On Thu, May 5, 2011 at 5:49 PM, Xiuyuan Bi <[email protected]> wrote:
> Hi, everyone.
>
> I compiled marss-0.1.2 with "scons -Q debug=1 c=4".
> And I run the benchmark "blackscholes" with thread number = 8, from
> the parsec-2.1 suite, which is provided by the marss86 website.
> I run this benchmark with 100m instructions two times, with the same
> configuration. When I run the benchmark, only ROI (Region of Interest) is
> logged.
> But the result of each time is significantly different. One IPC is
> 1.79, and the other is 4.29. I read the IPC from ptlsim.log file.
>
> The IPC in ptlsim.log file is not correct, please disregard that.
> When I compare their stats file, I found that the one with Higher
> ipc has 99.9% instructions in kernal64 mode (stats file -> external -> total
> -> insns_in_mode), while the one with Lower ipc has only 1.8% instructions
> in kernal64 mode.
>
> There was a bug in 0.1.2 release that has fixed recently in 'master'
branch. This bug was affecting scheduling of processes in single-core and
multi-core mode, and keep the simulation running in kernel mode
for infinite time. Grab the latest code from github and see if you still see
similar results or not.
- Avadh
Does anyone have any idea why the ipc differ so much? and how to
> solve it?
>
> Xiuyuan
> _______________________________________________
> 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
