Gathering Statistics from Simulations

[Pavan Chippigiri Acharya]

Hello,

I'm trying to collect statistics of execution on the simulator. I'm looking to plot data like Instructions per core vs time, Cache miss rate vs time. Is it possible to collect such data?

If so how would I go about doing it.

Right now, I'm running the fft.c in the /test/fft/ directory with a custom config file.

The sim.out, sim.info and sim.cfg has some information but it is not detailed enough. 

Can I run one of the scripts from the /tools directory to get the data i'm looking for?

Thanks,

Pavan AC

[Wim Heirman]

Hi Pavan,

You can use the periodic-stats.py script to dump all statistics through time, then process the snapshots using dumpstats.py --partial or a custom python script to extract the values. A less heavyweight approach that doesn't dump all the stats but just those that you need can be built using the ipctrace.py and stattrace.py scripts. All of these will need to run during the simulation (using run-sniper -s).

Regards,

Wim

--
--
--
You received this message because you are subscribed to the Google
Groups "Sniper simulator" group.
To post to this group, send email to snip...@googlegroups.com
To unsubscribe from this group, send email to
snipersim+...@googlegroups.com
For more options, visit this group at
http://groups.google.com/group/snipersim?hl=en

---
You received this message because you are subscribed to the Google Groups "Sniper simulator" group.
To unsubscribe from this group and stop receiving emails from it, send an email to snipersim+...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[Pavan Chippigiri Acharya]

Hello Wim,

I forgot to mention that I am trying to plot the number of instructions per core with respect to time. 

Sincerely,

Pavan AC

On Sun, Mar 15, 2015 at 4:40 PM, Pavan Chippigiri Acharya <pavan....@nyu.edu> wrote:

Hello Wim,

Thank you for the response. 

I would like to confirm this if the following is what I ought to do.

I have changed the make file in the /test/fft/ directory to the following

run$(TARGET):

        ../../run-sniper -n -2 -c pavan -s periodic-stats.py: --roi --viz -- ./fft -p 4

Here pavan is a custom configuration file.

I have question regarding some configuration options.

the section [core/hook_periodic_ins]

ins_per_core =

ins_global = 

Does the above configuration options have any bearing on the information being gathered with the script periodic-stats.py or periodicins-stats.py?

I could not locate iptrace.py and the stattrace.py in the /scripts directory. Where is it located? 

I used the command

 ../../tools/dumpstats.py -tt core.instructions > output.txt

The data i required is captured every 1 second. I want to change the rate at which the data is captured.

Do I use the following?

-s periodic-stats.py:"10:1000" where 10ns is the periodic time and 1000 is the snapshot? or do I change the configuration file?

Sincerely,

Pavan AC

[Pavan Chippigiri Acharya]

Hello Wim,

Thank you for the response. 

I would like to confirm this if the following is what I ought to do.

I have changed the make file in the /test/fft/ directory to the following

run$(TARGET):

        ../../run-sniper -n -2 -c pavan -s periodic-stats.py: --roi --viz -- ./fft -p 4

Here pavan is a custom configuration file.

I have question regarding some configuration options.

the section [core/hook_periodic_ins]

ins_per_core =

ins_global = 

Does the above configuration options have any bearing on the information being gathered with the script periodic-stats.py or periodicins-stats.py?

I could not locate iptrace.py and the stattrace.py in the /scripts directory. Where is it located? 

I used the command

 ../../tools/dumpstats.py -tt core.instructions > output.txt

The data i required is captured every 1 second. I want to change the rate at which the data is captured.

Do I use the following?

-s periodic-stats.py:"10:1000" where 10ns is the periodic time and 1000 is the snapshot? or do I change the configuration file?

Sincerely,

Pavan AC

On Sat, Mar 14, 2015 at 7:44 AM, Wim Heirman <w...@heirman.net> wrote: