Good catch, thanks... but then how did it work correctly in b3? Did that change recently?
On Nov 15, 2007 12:42 AM, Ali Saidi <[EMAIL PROTECTED]> wrote: > The args are fine, you're just passing the arguments to m5 > incorrectly. The command should be an array of strings (like argv) not > one big string. So changing the cmd line to: cmd = ['FFT', '-p', > str(options.numcpus), ' -m18'] should fix the problem. > > Ali > > > >>> > On Nov 15, 2007, at 2:25 AM, Gabe Black wrote: > > > I don't remember changing anything having to do with Alpha process > > initialization other than the changes I made to paging, aka TLBs in SE > > mode and everything that went with it. I don't want to say for sure > > that > > I didn't since it was a fairly long time between b3 and b4, but I > > would > > be surprised if I did. > > > > Gabe > > > > Steve Reinhardt wrote: > >> Interesting... from what you're reporting, it sounds like the problem > >> is likely in the code that sets up the stack (argv, argc, and all > >> that > >> good stuff). That's in LiveProcess::argsInit() in sim/process.cc. I > >> don't know what if anything has changed in there from b3 to b4 > >> though... I don't recall touching it, but Gabe might have. > >> > >> The next step would be to put a breakpoint in that function and see > >> if > >> the argv array really does have the right number of entries, and that > >> the entries point to the arg strings. > >> > >> Steve > >> > >> On Nov 14, 2007 10:36 PM, Sujay Phadke <[EMAIL PROTECTED]> > >> wrote: > >> > >>> A typo. In beta, the number of complex doubles reported is: > >>> 262144 Complex > >>> Doubles, which is correct since it should be 2^18. (the -m18 > >>> switch passes > >>> this value). > >>> > >>> So it there some problem passing these values to the benchmarks > >>> (atleast in > >>> SE mode in beta4?) HAs something is parameter passing changed > >>> between beta 3 > >>> and beta 4? > >>> > >>> ----- Original Message ----- > >>> From: "Sujay Phadke" <[EMAIL PROTECTED]> > >>> To: "M5 users mailing list" <m5-users@m5sim.org> > >>> > >>> Sent: Thursday, November 15, 2007 1:29 AM > >>> Subject: Re: [m5-users] sims with m5 beta4 - cache stats - any > >>> update on > >>> this? > >>> > >>> > >>> > >>>> ok I looked at the output and I think what is happening only 1 > >>>> processor > >>>> ie being assigned to the benchmark, ragardless of what value i > >>>> put. But I > >>>> dont know why this is happening. I type: > >>>> > >>>> ./build/ALPHA_SE/m5.opt configs/splash2/runsplash.py -n 4 -k 4 -b > >>>> FFT > >>>> > >>>> and still at the end of simulation using beta 4, it reports this: > >>>> 1024 Complex Doubles > >>>> 1 Processors > >>>> 65536 Cache lines > >>>> 16 Byte line size > >>>> 4096 Bytes per page > >>>> > >>>> if I use beta3, this is what is reported: > >>>> 1024 Complex Doubles > >>>> 4 Processors > >>>> 65536 Cache lines > >>>> 16 Byte line size > >>>> 4096 Bytes per page > >>>> > >>>> the line in the config script that invokes fft is: > >>>> class FFT(LiveProcess): > >>>> cwd = options.rootdir + '/kernels/fft' > >>>> executable = options.rootdir + '/kernels/fft/FFT' > >>>> cmd = 'FFT -p' + str(options.numcpus) + ' -m18' > >>>> > >>>> If I do a print options.numcpus here, it correctly prints out > >>>> "4". So n=4 > >>>> is definitely being passed onto fft. > >>>> > >>>> I dont know why in beta 4, the correct numcpus is not being used > >>>> by fft, > >>>> or some other benchmarks. > >>>> > >>>> Any ideas? > >>>> > >>>> Thanks for your help. > >>>> > >>>> - Sujay > >>>> > >>>> > >>>> ----- Original Message ----- > >>>> From: "Steve Reinhardt" <[EMAIL PROTECTED]> > >>>> To: "M5 users mailing list" <m5-users@m5sim.org> > >>>> Sent: Thursday, November 15, 2007 12:03 AM > >>>> Subject: Re: [m5-users] sims with m5 beta4 - cache stats - any > >>>> update on > >>>> this? > >>>> > >>>> > >>>> > >>>>> I don't recall what the output files are called... each > >>>>> benchmark is a > >>>>> little different in splash. You're looking for the benchmark > >>>>> output > >>>>> in addition to the m5 output... I'm guessing something bad > >>>>> happened > >>>>> and the benchmark quit because it encountered an error. > >>>>> > >>>>> On Nov 14, 2007 4:26 PM, Sujay Phadke <[EMAIL PROTECTED]> > >>>>> wrote: > >>>>> > >>>>>> Ok. Though I dont know what to look for, since there's nothing > >>>>>> that I > >>>>>> have > >>>>>> changed since beta3. Which output files should I examine? > >>>>>> The benchmarks ends because m5 reports "target called exit()". > >>>>>> > >>>>>> - Sujay > >>>>>> > >>>>>> ----- Original Message ----- > >>>>>> From: "Steve Reinhardt" <[EMAIL PROTECTED]> > >>>>>> To: "M5 users mailing list" <m5-users@m5sim.org> > >>>>>> Sent: Wednesday, November 14, 2007 6:31 PM > >>>>>> Subject: Re: [m5-users] sims with m5 beta4 - cache stats - any > >>>>>> update on > >>>>>> this? > >>>>>> > >>>>>> > >>>>>> > >>>>>> > >>>>>>> Thanks, this is very interesting. Have you looked at the > >>>>>>> output from > >>>>>>> the benchmarks that are producing weird results? I'd say from > >>>>>>> those > >>>>>>> stats (and your comment that those benchmarks are finishing > >>>>>>> "extremely > >>>>>>> fast") that the real problem is that the benchmark is > >>>>>>> terminating > >>>>>>> early, before it even gets to forking off threads on the other > >>>>>>> CPUs, > >>>>>>> due to some error. Presumably the output is indicating what the > >>>>>>> problem is if you look in the right place. > >>>>>>> > >>>>>>> So there's still some problem with b4 if you weren't > >>>>>>> encountering that > >>>>>>> error in b3, but it's not that the cache statistics are broken. > >>>>>>> > >>>>>>> Steve > >>>>>>> > >>>>>>> _______________________________________________ > >>>>>>> m5-users mailing list > >>>>>>> m5-users@m5sim.org > >>>>>>> http://m5sim.org/cgi-bin/mailman/listinfo/m5-users > >>>>>>> > >>>>>>> > >>>>>> _______________________________________________ > >>>>>> m5-users mailing list > >>>>>> m5-users@m5sim.org > >>>>>> http://m5sim.org/cgi-bin/mailman/listinfo/m5-users > >>>>>> > >>>>>> > >>>>> _______________________________________________ > >>>>> m5-users mailing list > >>>>> m5-users@m5sim.org > >>>>> http://m5sim.org/cgi-bin/mailman/listinfo/m5-users > >>>>> > >>>>> > >>>> _______________________________________________ > >>>> m5-users mailing list > >>>> m5-users@m5sim.org > >>>> http://m5sim.org/cgi-bin/mailman/listinfo/m5-users > >>>> > >>>> > >>> _______________________________________________ > >>> m5-users mailing list > >>> m5-users@m5sim.org > >>> http://m5sim.org/cgi-bin/mailman/listinfo/m5-users > >>> > >>> > >> _______________________________________________ > >> m5-users mailing list > >> m5-users@m5sim.org > >> http://m5sim.org/cgi-bin/mailman/listinfo/m5-users > >> > > > > _______________________________________________ > > m5-users mailing list > > m5-users@m5sim.org > > http://m5sim.org/cgi-bin/mailman/listinfo/m5-users > > > > _______________________________________________ > m5-users mailing list > m5-users@m5sim.org > http://m5sim.org/cgi-bin/mailman/listinfo/m5-users > _______________________________________________ m5-users mailing list m5-users@m5sim.org http://m5sim.org/cgi-bin/mailman/listinfo/m5-users
