Hi Rick As I have said earlier we have assigning each process to different cores with the following set of spec cpu 2006 benchmarks --sphinx3, gobmk, bwaves, bzip2, bwaves, bwaves, sphinx3, libquantum, gobmk, h264ref, namd, bzip2, libquantum, h264ref, libquantum, namd. assigning in following way:
max_insts_all_threads = 10000000 num of cpus = 8 num threads per cpu = 2 system.cpu[0].workload = [Mybench.sphinx3, Mybench.gobmk] system.cpu[1].workload = [Mybench.bwaves, Mybench.bzip2] system.cpu[2].workload = [Mybench.bwaves, Mybench.bwaves] system.cpu[3].workload = [Mybench.sphinx3, Mybench.libquantum] system.cpu[4].workload = [Mybench.gobmk, Mybench.h264ref] system.cpu[5].workload = [Mybench.namd, Mybench.bzip2] system.cpu[6].workload = [Mybench.libquantum, Mybench.h264ref] system.cpu[7].workload = [Mybench.libquantum, Mybench.namd] We got results on m5stats file with some cpu's values to be 0. But on the otherhand when I changed the order of the benchmarks, it produces the desired result with values in all the cpu's. I have tested the benchmarks on m5 and we are using only those benchmarks that works ok when tested for a uniprocessor(single core single thread). We are not understanding why we got a such result *where for one set of benchmarks assigned to different cores, it didn't work but on the otherhand with the same set of benchmarks but different assignments, it worked.* Please let us out from such an ambiguous situation Thanks Ashutosh Jain On Thu, Feb 12, 2009 at 4:03 PM, Ashutosh Jain <[email protected]> wrote: > HI > > > > On Tue, Feb 10, 2009 at 12:13 PM, Rick Strong <[email protected]> wrote: > >> >> ---------------------------------------------------------------------------------------------------------------------------- >> > >> > *Do i need to hack configs/example/se.py or use the use the patch by >> > Jiayuan Meng ?* >> > >> > Thanks >> > HARI >> >> You need to do both. Jiayuan's patch is designed to run a single >> benchmark with a mesh model. If you want to run a workload, you need to >> assign each process to a different core in configs/example/se.py. > > > > While starting simulation we have....... > > max_insts_all_threads = 100000000 > num of cpus = 8 > num threads per cpu = 2 > system.cpu[0].workload = [Mybench.sphinx3, Mybench.gobmk] > system.cpu[1].workload = [Mybench.bwaves, Mybench.bzip2] > system.cpu[2].workload = [Mybench.bwaves, Mybench.bwaves] > system.cpu[3].workload = [Mybench.sphinx3, Mybench.libquantum] > system.cpu[4].workload = [Mybench.gobmk, Mybench.h264ref] > system.cpu[5].workload = [Mybench.namd, Mybench.bzip2] > system.cpu[6].workload = [Mybench.libquantum, Mybench.h264ref] > system.cpu[7].workload = [Mybench.libquantum, Mybench.namd] > > With the above running simulation, its clear that we are assigning each > process to different cores. Again the situation is still the same as while > running the simulation we got statistics file(m5stats.txt) with some cpu(s) > have all the result values to be 0. We dont understand why this is happening > even when we are assigning each process to different cores. > > > > Thanks > Ashutosh Jain > > >
_______________________________________________ m5-users mailing list [email protected] http://m5sim.org/cgi-bin/mailman/listinfo/m5-users
