ok Gabe,
i'm sorry i'm switching mailing list.
Paolo
Il 26/07/2011 11:48, Gabe Black ha scritto:
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
--
====================================================
The devil is in the details
====================================================
Ing. Paolo Grani
Dipartimento di Ingegneria dell'Informazione
via Roma 56, 53100 Siena, Italia
Laboratorio di Architettura dei Calcolatori ---> int. 223
mailto:[email protected]
http://www.dii.unisi.it
====================================================
_______________________________________________
gem5-dev mailing list
[email protected]
http://m5sim.org/mailman/listinfo/gem5-dev
