I found the deadlock during my development of modified o3 model based on M5.
The deadlock condition rarely happened so that M5's o3 could work with all
splash2 kernels. However, the deadlock did show up in my modified version of
o3 when running Cholesky. After reviewing M5's o3, the deadlock may apply to
original o3 as well.

I attach a diff file generated by "hg export" based on the m5-r7088. I tried
the fix with Choelsky, FFT, Radix, all using default configurations in M5
release. Those three kernels work ed fine with the fix.

Thanks.
Liang

 *------------------
Wang, Liang
Dept. of Computer Science, SEAS
Univ. of Virginia, Charlottesville, VA*


On Wed, Oct 13, 2010 at 11:42 PM, Ali Saidi <[email protected]> wrote:

> HI Liang,
>
> Thank you for reporting the issue. It appears as though you've also
> included a fix in the source file. Have you tried the fix? Does it work? Are
> there any issues? Regression changes? Would you provide us with a diff?
>
> Thanks,
> Ali
>
> On Oct 11, 2010, at 8:56 PM, Liang Wang wrote:
>
> > The problematic code segment is in src/cpu/o3/iew_impl.hh with latest m5
> (r7704).
> >
> > Details are presented in the source as comment at line started form 1508.
> > ------------------
> > Wang, Liang
> > Dept. of Computer Science, SEAS
> > Univ. of Virginia, Charlottesville, VA
> > <iew_impl.hh>_______________________________________________
> > m5-users mailing list
> > [email protected]
> > http://m5sim.org/cgi-bin/mailman/listinfo/m5-users
>
> _______________________________________________
> m5-users mailing list
> [email protected]
> http://m5sim.org/cgi-bin/mailman/listinfo/m5-users
>

Attachment: deadlock.diff
Description: Binary data

_______________________________________________
m5-users mailing list
[email protected]
http://m5sim.org/cgi-bin/mailman/listinfo/m5-users

Reply via email to