On 07/26/11 02:12, Paolo Grani wrote: > Hi all, > > i need some clarification about the ruby statistics. > I read in a previous post that the miss counts for the ruby cache only > works with the MOESI_hammer coherence protocol. > I would like to know if there is some patch to get the > MOESI_CMP_directory protocol working with this information. > In fact when a try to run a simulation whit this protocol i get this > output for any CacheMemory in Ruby: > > "Cache Stats: system.l1_cntrl0.L1DcacheMemory > system.l1_cntrl0.L1DcacheMemory_total_misses: 0 > system.l1_cntrl0.L1DcacheMemory_total_demand_misses: 0 > system.l1_cntrl0.L1DcacheMemory_total_prefetches: 0 > system.l1_cntrl0.L1DcacheMemory_total_sw_prefetches: 0 > system.l1_cntrl0.L1DcacheMemory_total_hw_prefetches: 0" > > > In addition i'm also interested in latency informations on memory > accesses. > Running a simulation always with MOESI_CMP_directory i get this: > > "miss_latency: [binsize: 8 max: 825 count: 19954053 average: 3.10522 | > standard deviation: 3.91114 | 19923362 3729 6244 6396 3738 1081 119 79 > 121 88 87 31 31 27 34 30 36 31 35 371 2435 2496 957 663 874 529 110 29 > 27 19 21 17 13 19 16 12 15 16 11 7 9 4 5 5 7 3 9 7 1 5 2 3 0 4 2 3 3 0 > 1 1 1 1 3 1 1 1 0 1 0 0 0 1 2 0 1 0 2 1 2 0 1 1 1 0 0 0 0 0 0 0 0 1 0 > 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 > 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 > 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 > 0 0 ]" > > For what i can understand this information says "how many requests > took 1 cycle, 2 cycles, 3 cycles... to be accomplished" also reporting > some aggregate information (count,average....). > > But the the coherence protocol models a 2-level cache architecture so > i would like to know if this informations are for the L2 shared cache > or for the L1 level. > If the answer is L1 level this information is an aggregate of all the > private L1 caches (i'm running a 8 cores simulation so there are 8 > private L1 caches) or is specific for some L1 cache? > > > Finally i found a patch that tries to convert the Ruby's stats to the > M5-style Stat class. > I can correclty compile the code but when i try to run it a memory > leak occurs. > Someboy have tried to fix it? > > Thank's in advance, > > Paolo > >
This should really be on gem5-users. Gabe _______________________________________________ gem5-dev mailing list [email protected] http://m5sim.org/mailman/listinfo/gem5-dev
