Okay thanks let me try internally. Thanks Brendan. -------------------------------------------- On Wed, 15/6/16, Brendan Fitzgerald <[email protected]> wrote:
Subject: Re: [marss86-devel] strange IPC results with SPEC CPU2006 To: "saravanan s" <[email protected]> Cc: "msc" <[email protected]>, "MARSS List" <[email protected]> Date: Wednesday, 15 June, 2016, 7:24 PM Because SPEC is a licensed piece of software, we cannot just give it out. You have to purchase a license yourself if you wish to use it. On Wed, Jun 15, 2016 at 9:49 AM, saravanan s <[email protected]> wrote: Hello: Could you please share the steps to get the CPU 2006 licensed version or image where i can download ? and I will follow the steps to integrate into ubunty.natty.qcow2 image. Thank you for your help! I have only bzip2 which comes with default. But I would like to run all other workload as well. Please help me with that. Thank you! I am able to run ParsecROI, Stream, Splash2..yet to get the SPEC CPU 2006. Please provide more details to me. Thank you! With Best Regards, SARAVANAN.S Senior Research Engineer, IBM India Ltd Systems and Technologies Group, Embassy Golf Links Business Park, Bangalore-560 071 Tel(Direct):91-80-41777346 -------------------------------------------- On Tue, 31/5/16, Brendan Fitzgerald <[email protected]> wrote: Subject: Re: [marss86-devel] strange IPC results with SPEC CPU2006 To: "msc" <[email protected]> Cc: "MARSS List" <[email protected]> Date: Tuesday, 31 May, 2016, 5:37 PM With only 100M cycles, I can' be sure the program got to any meaningful portions. There are a few options you can try:1) Increase the number of cycles to something much higher. Think over a billion2) Run the benchmark directly instead of using the runspec command.3) Insert the ptl_call to start the simulation within the spec benchmarks When I run spec benchmarks, I can see IPC differences from 0.27 to 1.5 On Tue, May 31, 2016 at 12:00 AM, msc <[email protected]> wrote: Hi Yesterday, I created a SPEC_CPU2006.qcow2 image disk based on ubuntu-natty.qcow2 provided from MARSSx86 official website. I installed CPU 2006 files on '/root/' directory to make simulations easy with 'start_sim, stop_sim, kill_sim'. After the installation was completed, I followed CPU2006 documentation to build and run the workloads. The build operations were completed without any errors, and I used 'runspec' command as follows. ./start_sim ; runspec --config=myconfig.cfg --size=ref --noreportable --iterations=1 bzip2; ./stop_sim ;./kill_sim also, I ran all workloads based on that command lines. (additionally, I specified stopcycle 100M and single_core, ) However, today, when I check the simulation results, I show some strange results. All IPC results in log files are similar. To show what I say, I give you some IPC results as follows. GemsFDTD lbm wrf sphinx3 h264ref gcc libquantumuser 0.97849 0.97876 1.01513 0.97197 0.96678 0.99265 1.01742kernel 0.57844 0.59495 0.57975 0.57746 0.59781 0.59026 0.60412total 0.86600 0.87202 0.88925 0.85880 0.86203 0.87850 0.89806 (other workloads also have similar results) Recently, when I used parsecROI.qcow2, all workloads show different IPC results. I give you PARSEC workloads' IPC results (I ran the workloads with 10B cycles and single_core). swaptions fluidanimate vips freqmineuser 1.14556 1.28826 1.71456 1.12002kernel 0.37523 0.66409 0.91947 0.79926total 1.13923 1.27027 1.64984 1.10204 With the PARSEC results, I can't be sure that the SPEC CPU2006 disk I made is operated correctly, and I don't know what I missed if there are some problems. So, I want you to answer me about its correctness. If some problems are, please let me know how to solve the problems. Thanks in advance. _______________________________________________ http://www.marss86.org Marss86-Devel mailing list [email protected] https://www.cs.binghamton.edu/mailman/listinfo/marss86-devel -----Inline Attachment Follows----- _______________________________________________ http://www.marss86.org Marss86-Devel mailing list [email protected] https://www.cs.binghamton.edu/mailman/listinfo/marss86-devel _______________________________________________ http://www.marss86.org Marss86-Devel mailing list [email protected] https://www.cs.binghamton.edu/mailman/listinfo/marss86-devel
