<div class="gmail_quote">Pieter, you should pull the latest to get the fix for this ticket. I don't know whether it could affect your current observations, but it could trip you up anyway (depending on how your implementation works).</div>
<div class="gmail_quote"><br></div><div class="gmail_quote"><a href="http://trac.mcs.anl.gov/projects/mpich2/ticket/1601">http://trac.mcs.anl.gov/projects/mpich2/ticket/1601</a></div><div class="gmail_quote"><br></div><div class="gmail_quote">
On Sat, Apr 7, 2012 at 02:42, Pieter Ghysels <span dir="ltr"><<a href="mailto:pieter.ghysels@ua.ac.be">pieter.ghysels@ua.ac.be</a>></span> wrote:<br><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">
Ah, yes, sorry for the confusion. I am also overlapping the<br>
MPIX_Iallreduce with other (one-sided) MPI calls. But these calls are<br>
buried in the library I use, and I'm not so familiar with one sided<br>
communication. But I was told that there are no calls to MPI_Win_create<br>
or MPI_Win_free, except at the very beginning and end of the program.<br>
<br>
Is it possible to overlap different MPI calls like that?<br>
<br>
I configured MPICH2 with<br>
--enable-threads=runtime<br>
and initialized with<br>
MPI_Init_thread(argc, argv, MPI_THREAD_MULTIPLE, &thread_support);<br>
<br>
<br>
Pieter<br>
<div class="HOEnZb"><div class="h5"><br>
On Fri, 2012-04-06 at 19:27 -0500, Jeff Hammond wrote:<br>
> One way to implement nonblocking allreduce would be to use<br>
> MPI_Accumulate, so it doesn't seem entirely nuts. I presume you know<br>
> the MPICH2 implementation of MPIX_Iallreduce does not use RMA...<br>
><br>
> Jeff<br>
><br>
> On Fri, Apr 6, 2012 at 6:24 PM, Rajeev Thakur <<a href="mailto:thakur@mcs.anl.gov">thakur@mcs.anl.gov</a>> wrote:<br>
> > You mean your program doesn't call MPI_Win_create and MPI_win_free but Jumpshot shows them? No, that's not normal.<br>
> ><br>
> > Rajeev<br>
> ><br>
> > On Apr 6, 2012, at 3:14 PM, Pieter Ghysels wrote:<br>
> ><br>
> >> Dear,<br>
> >><br>
> >> I'm replacing calls to MPI_Allreduce by calls to MPIX_Iallreduce.<br>
> >> I'm compiling with -mpe=mpilog and look at the communication pattern<br>
> >> with jumpshot. The MPI_Allreduce's are gone now of course, but<br>
> >> mpe/jumpshot does not show MPIX_Iallreduce. Instead, where the<br>
> >> MPI_Allreduce used to be, now a lot of time is being spent in<br>
> >> MPI_Win_create and MPI_Win_free, where before the MPI_Allreduce was<br>
> >> taking place. Is this normal?<br>
> >><br>
> >> Can I configure MPICH2 to use a progress thread for the nonblocking<br>
> >> collectives, as in libNBC?<br>
> >> Do I have to use MPI_Init_thread()?<br>
> >><br>
> >> Pieter<br>
> >><br>
> >> _______________________________________________<br>
> >> mpich-discuss mailing list <a href="mailto:mpich-discuss@mcs.anl.gov">mpich-discuss@mcs.anl.gov</a><br>
> >> To manage subscription options or unsubscribe:<br>
> >> <a href="https://lists.mcs.anl.gov/mailman/listinfo/mpich-discuss" target="_blank">https://lists.mcs.anl.gov/mailman/listinfo/mpich-discuss</a><br>
> ><br>
> > _______________________________________________<br>
> > mpich-discuss mailing list <a href="mailto:mpich-discuss@mcs.anl.gov">mpich-discuss@mcs.anl.gov</a><br>
> > To manage subscription options or unsubscribe:<br>
> > <a href="https://lists.mcs.anl.gov/mailman/listinfo/mpich-discuss" target="_blank">https://lists.mcs.anl.gov/mailman/listinfo/mpich-discuss</a><br>
><br>
><br>
><br>
<br>
<br>
_______________________________________________<br>
mpich-discuss mailing list <a href="mailto:mpich-discuss@mcs.anl.gov">mpich-discuss@mcs.anl.gov</a><br>
To manage subscription options or unsubscribe:<br>
<a href="https://lists.mcs.anl.gov/mailman/listinfo/mpich-discuss" target="_blank">https://lists.mcs.anl.gov/mailman/listinfo/mpich-discuss</a><br>
</div></div></blockquote></div><br>