Hi all,

I think the m5op dumpresetstats [delay [period]] should be transparent to
the GEM5 multi-core simulation. For example, setting  dumpresetstats with
different period in the .rcS runscript will only impact the output
stats.txt, but the cycle-by-cycle system state transition shouldn't be
affected, and therefore the CPU runtime shouldn't be changed.

However, when I run the bodytrack application from Parsec benchmark on a
4-core alpha in FS, I found the time spent in ROI with 1us dumping period
and 100us dumping period are 0.143s and 0.105s, respectively. Also, I found
the workload distribution among 4 cores is different for the two cases.

Any one has any idea about the reason behind? Thanks a lot!
gem5-users mailing list

Reply via email to