Re: [Gems-users] instruction count in different GEMS protocols?

[Konstantinos Aisopos <kaisopos@xxxxxxxxx>]

hi again,

I have a follow up question related to instructions:

When simulating with MOESI_SMP, overall CPI is ~0.5 ( i.e., (number of
cycles*number of cores)/total instructions  ) when simulating any
splash benchmark.  For istance in fft, cycle count is 24072104 and
total instructions (in all cores) are 3083116137, meaning that CPI is
24072104 / (3083116137/64cores) = 0.5. This means that cores (on
average) execute 2 instructions per cycle. Is this possible (I am not
using Opal)?

Also, has ruby visibility on the instruction count variable or is it
in the simics binary? If ruby uses the instruction variable, can
someone give me a pointer in which files I should look for this?

thanks again
-Kostas

On Sat, Mar 14, 2009 at 3:30 PM, Konstantinos Aisopos
<kaisopos@xxxxxxxxx> wrote:
> thnaks for your help Dan!
>
> On Sat, Mar 14, 2009 at 3:16 PM, Dan Gibson <degibson@xxxxxxxx> wrote:
>> As I said before, this is expected. Instruction counts are not often useful
>> measures for multiprocessor and/or full-system simulations.
>>
>> Regards,
>> Dan
>>
>> On Sat, Mar 14, 2009 at 4:09 PM, Konstantinos Aisopos <kaisopos@xxxxxxxxx>
>> wrote:
>>>
>>> Thanks for the quick respose Dan! All my simulations have the same
>>> random seed (g_RANDOM_SEED: 1). I know that I should have
>>> randomization enabled and run many simulations to get accurate
>>> results, but currently I am avoiding it for simlicity.
>>> The thing that troubles me most is that *ALL* splash specs in
>>> MOESI_SMP have 500%-800% the number of instructions in MESI_SCMP. If I
>>> would see such big differences in the instruction count but random, I
>>> wouldn't bring up this question. Any additional thought you may have
>>> please let me know. If it still seems normal to you also let me know
>>> :)
>>>
>>> thanks for your time!
>>> -Kostas
>>>
>>> On Sat, Mar 14, 2009 at 12:36 PM, Dan Gibson <degibson@xxxxxxxx> wrote:
>>> > You should expect instruction count to vary widely even with the same
>>> > protocol for different RANDOM_SEED, because of spinloops.
>>> >
>>> > Regards,
>>> > Dan
>>> >
>>> > On Sat, Mar 14, 2009 at 11:44 AM, Konstantinos Aisopos
>>> > <kaisopos@xxxxxxxxx>
>>> > wrote:
>>> >>
>>> >> hi all,
>>> >>
>>> >> I am running splash specs in MESI_SCMP and MOESI_SMP and observing the
>>> >> following: Ruby_cycles are in the same order of magnitute (+- 20%).
>>> >> However, instructions in MOESI_SMP = (~800%) instructions in
>>> >> MESI_SCMP.  How can the same specs with the same inputs generate 8x
>>> >> more instructions in a different protocol? (roughly the same
>>> >> proportion for all splash specs). Is the variable
>>> >> *instruction_executed* trustworthy? Should I use another variable to
>>> >> measure instruction count?
>>> >>
>>> >> thanks,
>>> >> Kostas
>>> >> _______________________________________________
>>> >> Gems-users mailing list
>>> >> Gems-users@xxxxxxxxxxx
>>> >> https://lists.cs.wisc.edu/mailman/listinfo/gems-users
>>> >> Use Google to search the GEMS Users mailing list by adding
>>> >> "site:https://lists.cs.wisc.edu/archive/gems-users/"; to your search.
>>> >>
>>> >
>>> >
>>> >
>>> > --
>>> > http://www.cs.wisc.edu/~gibson [esc]:wq!
>>> >
>>> > _______________________________________________
>>> > Gems-users mailing list
>>> > Gems-users@xxxxxxxxxxx
>>> > https://lists.cs.wisc.edu/mailman/listinfo/gems-users
>>> > Use Google to search the GEMS Users mailing list by adding
>>> > "site:https://lists.cs.wisc.edu/archive/gems-users/"; to your search.
>>> >
>>> >
>>> >
>>> _______________________________________________
>>> Gems-users mailing list
>>> Gems-users@xxxxxxxxxxx
>>> https://lists.cs.wisc.edu/mailman/listinfo/gems-users
>>> Use Google to search the GEMS Users mailing list by adding
>>> "site:https://lists.cs.wisc.edu/archive/gems-users/"; to your search.
>>>
>>
>>
>>
>> --
>> http://www.cs.wisc.edu/~gibson [esc]:wq!
>>
>> _______________________________________________
>> Gems-users mailing list
>> Gems-users@xxxxxxxxxxx
>> https://lists.cs.wisc.edu/mailman/listinfo/gems-users
>> Use Google to search the GEMS Users mailing list by adding
>> "site:https://lists.cs.wisc.edu/archive/gems-users/"; to your search.
>>
>>
>>
>