[mpiwg-rma] Memory barriers in passive-target RMA

Jeff Hammond jeff.science at gmail.com
Mon Jul 14 13:17:48 CDT 2014

Can you resend without the awful HTML punctuation?

=3D=3D is not easy to parse. 


Sent from my iPhone

> On Jul 14, 2014, at 2:15 PM, "Balaji, Pavan" <balaji at anl.gov> wrote:
> P0:
>    Win_lock(P1)
>    Put(X, 1)
>    Win_unlock(P1)
>    MPI_Send(P1)
> P1:
>    MPI_Recv(P0)
>    assert(X =3D=3D 1)
> If the above is correct, shouldn=92t there be a memory read barrier on P1 s=
> omewhere?  Since P1 is not making any RMA calls, I=92d assume that=92ll nee=
> d to somehow come from the lock and unlock operations.  That is, the MPI im=
> plementation will need to do an active message in Win_lock and Win_unlock f=
> orcing a memory barrier at the target.  Assuming that=92s correct, I'll hav=
> e to send out a lock packet even if the user gave the MPI_MODE_NOCHECK hint=
> , for memory consistency reasons.  That sounds awful, so I=92m really hopin=
> g that I missed something in the standard which will say I don=92t need to =
> do all this.
> Note that all this active message problem will go away if I need P1 to do a=
> lock/unlock to itself in order to access X.
> Thanks,
>  =97 Pavan
> _______________________________________________
> mpiwg-rma mailing list
> mpiwg-rma at lists.mpi-forum.org
> http://lists.mpi-forum.org/mailman/listinfo.cgi/mpiwg-rma

More information about the mpiwg-rma mailing list