I put together some preliminary DVFS support for MARSS awhile ago: http://cs.binghamton.edu/~tstache1/marss.nobackup/marss_dvfs.tgz
It'll include the number of cycles spent in each ACPI P-state in the stats. Tyler > I believe it's fairly difficult to distinguish what it means to be > "active". I don't believe marss models dvfs, so there's always something > using a core -- even if it is the idle thread that is doing nothing. > > > On Wed, Apr 30, 2014 at 4:53 PM, alireza nazari > <[email protected]>wrote: > >> Hello, >> >> I am confused about some stats. I want to know how many cycles a >> benchmark >> takes to execute in parallel. Does " ooo_x_x : cycle:xxxxxxxx " show all >> the cycles that a specific core is active or it just counts all the sim >> cycles regardless of if the core is used or not? I assume that when a >> part >> of application is sequential, one core is used and the rest are idle, >> this >> means one of the cores should have a larger cycle than the rest. I do >> not >> see that in the benchmark that I am running now. Am I wrong? how can I >> find >> the number of cycles a core was active and for how many cycles it was >> idle(not waiting for dependency but actual idle because of no >> instruction >> to execute) >> >> Thank you >> Alireza >> >> >> >> _______________________________________________ >> http://www.marss86.org >> Marss86-Devel mailing list >> [email protected] >> https://www.cs.binghamton.edu/mailman/listinfo/marss86-devel >> >> > _______________________________________________ > http://www.marss86.org > Marss86-Devel mailing list > [email protected] > https://www.cs.binghamton.edu/mailman/listinfo/marss86-devel > _______________________________________________ http://www.marss86.org Marss86-Devel mailing list [email protected] https://www.cs.binghamton.edu/mailman/listinfo/marss86-devel
