[mvapich-discuss] MPI_THREAD_MULTIPLE support
Martin Cuma
martin.cuma at utah.edu
Mon Nov 7 15:50:42 EST 2011
Hi Krishna,
thanks for your reply. I do disable the affinity, use MVAPICH2 1.7 built
with PGI compilers, though, I get this with Intel as well.
Later on last night, I discovered that the Nemesis device works fine and
even performs faster (my code is heavy in Allgatherv). So, I am happy to
use the OFA-IB-Nemesis device instead of the OFA-IB-CH3 where I see this
error.
If you still want to pursue this and file a bug report, let me know and
I'll try to package up the code so that you can reproduce it.
Thanks,
MC
On Mon, 7 Nov 2011, Krishna Kandalla wrote:
> Hi Martin,
> Thanks for reporting this issue. Could you please let us know which version of MVAPICH2
> you are using? Also, as described in our userguide
> (http://mvapich.cse.ohio-state.edu/support/user_guide_mvapich2-1.7.html#x1-700006.14), could
> you please set MV2_ENABLE_AFFINITY=0 and try re-running your application.
>
> Thanks,
> Krishna
>
> On Sun, Nov 6, 2011 at 10:49 PM, Martin Cuma <martin.cuma at utah.edu> wrote:
> Hello,
>
> I am wondering how well is MPI_THREAD_MULTIPLE supported in MVAPICH2 built with
> --with-rdma=gen2. The reason I am asking is that I am getting deadlock in my code,
> which does communication from multiple threads/processes. The code is running fine
> with MPICH2, nemesis device.
>
> The code does bunch of isends/irecvs, the thread number is always the same among
> the communication pair, the processes involved differ. For example, if process 0
> sends to 1, then all threads (say 0-3) on process 0 do one send each, addressed to
> the respective thread on process 1. So, in total, I do 4 sends, thr0 proc0 to thr0
> proc1, thr1 proc0 to thr1 proc2, etc. I am ensuring the right recipient via
> different tags for each thread.
>
> Can someone please comment their experience with communication pattern like that
> and what could be the cause of the problem I am seeing?
>
> Thanks,
> MC
>
> --
> Martin Cuma
> Center for High Performance Computing
> University of Utah
> _______________________________________________
> mvapich-discuss mailing list
> mvapich-discuss at cse.ohio-state.edu
> http://mail.cse.ohio-state.edu/mailman/listinfo/mvapich-discuss
>
>
>
>
--
Martin Cuma
Center for High Performance Computing
University of Utah
More information about the mvapich-discuss
mailing list