Hey Everyone, I am running m5 in ALPHA_FS mode and I am trying to take a checkpoint after 17,175M instructions into the sjeng spec2006 benchmark. (This instruction count is a reported Simpoint for the benchmark and is thus an ideal portion of the benchmark to simulate instead of the entire run). I am properly compiling and running ALPHA_FS, and I have the binary on a mounted disk image file and am able to run the benchmark from the command line inside the simulator -- I am just unsure how best to go about taking the checkpoint.
I know the 'm5 benchmark [delay [period]]' command takes "ticks" as its arguments. Is there a better way to go about taking a checkpoint after 17,175M instructions than guessing how many simulator ticks it will take to execute this many instructions? I am running this all on a single TimingSimpleCPU with caches. Some digging has shown that the stats "system.cpu.num_insts" statistic does not differentiate between kernel instructions and user instructions. And also that the num_insts statistic does not necessarily correlate as 1 cycle = 1 instruction because TimingSimpleCPU can stall. It would seem that m5 would be able to differentiate between kernel instructions and user instructions. Is there a smarter way to set up my checkpoint 17,175 M instructions into the spec program? Or should I abandon this plan entirely and simply run ALPHA_SE mode with the binary where instruction count-based checkpointing is already readily implemented? Thanks so much, Alex Edwards
_______________________________________________ gem5-users mailing list gem5-users@m5sim.org http://m5sim.org/cgi-bin/mailman/listinfo/gem5-users
