Re: [Gems-users] cache misses over a period of time

["Abhishek Ranjan" <absk82@xxxxxxxxx>]

Hi,
I made some changes(simple writes to output file) in Profiler.C(in the function wakeup() where periodic dumps take place) and recompiled it but I don't see my changes in the output file. This is the first time, I was recompiling it. Is there anything which has to be removed before recompilation? I also tried removing the object as well as the source files in the generated folder but it doesn't happen. I am using the same checkpoints which I was using before recompiling. Is there something I am missing here?
thanks in advance,
Abhishek

On 9/25/06, Abhishek Ranjan <absk82@xxxxxxxxx> wrote:

Thanks Dan,

I will try the second approach, so I will get back with more questions :).

Abhishek

 

On 9/24/06, Dan Gibson <degibson@xxxxxxxx> wrote:

Abhishek,
You can get your data for the longer period relatively easily, using steps like those in the WIKI QuickStart.

As for the short periods, here is what I suggest.

Start your simulation, load (and init) ruby and do not load opal.
Use a fixed random seed for ruby (non-zero) -- work to make the cache images as constant as possible at the beginnings of the various "small periods"
Advance the simulation to the beginning of the first "small period."
ruby0.clear-stats
Load (and init) opal.
Run for the duration of first small period.
ruby0.dump-stats
Repeat for all such small stats (eg quit simics, start over)

Another option would be to modify Ruby's profiler (Profiler.[Ch]) to dump the cache miss counts versus time to a file, then use Matlab or some other tool to generate histogram-like statistics from that file. This might be a better option if the number of "short periods" is large.

Regards,
Dan Gibson

Abhishek Ranjan wrote:

Hi,
I have an application and I want to see the number of cache misses generated by it over a period of time and I want to get this data for a long period of time as well as multiple small periods to get the standard deviation. I tried using opal for this and even though opal gives the data for one particular period, it cannot give the corresponding data for multiple small periods because once opal encounters magic break point, we cannot step it as mentioned in one of the previous threads. Also, using just ruby would not be accurate as it assumes no pipeline stalls, and a simple in-order processor modelled by simics. Could anyone please give me some suggestion on what can be done for this?
Thanks a lot,
Abhishek

_______________________________________________
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.