Hello, everyone Recently I now using M5 and Parsec to do some experiments about caches. But running full Parsec benchmark with simmedium and simlarge in detailed O3 CPU mode takes too too long time. In my experience, X264, facesim, and freqmine will be more than 1 week to simulate. So I have to limit the maximum instructions when M5 restore from ROI checkpoint. I did some experiments with M5 cmd, but it doesn't work. Here are the cmd line
1.first create ROI checkpoint ./build/ALPHA_FS/m5.opt ./configs/example/fs.py -n 4 --script=./parsec_script/bodytrack/bodytrack_4c_simmedium_ckpts.rcS 2.restore from ROI checkpoint, run extra 1 billion instructions for ROI part using --max-inst option. ./build/ALPHA_FS/m5.opt ./configs/example/fs.py -n 4 -d --caches --l2cache --max-inst=1000000000 --checkpoint-restore=1 The simulation successfully running with detailed mode, but it does not stop after arrive at the 1 billion instruction, here are some results shown in m5out/stats.txt. It seems the M5 already exceeded 1 billion instruction limitation (In this results, m5 already run more than 3.8 billion instructions). So my question, how could I control max instruction for PARSEC after its restoring from ROI checkpoint? ---------- Begin Simulation Statistics ---------- host_inst_rate 142861 # Simulator instruction rate (inst/s) host_mem_usage 701880 # Number of bytes of host memory used host_seconds 22973.84 # Real time elapsed on the host host_tick_rate 18711184 # Simulator tick rate (ticks/s) sim_freq 1000000000000 # Frequency of simulated ticks sim_insts 3282054759 # Number of instructions simulated sim_seconds 0.429868 # Number of seconds simulated sim_ticks 429867680500 # Number of ticks simulated Many thinks in advance. David
_______________________________________________ m5-users mailing list [email protected] http://m5sim.org/cgi-bin/mailman/listinfo/m5-users
