I agree that this is surprising, but it's hard for me to say if it is "right" or "wrong" without a lot more digging. If you find that this is a bug, please post a patch! We'd be happy to review and incorporate a fix.
Jason On Mon, Feb 13, 2017 at 11:26 AM Ayaz Akram <aaq...@gmail.com> wrote: > Hi Jason, thanks for replying. Actually, I am using SE mode and currently > relying on printings from the benchmarks themselves (these are SPEC2006 > benchmarks). I expect that due to different implementation of syscalls, > some executed portion can be different, but even the printings from > benchmarks for ARM and x86 (say till X number of pseudo instructions calls) > are not in sync. In other words, I expect that when a particular function > in the benchmark (where I have placed pseudo instruction) is called Xth > time, the same work should have been done by both binaries if the input is > same in both cases. That does not seem to be the case ! > > On Mon, Feb 13, 2017 at 12:05 PM, Jason Lowe-Power <ja...@lowepower.com> > wrote: > > Hi, > > Another difference between ARM and x86 is the OS that you're using. Maybe > the different paths are actually different paths through the OS (or > interrupts, etc.) and not different paths through your binaries. Even in SE > mode there may be different code paths based on how the syscalls are > implemented. > > I would check the output of your benchmarks (maybe add some debug printfs > in the benchmarks) to check to see what is happening. > > Jason > > On Sun, Feb 12, 2017 at 1:12 PM Ayaz Akram <aaq...@gmail.com> wrote: > > Hello ! > > I am trying to use gem5 pseudo instructions inside code of a benchmark to > create checkpoints at certain points in program execution (after pseudo > instruction is called for a specific number of times). The benchmark is > compiled for both Arm and x86. Once checkpoint is created I run the > benchmark till one more call to the pseudo instruction. The problem is > that x86 and arm differ in their path of execution (when run from > checkpoint), as x86 gets the one more pseduo-ins call after hundreds of > millions of instructions while arm does that after only a few million > instructions. I don't see any issues with the benchmark binaries, > compilation process, path of the input files. I wonder if the way the > checkpoints are created for different ISAs can result into such an issue ? > I will appreciate if someone can point out something I might be missing. > > > -Thanks for your time ! > _______________________________________________ > gem5-users mailing list > gem5-users@gem5.org > http://m5sim.org/cgi-bin/mailman/listinfo/gem5-users > > > _______________________________________________ > gem5-users mailing list > gem5-users@gem5.org > http://m5sim.org/cgi-bin/mailman/listinfo/gem5-users > > > _______________________________________________ > gem5-users mailing list > gem5-users@gem5.org > http://m5sim.org/cgi-bin/mailman/listinfo/gem5-users
_______________________________________________ gem5-users mailing list gem5-users@gem5.org http://m5sim.org/cgi-bin/mailman/listinfo/gem5-users
