[mvapich-discuss] mpi_probe hangs on PSM

[Hari Subramoni]

Hello Dr. Low,

Can you please try this patch and let me know if it works fine?

diff --git a/src/mpid/ch3/channels/psm/src/psm_queue.c
b/src/mpid/ch3/chindex a66a9ed..d574ab9 100644
--- a/src/mpid/ch3/channels/psm/src/psm_queue.c
+++ b/src/mpid/ch3/channels/psm/src/psm_queue.c
@@ -314,10 +314,10 @@ PSM_ERROR_T psm_probe(int src, int tag, int contex
  #else
         rtag = 0;
         rtagsel = MQ_TAGSEL_ALL;
-        if(unlikely(tag == MPI_ANY_TAG))
-            rtagsel = rtagsel & MQ_TAGSEL_ANY_TAG;
         if(unlikely(src == MPI_ANY_SOURCE))
             rtagsel = MQ_TAGSEL_ANY_SOURCE;
+        if(unlikely(tag == MPI_ANY_TAG))
+            rtagsel = rtagsel & MQ_TAGSEL_ANY_TAG;
     #endif

     MAKE_PSM_SELECTOR(rtag, context, tag, src);

Regards,
Hari.

On Mar 23, 2017 5:09 PM, "Low, John J." <jlow at mcs.anl.gov> wrote:

> To Whom It May Concern:
>
>
>
> I have noticed that MPI_PROBE hangs on the LCRC cluster blues (
> https://www.lcrc.anl.gov/systems/resources/blues/) with mvapich2 and
> openmpi but not mpich for a few applications like parallel-HDF5 and the
> attached example.
>
>
>
> The attached example hangs when linked with mvapich2-v2.2 and
> openmpi-2.0.2 and not mpich-3.2 on blues.  Blues is a sandybridge cluster
> connected with PSM/InfiniPath. All of these libraries were compiled with
> intel-17.0.2.
>
>
>
> The example ran fine on the JLSE (http://www.jlse.anl.gov/) login node
> linked with intel-mpi, which are sandybridge nodes connected by Mellanox
> Infiniband.
>
>
>
> I suspect there is an issue with the PSM interface and MPI_PROBE, because
> the test runs fine when not using the PSM interface.
>
>
>
> Could someone explain what is happening in this case and if there is patch
> or workaround for this issue?
>
>
>
> John J. Low
>
> Principal Computational Science Specialist
>
> Computing, Environment and Life Sciences
>
> Building 240, 2143
>
> 9700 South Cass Avenue
>
> Argonne National Laboratory
>
> Argonne, IL 60439.
>
> 630-252-0045 <(630)%20252-0045>
>
> www.linkedin.com/pub/john-low/15/8b0/5aa/
>
> _______________________________________________
> mvapich-discuss mailing list
> mvapich-discuss at cse.ohio-state.edu
> http://mailman.cse.ohio-state.edu/mailman/listinfo/mvapich-discuss
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.cse.ohio-state.edu/pipermail/mvapich-discuss/attachments/20170323/8d29de9b/attachment-0001.html>