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

Reply via email to