It has to be something to do with initialization... note that there's a Process class that's the common base class for SE workloads, and two classes derived from that, LiveProcess (for "regular" SE workloads) and EioProcess (for EIO traces). The first two are defined in sim/process.*, and the last one in eio/eio.*. There must be some initialization step that the InOrder CPU assumes that's happening in LiveProcess and either needs to be moved up to Process so it happens for EIO traces as well, or some equivalent but different step needs to be added to EioProcess.
Steve On Thu, Sep 24, 2009 at 5:14 PM, Korey Sewell <[email protected]> wrote: > Currently, InOrder doesnt support EIO traces as inputs. > > However, this might just be a matter of importing whatever code is used to > import EIO traces (encumbered?) and merging that into the InOrder stuff. > > I'm unfamiliar with how the EIO traces worked originally, but if the PCs > arent getting set up at all then it's not being loaded correctly. I would > check the constructors for that (inorder/cpu.cc). > > It seems like the problem you are describing is happening midstream with the > code. Maybe you can identify the problem instructions by comparing traces > from the AtomicSimpleCPU and the InOrderCPU? The tracediff tool is pretty > good (utils/tracediff) or if you want to do it manually, run both > simulations with the "ExecNoTicks" traceflag on for and then diff the > traces. > > I would guess the problem is right after a branch or syscall instruction... > > Can someone more familiar with EIO share some thoughts? > > On Thu, Sep 24, 2009 at 7:03 PM, Jing Xin <[email protected]> > wrote: >> >> Dear m5-users, >> >> I am trying to get the SPEC2000 eio traces to run on M5 with >> InOrderCPU in SE mode. I built m5 with "EXTRAS" support using this >> command: >> >> scons EXTRAS=/path/to/m5_2.0_encumbered build/ALPHA_SE/m5.debug >> >> To run the eio traces, I modified the "se.py" script following >> Michael's direction >> (http://osdir.com/ml/emulators.m5.users/2006-10/msg00013.html). Then I >> use the following command: >> >> ./build/ALPHA_SE/m5.debug ./configs/example/se.py --inorder --caches >> --l2cache --eio=/path/to/EIO_traces >> >> But I got core dump and the error message as following: >> **** REAL SIMULATION **** >> For more information see: http://www.m5sim.org/warn/7f11c91a >> info: Entering event queue @ 0. Starting simulation... >> panic: Tried to execute unmapped address 0. >> @ cycle 58000 >> [invoke:build/ALPHA_SE/arch/alpha/faults.cc, line 186] >> Memory Usage: 547308 KBytes >> For more information see: http://www.m5sim.org/panic/95542d88 >> Program aborted at cycle 58000 >> >> gzip: Abort (core dumped) >> stdout: Broken pipe >> ================================================== >> >> I installed several versions of M5 (with support for InOrderCPU model) >> from the repository and got the same error message. I printed some >> traces, it seems that there is bug in "Fetch-Seq-Unit" which assigns >> dynamic instruction to PC 0x000000 (From trace: "58000: >> system.cpu.Fetch-Seq-Unit: [tid:0]: Assigning [sn:3] to PC 0x000000, >> NPC 0x000004, NNPC 0x000008"). I looked into the source code of >> "cpu/inorder/first_stage.cc" and related files but still cannot figure >> out the problem. >> >> I wonder if someone met the same problem before (at least I haven't >> found solution from the list archive) and how to make it work >> correctly. Any help would be greatly appreciated. Thanks in advance. >> >> >> Best, >> Jing >> _______________________________________________ >> m5-users mailing list >> [email protected] >> http://m5sim.org/cgi-bin/mailman/listinfo/m5-users > > > > -- > - Korey > > _______________________________________________ > 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
