My guess is one of the threads in Simulation B started committed instructions from the wrong path but I cant be sure...
The way I would debug it is quite involved since there are many threads running simultaneously. You need to figure out which CPU and which thread is causing the error... ***Assuming your benchmarks already work on an AtomicSimpleCPU ***, First thing I would do is run GDB on simulation B, then backtrace to the frame that has "faults.cc". In that frame, you can query the CPU and hopefully the thread that caused the fault through a member function of the fault class. Or maybe, you can backtrace some more to the CPU code that calls the fault function and you can query that information there. Somewhere a "p this->cpuId()" in GDB would do the trick. The ThreadContext can be queried to figure out which thread. Once you figure out the CPU, I would then get traces of your benchmarks in AtomicCPU without the ticks (so you can compare). Run each sim individually with a traceflag like "Exec,-ExecTicks". I'm not sure which flag removes the thread number but you can find that in src/cpu/SConscript for the cpu trace flags. Next, get execution traces from your O3CPU simulation run and separate the the tracedata for each benchmark. Finally, you can diff the SMT (O3) individual traces with the Atomic individual traces to figure out where the first bad instruction is (which will lead you to the problem). Yea, I know that's quite involved but since the error message is so bland, I dont know what the quick fix would be. On Sat, May 9, 2009 at 2:22 PM, Devraj Chapagain <[email protected]> wrote: > hi everyone, > I am testing simulations using SPEC CPU 2006 in SE mode. > Here, i have posted two simulations where simulation A works and Simulation > B doesn't work (Please see error message attached below in simulation B). > > Simulation A: > > num of cpus = 2 > num threads per cpu = 2 > system.cpu[0].workload = [Mybench.bzip2(), Mybench.libquantum()] > system.cpu[1].workload = [Mybench.gobmk(), Mybench.soplex()] > ........ > L1 I cache: size= 32KB, L1 D cache: size= 32KB, block size= 64 > L2 cache: size= 1MB, block size= 64 > L3 cache: size= 2MB, block size= 64 > > Simulation B: > > num of cpus = 2 > num threads per cpu = 4 > system.cpu[0].workload = [Mybench.milc(), Mybench.soplex(), > Mybench.gromacs(), Mybench.cactusADM()] > system.cpu[1].workload = [Mybench.gcc(), Mybench.leslie3d(), > Mybench.gobmk(), Mybench.hmmer()] > ......... > L1 I cache: size= 32KB, L1 D cache: size= 32KB, block size= 64 > L2 cache: size= 1MB, block size= 64 > L3 cache: size= 2MB, block size= 64 > > error message: > ...................................................................................... > warn: Increasing stack size by one page. > panic: Tried to execute unmapped address 0. > @ cycle 7515000 > [invoke:build/ALPHA_SE/arch/alpha/faults.cc, line 186] > Program aborted at cycle 7515000 > ........................................................................................ > > Could anyone please help me to figure out how to solve this problem? > > Thanks in advance, > devraj. > _______________________________________________ > 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
