bzip2: host_inst_rate = 27400 lbm: host_inst_rate = 17670 Number of sim ticks is 100,000,000,000 and sim freq is 1,000,000,000,000 which I think are reasonable (long enough to discoount startup overheads).
as I asked before, I wonder what happen in case of an unimplemented x87 instruction, then it will be ignored. Is that right? Then they are skipped and simulated ended quickly. What is the probability of that? On 1/14/12, Steve Reinhardt <ste...@gmail.com> wrote: > Simulator runtime is primarily determined by the number of events > processed, like instructions executed, cache misses, etc. The complexity > of those individual events is typically a second-order effect. The number > of ticks simulated is only loosely correlated with this. > > The host_inst_rate stat reports simulated instructions per wall clock > second; for a given CPU model, this should be roughly similar across > different benchmarks on a given host, assuming your run is long enough to > discount startup overheads. > > Steve > > On Fri, Jan 13, 2012 at 12:02 PM, Mahmood Naderan > <mahmood...@gmail.com>wrote: > >> that is also possible. Since lbm is floating point benchmark, my >> concern is that some x87 instruction are not implemented so they are >> actually *skipped*. >> >> Hope that I am able to state the problem. >> >> On 1/13/12, Paul Rosenfeld <dramnin...@gmail.com> wrote: >> > Well, you have to look at the complexity of the instructions in the >> binary >> > in terms of how they are simulated. If bzip2 uses some specific >> > instructions in ISA X which are then simulated in an inefficient way on >> > host machine Y, you could have a big discrepancy in how long each tick >> > takes to execute. >> > >> > So I'd say this is perfectly possible. >> > >> > On Fri, Jan 13, 2012 at 9:03 AM, Mahmood Naderan >> > <mahmood...@gmail.com>wrote: >> > >> >> with the above config, ipc of bzip2 is 1.53 and lbm is 0.5 >> >> >> >> On 1/13/12, Mehmet Burak Aykenar <mbayke...@etu.edu.tr> wrote: >> >> > I am not quite sure but the "IPC" of 'lbm' might be higher than >> 'bzip2' >> >> .. >> >> > In my tests, benchmarks having higher "IPC" usually finish earlier.. >> >> > >> >> > Regards, >> >> > Burak >> >> > >> >> > 2012/1/13 Mahmood Naderan <mahmood...@gmail.com> >> >> > >> >> >> hi >> >> >> I ran two benchmarks at the same time with similar number of fast >> >> >> forwards and maxticks. But the lbm (with reference input which is >> big) >> >> >> reaches the end much much sooner than bzip2 (with test input). >> >> >> >> >> >> >> >> >> configs/example/cmp.py --prog-interval=1000000 -F 1000000 --maxtick >> >> >> 100000000000 -d --caches --l2cache -b lbm >> >> >> >> >> >> configs/example/cmp.py --prog-interval=1000000 -F 1000000 --maxtick >> >> >> 100000000000 -d --caches --l2cache -b bzip2_dryer >> >> >> >> >> >> Is that normal? >> >> >> -- >> >> >> // Naderan *Mahmood; >> >> >> _______________________________________________ >> >> >> gem5-users mailing list >> >> >> gem5-users@gem5.org >> >> >> http://m5sim.org/cgi-bin/mailman/listinfo/gem5-users >> >> >> >> >> >> >> >> >> !DSPAM:6374,4f101b927441533729461! >> >> >> >> >> >> >> >> >> >> >> > >> >> > >> >> > -- >> >> > *Mehmet Burak AYKENAR >> >> > Graduate Student of Computer Engineering >> >> > Research Assistant >> >> > **TOBB ETU Z-10 >> >> > Microprocessors Laboratory* >> >> > * >> >> > mbaykenar.etu.edu.tr >> >> > mbayke...@etu.edu.tr* >> >> > >> >> >> >> >> >> -- >> >> -- >> >> // Naderan *Mahmood; >> >> _______________________________________________ >> >> gem5-users mailing list >> >> gem5-users@gem5.org >> >> http://m5sim.org/cgi-bin/mailman/listinfo/gem5-users >> >> >> > >> >> >> -- >> -- >> // Naderan *Mahmood; >> _______________________________________________ >> gem5-users mailing list >> gem5-users@gem5.org >> http://m5sim.org/cgi-bin/mailman/listinfo/gem5-users >> > -- -- // Naderan *Mahmood; _______________________________________________ gem5-users mailing list gem5-users@gem5.org http://m5sim.org/cgi-bin/mailman/listinfo/gem5-users
