On Sat, Sep 29, 2012 at 4:18 PM, Ankita (Garg) Goel <[email protected]>wrote:

> Hi Avadh,
>
> True, the reset to 0 is happening once and that is when the context switch
> happens. Thanks!
>
> I am trying to gather per-cycle statistics to compute per-cycle power
> consumption from within the simulator (not by post-processing). So at this
> point I am computing the difference and capturing the data. Can this be
> simplified ?
>
> To get the total stats count value, sum of user and kernel phase
activities, look at the macros 'N_STATS_UPDATE' in cache folder.  They use
'kernel_stats' and 'user_stats' as parameters to these stats to get each
phase related stats counter.  You can use similar technique to calculate
total value of your counters and then find differences.

- Avadh


> Regards,
> Ankita
>
>
> On Sat, Sep 29, 2012 at 5:54 PM, avadh patel <[email protected]> wrote:
>
>>
>>
>> On Sat, Sep 29, 2012 at 10:35 AM, Ankita (Garg) Goel <
>> [email protected]> wrote:
>>
>>> Hi,
>>>
>>> I was interesting in collecting per-cycle statistics. While doing, I
>>> found that the total number of instructions being dispatched suddenly got
>>> reset to zero:
>>>
>>> W64 total_insns = 0;
>>>  for (W32 i = 0; i < OPCLASS_COUNT; i++) {
>>>                 total_insns += CORE_STATS(dispatch.opclass)[i];
>>>  }
>>>
>>> So total_insns would typically increase from 0 to a 1000+ value and then
>>> somehow in some random cycle in between, it seems to be getting a value of
>>> zero. Am I gathering the above data incorrectly ? I did look at the source
>>> for the core_stats, but could not find a place where the stats are being
>>> reset. Any thoughts ?
>>>
>>> If these values starts from 0 only once it must be change of phase from
>> user level to kernel or vise versa.  Because we store separate stats for
>> user level and kernel level, when core switches to other level, it uses
>> separate counter storage.
>>
>> what is your aim to collect the total_insns count.  If you are looking
>> into periodic counter value then checkout time-stats options which dump
>> specific stats counter's value after fix interval in to csv file.
>>
>> - Avadh
>>
>>
>>> Appreciate your help!
>>>
>>> --
>>> Regards,
>>> Ankita
>>> Graduate Student
>>> Department of Computer Science
>>> University of Texas at Austin
>>>
>>>
>>>
>>> _______________________________________________
>>> http://www.marss86.org
>>> Marss86-Devel mailing list
>>> [email protected]
>>> https://www.cs.binghamton.edu/mailman/listinfo/marss86-devel
>>>
>>>
>>
>
>
> --
> Regards,
> Ankita
>
>
>
_______________________________________________
http://www.marss86.org
Marss86-Devel mailing list
[email protected]
https://www.cs.binghamton.edu/mailman/listinfo/marss86-devel

Reply via email to