Can you run the SPARC_FS regression successfully? Ali
On Feb 16, 2010, at 8:11 PM, prasun gera wrote: > Forgot that I edited a cc file and not a script and hence didn't > rebuild. I suppose this won't happen once i rebuild m5. > > On Wed, Feb 17, 2010 at 7:19 AM, prasun gera <[email protected]> > wrote: >> Ali, >> I saw the simulate(Tick num_cycles) function in >> build/SPARC_FS/sim/simulate.cc and it has the following lines of code >> (lines 58 to 60) >> >> Event *limit_event = >> new SimLoopExitEvent("simulate() limit reached", 0); >> mainEventQueue.schedule(limit_event, num_cycles); >> >> As far as I can tell, this is the only place where a limit_event is >> added to the event queue. (and should be the only one right?) So I >> commented the aforementioned lines out just to see what happens. >> However, m5 still exit with same error message about limit being >> reached. I expected m5 to exit with an assert failure(inside the >> following while loop) since the queue would be empty after the event >> before the limit_event is executed, but that didn't happen. So does >> it mean that another(possibly interfering) limit_event was added to >> the queue earlier? >> >> Thanks, >> Prasun >> >> On Mon, Feb 15, 2010 at 3:08 AM, Ali Saidi <[email protected]> wrote: >>> Prasun, >>> >>> I imagine what is happening is you're running the simple cpu, >>> booting >>> Solaris and then there is nothing to do, since you didn't specify >>> anything. The only think that occurs after that point are timer >>> interrupts which makes the simulation tick quite quickly up until >>> you >>> reach MaxTick. Have you looked at the terminal? What is the output >>> there? >>> >>> Ali >>> >>> On Feb 14, 2010, at 2:33 PM, prasun gera wrote: >>> >>>> Hi, >>>> You mentioned that I'm using the O3 CPU model. Isn't the default >>>> model >>>> simple atomic? I mean, I didn't pass any arguments to the script >>>> fs.py >>>> and from setCPUClass, it seemed as though it is using the simple >>>> atomic model. >>>> In fact, later I tried the command line >>>> >>>> build/SPARC_FS/m5.opt -v -d /tmp/output/ configs/example/fs.py -d >>>> -- >>>> caches >>>> >>>> to use the detailed CPU model but it threw an error >>>> NameError: global name 'DerivO3CPU' is not defined. >>>> >>>> >>>> On Sat, Feb 13, 2010 at 6:56 AM, Gabriel Michael Black >>>> <[email protected]> wrote: >>>>> It looks like the simulation ran out of things to do and stopped >>>>> at >>>>> the end of simulated time. You could use the Exec trace flag to >>>>> see >>>>> what, if anything, is executing when that happens. If the >>>>> simulation >>>>> runs for a while before failing, Exec will output a lot of text. >>>>> You'll want to start tracing close to the interesting point. >>>>> >>>>> One other thing I notice is that you're using the O3 CPU model >>>>> with >>>>> SPARC_FS. While that model should work with SPARC_SE and SPARC_FS >>>>> works with the simple CPUs, I don't know if anyone ever got the >>>>> bugs >>>>> worked out of that particular combination (someone please say >>>>> something if you know otherwise). That makes me think that O3 is >>>>> losing track of work that it needs to do, thinks it should become >>>>> idle, and effectively goes to sleep and never wakes up. >>>>> >>>>> Gabe >>>>> >>>>> Quoting prasun gera <[email protected]>: >>>>> >>>>>> I could boot solaris in SPARC_FS, but m5 exited abruptly after >>>>>> that >>>>>> with the following message: >>>>>> Exiting @ cycle 9223372036854775807 because simulate() limit >>>>>> reached >>>>>> >>>>>> The command line I executed was: >>>>>> build/SPARC_FS/m5.opt -v -d /tmp/output/ configs/example/fs.py >>>>>> >>>>>> Host system: Ubuntu 32 bit >>>>>> >>>>>> I tried it twice, and it quit at the same cycle count both the >>>>>> times. >>>>>> To ascertain whether the error was caused because of something I >>>>>> did, >>>>>> I didn't enter anything on the solaris terminal the second time. >>>>>> i.e. >>>>>> The computer was idle for the entire duration except for the boot >>>>>> command on opb. Has anyone run into a similar error? Or any hints >>>>>> regarding debugging this? >>>>>> >>>>>> >>>>>> On Fri, Feb 12, 2010 at 10:26 PM, Ali Saidi <[email protected]> >>>>>> wrote: >>>>>>> >>>>>>> The original binaries should work just fine, the _new versions >>>>>>> were ones >>>>>>> that we verified we could compile from source. >>>>>>> >>>>>>> Ali >>>>>>> >>>>>>> >>>>>>> On Fri, 12 Feb 2010 20:50:07 +0530, prasun gera <[email protected] >>>>>>>> >>>>>>> wrote: >>>>>>>> Figured it out. Copied the files to the binaries and disks >>>>>>>> directories >>>>>>>> and could run configs/example/fs.py after that. One small thing >>>>>>>> though. The names of the solaris binaries used in m5 have new >>>>>>>> as a >>>>>>>> suffix ( for eg. openboot_new.bin and q_new.bin). Does it mean >>>>>>>> that >>>>>>>> the original binaries from opensparc need to be modified in >>>>>>>> some >>>>>>>> way? >>>>>>>> _______________________________________________ >>>>>>>> m5-users mailing list >>>>>>>> [email protected] >>>>>>>> http://m5sim.org/cgi-bin/mailman/listinfo/m5-users >>>>>>> _______________________________________________ >>>>>>> m5-users mailing list >>>>>>> [email protected] >>>>>>> http://m5sim.org/cgi-bin/mailman/listinfo/m5-users >>>>>>> >>>>>> _______________________________________________ >>>>>> m5-users mailing list >>>>>> [email protected] >>>>>> http://m5sim.org/cgi-bin/mailman/listinfo/m5-users >>>>>> >>>>> >>>>> >>>>> _______________________________________________ >>>>> m5-users mailing list >>>>> [email protected] >>>>> http://m5sim.org/cgi-bin/mailman/listinfo/m5-users >>>>> >>>> _______________________________________________ >>>> m5-users mailing list >>>> [email protected] >>>> http://m5sim.org/cgi-bin/mailman/listinfo/m5-users >>>> >>> >>> _______________________________________________ >>> m5-users mailing list >>> [email protected] >>> http://m5sim.org/cgi-bin/mailman/listinfo/m5-users >>> >> > _______________________________________________ > m5-users mailing list > [email protected] > http://m5sim.org/cgi-bin/mailman/listinfo/m5-users > _______________________________________________ m5-users mailing list [email protected] http://m5sim.org/cgi-bin/mailman/listinfo/m5-users
