Charm++ parallel programming system

[Kate Isaacs <keisaacs AT ucdavis.edu>]

I haven't tested it on an AMPI application yet, but it does sound like Ravel-oriented time-space makes sense for comparison. My understanding is that scorep would trace the true calls made by the runtime, while Rafael is looking for timing on the application-level MPI calls which get translated and may show up in the Projections logs as a translated name (ampi::generic perhaps?) If these calls come with index values, they can be attributed to the application ranks. Is the goal to explore traces visually or calculate something per rank?

On Tue, Mar 17, 2015 at 2:01 PM, Phil Miller <mille121 AT illinois.edu> wrote:

This sounds like something that Kate Isaacs' chare oriented timeline view would handle quite nicely. The adaptation to show the MPI calls would seem minor in comparison to the soft in view that she's already implemented.

On Mar 17, 2015 1:36 PM, "Rafael Keller Tesser" <rktesser AT inf.ufrgs.br> wrote:

Hello Abhinav,

Projections doesn't seem to trace MPI/AMPI calls. At least, I couldn't find a way to visualize them on the visualization tool. Furthermore, it groups the traces by process instead of by MPI rank, which is not ideal for my purposes.

With Tau, I was able to see some calls to functions like ampi::generic(AmpiMsg* impl_msg)::165, but even those were grouped by process, as Tau can't see Charm's user threads as different ranks.

At minimum, I need to trace the duration of MPI calls, like MPI_Send, MPI_Recv, MPI_Wait. I also would like to pair MPI_Wait calls with the call that initiated the request (e.g. MPI_Isend, MPI_Irecv, MPI_Start).

Regards,

Rafael

2015-03-17 14:16 GMT-03:00 Abhinav Bhatele <bhatele AT illinoisalumni.org>:

Hello Rafael,

Have you considered using the tracing functionality that comes with Charm++? It generates detailed OTF-style traces for Charm++/AMPI programs.

The Projections tool can be used to visualize these traces:

http://charm.cs.illinois.edu/manuals/html/projections/manual-1p.html

- Abhinav

On Sun, Mar 15, 2015 at 4:48 PM, Rafael Keller Tesser <rktesser AT inf.ufrgs.br> wrote:

Good night!

Is there some way to trace MPI calls in AMPI applications. I'm especially interested in recording the duration of communication calls? My intention is to compare these traces with MPI traces of the same application.

I tried using a few tools, like Score-p and TAU. The problem is that they rely on PMPI to trace MPI calls, which is not implemented on AMPI.

Regards,

Rafael Tesser

_______________________________________________
charm mailing list
charm AT cs.uiuc.edu
http://lists.cs.uiuc.edu/mailman/listinfo/charm

_______________________________________________
ppl mailing list
ppl AT cs.uiuc.edu
http://lists.cs.uiuc.edu/mailman/listinfo/ppl

--

Abhinav Bhatele, people.llnl.gov/bhatele
Center for Applied Scientific Computing, Lawrence Livermore National Laboratory

_______________________________________________
charm mailing list
charm AT cs.uiuc.edu
http://lists.cs.uiuc.edu/mailman/listinfo/charm

_______________________________________________
charm mailing list
charm AT cs.uiuc.edu
http://lists.cs.uiuc.edu/mailman/listinfo/charm