_rate_0 0.652409 > # miss rate for overall accesses > system.l2a.overall_miss_rate_1 <err: div-0> > # miss rate for overall accesses > system.l2a.overall_miss_rate_2 <err: div-0> > # miss rate for overall accesses > system.l2a.overall_miss_rate_3 <err: div-0> > # miss rate for overall accesses > do not show the correct usage of threads. > > I am not understanding why only one overall_miss_rate is having the value, > rest doesn't. > > Thanks in advance > Ashutosh Jain > > _______________________________________________ > m5-users mailing list > [email protected] > http://m5sim.org/cgi-bin/mailman/listinfo/m5-users > _______________________________________________ m5-users mailing list [email protected] http://m5sim.org/cgi-bin/mailman/listinfo/m5-users
