If you can run the benchmarks I listed in the first post using ALPHA_SE, then the problem is with X86_SE syscalls. I wanted to be sure that they are runable with gem5. The benchmarks are compiled with static options.
For example, when I run tonto, I see that at the beginning of simulation the whole 32GB memory of host machine is eaten!! That mean there is a bug that eat all memory. For some syscalls that are not implemeted in X86 (ioctl, lstat), I did the same thing posted at http://www.csl.cornell.edu/~vince/projects/m5/m5_x86_64_se_status.html So maybe the problem is with *disabling the syscalls*. I don't know. On 2/13/12, [email protected] <[email protected]> wrote: > > There is a list including all the system calls implemented and > unimplemented. Apparently ioctl is not implemented. Either you can change > fatal to warning to see if it works, or run it in FS mode instead of SE > mode. Or you can implement it by yourself. > > On Mon, 13 Feb 2012 13:06:49 -0500, [email protected] wrote: >> Hi Matt, >> >> I am trying to run some SPEC2006 benchmarks as well and I've tried >> running with the se.py method that you mentioned but it seems to work >> for certain benchmarks but not others. For example, I was able to run >> bzip2 fine but when I tired to run perlbench or bwaves, I get the >> following error: >> >> fatal: syscall ioctl (#16) unimplemented. >> @ cycle 5818000 >> [unimplementedFunc:build/X86_SE/sim/syscall_emul.cc, line 83] >> >> >> Do you have any idea how to fix this? >> >> Thanks >> >> Quoting Matt Poremba <[email protected]>: >> >>> Hi Mahmood, >>> >>> >>> You need to compile SPEC2006 statically to start. After that you can >>> either >>> run directly using se.py, for example: >>> >>> build/ALPHA_MESI_CMP_directory/gem5.opt configs/example/se.py -c >>> /path/to/CPU2006/bin/dealII -o 23 >>> >>> or, you can use the LiveProcess variables defined here: >>> http://gem5.org/SPEC2006_benchmarks . These scripts don't exactly work > if >>> you copy and paste them from there (you'll need to set binary_dir and it >>> only supports 1 CPU among other things). The key thing is to set >>> "system.cpu[i].workload >>> = some_LiveProcess_variable". Personally, I wrote my own script to take > a >>> list of spec benchmarks to run and assigned the corresponding > LiveProcess >>> to each CPU. This worked fine. >>> >>> >>> -- >>> Matt Poremba >>> Ph.D. Candidate >>> 111N IST Building >>> Pennsylvania State University >>> University Park, PA 16802 >>> Phone: 814-689-9447 >>> Email: [email protected] >>> >>> >>> >>> On Mon, Feb 13, 2012 at 9:57 AM, Mahmood Naderan >>> <[email protected]>wrote: >>> >>>> Hi >>>> Has anyone tried the following benchmarks in SE mode (any ISA)? >>>> >>>> perlbench >>>> gamess >>>> leslie3d >>>> namd >>>> dealII >>>> tonto >>>> wrf >>>> sphinx3 >>>> >>>> If yes, please share your solution >>>> Thanks, >>>> -- >>>> // Naderan *Mahmood; >>>> _______________________________________________ >>>> gem5-users mailing list >>>> [email protected] >>>> http://m5sim.org/cgi-bin/mailman/listinfo/gem5-users >>>> >>> >> >> >> >> >> >> _______________________________________________ >> gem5-users mailing list >> [email protected] >> http://m5sim.org/cgi-bin/mailman/listinfo/gem5-users > _______________________________________________ > gem5-users mailing list > [email protected] > http://m5sim.org/cgi-bin/mailman/listinfo/gem5-users > -- -- // Naderan *Mahmood; _______________________________________________ gem5-users mailing list [email protected] http://m5sim.org/cgi-bin/mailman/listinfo/gem5-users
