Hi Hien,

For example if you do something like this in your run script:

/sbin/m5 resetstats
./benchmark
/sbin/m5 dumpstats
/sbin/m5 resetstats
./benchmark
/sbin/m5 dumpstats
/sbin/m5 resetstats
./benchmark

Then you will notice that in your stats.txt, you will have 3 sets of
statistics. Like this:

---------- Begin Simulation Statistics ----------
...stats for first benchmark run....
---------- End Simulation Statistics   ----------
---------- Begin Simulation Statistics ----------
...stats for second benchmark run....
---------- End Simulation Statistics   ----------
---------- Begin Simulation Statistics ----------
...stats for third benchmark run....
---------- End Simulation Statistics   ----------

Ivan


On Mon, Jun 30, 2014 at 5:58 AM, Nguyễn Hiển <[email protected]> wrote:

> Hi Ivan,
>
> Thanks for your help,
>
> I run like your command I have got the result.
> but when I run many times, I get only a result folder (m5out).
> I want to for each time i get a result folder ( e.g: m5out_0 for the
> first, m5out_1 for the second,...) to can compare between them to detect
> the change of caches ( miss, hit,...)
>
> Can you have an idea for that ?
>
> I used also "m5 dumpstats" and "m5 resetstats" but I don't understand so
> much about them. Can you explain for me ?
>
> Best,
> Hien
>
>
> 2014-06-26 16:45 GMT+02:00 Ivan Stalev <[email protected]>:
>
> Hi Hien,
>>
>> You can run multiple commands/benchmarks using the run script (basically
>> a bash script). If you want them to run one after another, simply do this
>> in your run script:
>>
>> cd benchmark-directory
>> ./benchmark
>> ./benchmark
>> ./benchmark
>>
>> or if you want them to run simultaneously, simply:
>>
>> ./benchmark &
>> ./benchmark &
>> ./benchmark
>>
>> You can also utilize the "/sbin/m5 dumpstats" and "/sbin/m5 resetstats"
>> commands to dump and reset stats.
>>
>> Ivan
>>
>>
>> On Thu, Jun 26, 2014 at 9:52 AM, Nguyễn Hiển <[email protected]> wrote:
>>
>>> Hi Soman,
>>>
>>> Thanks for your help.
>>>
>>> Can you give an example ( e.g command line ) to run a benchmark many
>>> times in multicore or SMT mode ?
>>>
>>> I can run a bechmark on se mode and fs mode one time: build gem5 and run.
>>> e.g: build/ALPHA/gem5.opt --debug-flags=Cache --debug-file=trace.out
>>> configs/example/fs.py --script=configs/boot/hello_hien_alpha_0.rcS --caches
>>> --cpu-type=timing --cacheline_size=64 --ruby
>>>
>>> I don't want to rebuild gem5 when I run a benchmark the second time but
>>> I don know how to do that.
>>>
>>> Best,
>>> Hien.
>>>
>>>
>>> 2014-06-26 15:27 GMT+02:00 Jyothish Soman <[email protected]>:
>>>
>>> Hi Nguyen,
>>>> As far as I can understand, there should not be variations across runs
>>>> for the same executable ran in SE mode due to the lack of interference from
>>>> any other source and gem5 acts as an exact and predictable system. In
>>>> multicore or SMT mode with multiple applications interfering with each
>>>> other will give you different results in every run (different combination
>>>> of other workloads each time).
>>>>
>>>> FS is a bit more tricky, I would let the devs reply to this one.
>>>> Best,
>>>> Jyothish
>>>>
>>>>
>>>>
>>>> On 26 June 2014 13:46, Nguyễn Hiển <[email protected]> wrote:
>>>>
>>>>> Hi all,
>>>>>
>>>>> I want to run a benchmark ( e.g "hello world" ) in gem5 SE mode or FS
>>>>> mode many times to get the output and compare them between different times
>>>>> ( e.g: compare to view the change of cache hits and misses,... )
>>>>>
>>>>> I tried so much but not successed. Can you give an idea ?
>>>>> Thanks advance,
>>>>> Hien.
>>>>>
>>>>> _______________________________________________
>>>>> gem5-users mailing list
>>>>> [email protected]
>>>>> http://m5sim.org/cgi-bin/mailman/listinfo/gem5-users
>>>>>
>>>>
>>>>
>>>
>>> _______________________________________________
>>> gem5-users mailing list
>>> [email protected]
>>> http://m5sim.org/cgi-bin/mailman/listinfo/gem5-users
>>>
>>
>>
>
_______________________________________________
gem5-users mailing list
[email protected]
http://m5sim.org/cgi-bin/mailman/listinfo/gem5-users

Reply via email to