Ali, thanks for the reply. I have a related question, hope you can give
some advice.
Instead of using setaffinity call, is it possible to hack the gem5
interface with the linux OS, so that I can assign any thread (with its
instruction workloads) to any cpu I want to? The reason to do this is
because I want to switch cpus & thread bindings in the middle of a
simulation, while m5 pin can't do it (I guess, without modifying benchmark
source).
For example, at the beginning I have 4 cores running one threads per core.
Then I want to move thread 2 to core 0 and thread 3 to core 1, leaving core
2&3 idle. Is it possible to do it on the  Gem5 side only, without changing
the Linux OS?

Thanks in advance
Hui


On Wed, Apr 10, 2013 at 11:27 PM, Ali Saidi <[email protected]> wrote:

> Where the instruction counts on the cores the same? If the application was
> small enough the instruction and ilc might be dominated by the system
> booting.
>
> Ali
>
> On Apr 9, 2013, at 9:54 PM, Hui Zhao <[email protected]> wrote:
>
> > I am running a parsec benchmark with 4 threads on 4 core Alpha FS
> machine, I used "m5 pin" to bind all 4 threads to the first 2 cores.
> However, when I checked the ipc in the stats, I saw all 4 cores have
> similar ipc. I am expecting that only first 2 cores have reasonable ipc,
> and other cores should have 0 or very low ipc.
> >  The linux kernel is vmlinux_2.6.27-gcc_4.3.4 which should have the
> sched_setaffinity funciton working.  Does anyone know why?
> >
> > Thanks
> > _______________________________________________
> > gem5-users mailing list
> > [email protected]
> > http://m5sim.org/cgi-bin/mailman/listinfo/gem5-users
>
> _______________________________________________
> gem5-users mailing list
> [email protected]
> http://m5sim.org/cgi-bin/mailman/listinfo/gem5-users
>
_______________________________________________
gem5-users mailing list
[email protected]
http://m5sim.org/cgi-bin/mailman/listinfo/gem5-users

Reply via email to