-
Notifications
You must be signed in to change notification settings - Fork 860
WeeklyTelcon_20220419
Geoffrey Paulsen edited this page Apr 19, 2022
·
1 revision
- Dialup Info: (Do not post to public mailing list or public wiki)
- Geoffrey Paulsen (IBM)
- Jeff Squyres (Cisco)
- Austen Lauria (IBM)
- Edgar Gabriel (UoH)
- Howard Pritchard (LANL)
- Matthew Dosanjh (Sandia)
- Todd Kordenbrock (Sandia)
- William Zhang (AWS)
- Brendan Cunningham (Cornelis Networks)
- Brian Barrett (AWS)
- Harumi Kuno (HPE)
- Tommy Janjusic (nVidia)
- David Bernhold (ORNL)
- Josh Fisher (Cornelis Networks)
- Akshay Venkatesh (NVIDIA)
- Hessam Mirsadeghi (UCX/nVidia)
- Christoph Niethammer (HLRS)
- Joseph Schuchart
- Josh Hursey (IBM)
- Thomas Naughton (ORNL)
- Artem Polyakov (nVidia)
- Aurelien Bouteiller (UTK)
- Brandon Yates (Intel)
- Charles Shereda (LLNL)
- Erik Zeiske
- Geoffroy Vallee (ARM)
- George Bosilca (UTK)
- Joshua Ladd (nVidia)
- Marisa Roman (Cornelius)
- Mark Allen (IBM)
- Matias Cabral (Intel)
- Michael Heinz (Cornelis Networks)
- Nathan Hjelm (Google)
- Noah Evans (Sandia)
- Raghu Raja (AWS)
- Ralph Castain (Intel)
- Sam Gutierrez (LLNL)
- Scott Breyer (Sandia?)
- Shintaro iwasaki
- Xin Zhao (nVidia)
- 4.1.4 - Looking to do a quick turnaround to get coll_ucc component released.
- A bug, William added a check to compare fabric names.
- Thinks the check is correct, but fabric names might not be?
- Hopefully today or tomorrow he'll have it fixed.
- Want to make an 4.1.4rc1 soon.
- Mellanox is testing this internally, but not in MTT.
- If anyone else wants to test coll_ucc, need to up the priority (it's default=0)
- v5.0.0rc6 went out last week.
- srun issue opened up and giles already PRed a fix.
- Nothing new yet for an rc this week.
- Remaining major issues
- orte -> prrte docs
- Ralph has a laundry list for prrte v2.1 that sound like we need fixed for OMPI v5.0.0
- And sound like a lot of work
- Two that stand out:
- mpirun (no DVM) - issue in how resources are being allocated for anything more complex than
-np 2
- Markalle opened a prrte issue on Power9
- Need to understand this disconnect
- Bouncing data back between PMIx and PRRTE repeatedly.
- mpirun (no DVM) - issue in how resources are being allocated for anything more complex than
- opal_show help not being aggregated correctly. Possibly above?
- Java binding issue isn't a blocker, but represents a bigger disconnect between us an PRRTE.
- Brian and Jeff are working on an mpirun requirements doc.
- Will need to beef up the mpirun before the exec of prterun.
- No hook in prrte for setting up path in back end daemons
- Brian opened an Issue #10252
- Brian wrote up a first cut at it, and Jeff hasn't yet read it.
- He will paste it into the Issue later today or tomorrow.
- v5.0.0 blocker items.
- Jeff gave the packagers a heads up that we're now including the html docs and they might want to package that.
- No plan for update
- alltoallv patches - Patch went into v4.1.x that
- Reminder: Delete
master
branch in your fork. It will save you much pain in the future.- Do it NOW!
- No longer master nightly tarballs, so if MTT isn't updated, you're not running recent code.
- Howard wants to talk about Fortran.
- He's been trying to use recessitate and what Giles started to do a while ago.
- OMPI doesn't doe f08 correctly at ALL.
- Something is wrong in the Fortran->C bindings for years.
- Howard rebased a fix that Giles did
- XL Fortran compiler is very inadequate in F08.
- configury that Giles added is based on the assumption (correct one).
- If you have cscoping, and #include <ISO_Fortran_binding_support.h> something header.
- Compiler will pull this in because it's part of the compiler.
- GNU <8 doesn't use this ISO
- Jeff, Howard and Mark will meet next Thursday to talk about this.
- Howard things the bindings generations are somewhat seperate.
- Jeff swears that Rolf told him that the descriptor way, and the older way of just taking a pointer... that the descriptor way is compatible, but since the first element of the struct is a pointer, then it's compatible.
- Howard agrees with this argument, but it can be a scalar, but scalar will be passed as a pointer as well.
- Don't think it'll misbehave, but if the compiler supports descriptors we're supposed to support it.
- Reviving Giles work is good and neccesary,
- This is post v5.0.0 topic.
- He's been trying to use recessitate and what Giles started to do a while ago.
- IBM asked nVidia if they would take over PGI compiler build/testing.
- nVidia is still looking into it.
- Wiki for face to face: https://github.com/open-mpi/ompi/wiki/Meeting-2022
- Should think about schedule, location, and topics.
- Some new topics added this week. Please consider adding more topics.
- Domestic travel might be opening up next few months.