Hi Martin, The idea to allow these is that I would like Tools to only intercept a few MPI calls (i.e., MPI_Init, MPI_Init_thread, MPI_Finalize) and be fully functional. This includes the possibility to allocate and free handles sometime in the middle of the execution of an MPI programm. If only Init and Finalize are intercepted tools would not have an opportunity of handle allocation/deallocation before the end of the program (at Finalize). Also discussed yesterday, we want to allow an implementation to return a specific error code (MPI_T_ERR_NOT_ACCESSIBLE) to indicate that a specific function may not be called at the time. The idea behind this is that both allocation functions are conceptionally non-blocking (as observed by the user). From the standpoint of a user, an ALLOC behaves similar to an MPI_Isend + MPI_Request_free, as the implementation will start to raise events for the handle, yet does not really know when. The FREE is actually explicitly non-blocking. In summary, in most situations, handle allocation may be postponed beyond the callback and offloaded to some other context. Still, if even the postponing is not possible, the implementation can return MPI_T_ERR_NOT_ACCESSIBLE in those cases (a tool may try again at a later time/different callback context). With this in mind. Do you think, we should extend the list of MPI_T calls also to other allocation/free functions? Cheers, Marc-Andre On 18.01.19 11:11, [email protected] wrote:
Hi Marc-Andre, all,
Thanks for the notes and sorry that I couldn’t make it yesterday.
Looking over the notes, I have one question (more out of curiosity): why include event handle alloc/free in the list of allowed calls and not the par/cvar counterparts? Do you have a particular use case in mind? Is the main goal to allow the removal of callbacks after events?
Thanks,
Martin
— Prof. Dr. Martin Schulz, Chair of Computer Architecture and Parallel Systems Department of Informatics, TU-Munich, Boltzmannstraße 3, D-85748 Garching Email: [email protected] <mailto:[email protected]>
On 18. Jan 2019, at 08:12, Marc-Andre Hermanns via mpiwg-tools <[email protected] <mailto:[email protected]>> wrote:
Hi everyone,
I put the notes of yesterdays meeting up on the wiki at: https://github.com/mpiwg-tools/tools-issues/wiki/Notes-2019-01-17
Cheers, Marc-Andre
On 17.01.19 09:10, Marc-Andre Hermanns wrote:
Hi everyone,
This is a reminder of our regularly scheduled call today January 17 at 8 am Pacific / 11 am Eastern / 5pm CET.
Webex Info: https://apps.fz-juelich.de/scalasca/mpiwg-tools-webex/index.php
*IMPORTANT*: The webex information has been updated for 2019, so be sure to get the new info.
Agenda: ------- - Events - Open Issues: CB Safety, MPI_Info parameters - Preparations for Virtual Meeting at Jan 23, 2019
Cheers, Marc-Andre
_______________________________________________ mpiwg-tools mailing list [email protected] https://lists.mpi-forum.org/mailman/listinfo/mpiwg-tools
-- Dr. Marc-Andre Hermanns Jülich Aachen Research Alliance, High Performance Computing (JARA-HPC) Jülich Supercomputing Centre (JSC)
Wilhelm-Johnen-Str. 52425 Jülich Germany
Phone: +49 2461 61 2509 | +49 241 80 24381 Fax: +49 2461 80 6 99753 www.jara.org/jara-hpc email: [email protected]
_______________________________________________ mpiwg-tools mailing list [email protected] https://lists.mpi-forum.org/mailman/listinfo/mpiwg-tools
-- Dr. Marc-Andre Hermanns Jülich Aachen Research Alliance, High Performance Computing (JARA-HPC) Jülich Supercomputing Centre (JSC) Wilhelm-Johnen-Str. 52425 Jülich Germany Phone: +49 2461 61 2509 | +49 241 80 24381 Fax: +49 2461 80 6 99753 www.jara.org/jara-hpc email: [email protected]