From a220ae7f8e325d523203931941cb7996b1bb6473 Mon Sep 17 00:00:00 2001 From: Pariksheet Nanda Date: Tue, 25 Aug 2026 12:12:27 -0400 Subject: [PATCH] Add meeting notes for WG Build and Packaging --- .../minutes/2026_08_11.md | 120 ++++++++++++++++++ .../minutes/2026_08_18.md | 103 +++++++++++++++ .../minutes/2026_08_25.md | 32 +++++ .../minutes/2026_09_01.md | 55 ++++++++ 4 files changed, 310 insertions(+) create mode 100644 meeting_notes/wg-education-training/minutes/2026_08_11.md create mode 100644 meeting_notes/wg-education-training/minutes/2026_08_18.md create mode 100644 meeting_notes/wg-education-training/minutes/2026_08_25.md create mode 100644 meeting_notes/wg-education-training/minutes/2026_09_01.md diff --git a/meeting_notes/wg-education-training/minutes/2026_08_11.md b/meeting_notes/wg-education-training/minutes/2026_08_11.md new file mode 100644 index 0000000..807ddf9 --- /dev/null +++ b/meeting_notes/wg-education-training/minutes/2026_08_11.md @@ -0,0 +1,120 @@ +# Education & Training working group meeting (2026-08-11) + +## Minutes + +1. Luc: Will do the blog post because Patrick may be busy. +1. Luc: [Regarding [issue #62][issue-62],] Didn't end up doing auto-build for + release briefing because [there is] not tremendous value in rebuilding it + per Damien. [We] don't change source code [after the fact] for release + briefing, because it is static and enough maintenance to do everywhere else +1. Luc: Working on running tutorial on GPU nodes. +1. Pariksheet: I don't see SC26 Kokkos tutorial in the 5.2 release briefing + slides. +1. Luc: I don't recall them mentioning it at the release briefing either. +1. Luc: [Shared screen working on a dual A100 node with fork of + [kokkos-tutorials][kokkos-tutorials].] + Installed kokkos 5.2.0. Installed kokkos-tools 5.2.0. +1. Pariksheet: Slide 51/73: Exercise#1: logistics "Compiling for CPU". +1. Pariksheet: Missing line of `cd kokkos-tutorials/Exercises/01/Begin/` +1. Pariksheet: Not using spack at all; using kokkos repostiory. +1. Luc: Missing cmake module; loaded cmake@3.??.??. Missing gcc module; loaded + gcc@11.3.0 (using OpenMP 4.5). +1. Pariksheet: CMake output says it's using Kokkos 5.0.0!? +1. Christian: Checks out a specific version in `common.cmake`. When tutorial + is done we might have 5.3, but have for now using 5.2. "Fetch content" tied + to a specific thing. +1. Luc: Even after changing still grabs 5.0. +1. Christian: Because you're inside a specific exercise. `common.cmake` puts + it somewhere else. According to git status, it's in `../../dep`. +1. Luc: Now fetches and uses Kokkos 5.2.0. +1. Pariksheet: So if I understand correctly you don't even need to setup kokkos + beforehand because the tutorial fetches kokkos for you, correct? +1. Luc: Correct. +1. Christian: It's nice, but can't switch the Kokkos version. Might want to + add a "clean all" target to delete the dep and start over. +1. Luc: CPU exercise [01] works. +1. Luc: Did the same for ../Solution. +1. Christian: Rebuilds Kokkos because different exercises could use different + backends. +1. Luc: Solution has 72 GB bandwidth and serial had 8 GB bandwidth. +1. Luc: Let's do MI250 node. Loaded rocm@7.2.0 Went back to exercise 01 + solution. +1. Pariksheet: Edit the cpp to apply the suggestions on slide 53 to use GPU? +1. Luc: Do we have auto-detection for AMD GPUs? I think so? +1. Christian: Yes. You potentially need to use `hipcc` for the compiler. + Didn't complain here because you're using `amdclang++` compiler from your + module. +1. Luc: Lucky me; do we want to make a note of that? For CUDA `nvcc` wrapper + takes over. +1. Christian: For CUDA, works automatically as long as `nvcc` in the `PATH`. +1. Luc: `echo $CXX` gives absolute path to `amdclang++`. Build with HIP + slightly slower. Just building the solution, not the one you have to + modify. `cd build-hip/`. But crashes. +1. Christian: Yes, supposed to crash because we haven't introduced Views yet. + Accessing memory that was `malloc`'d. People trying that is a learning + experience. +1. Luc: [Confirmed replacing `std::malloc` and `std::free` with `Kokkos::` + equivalents work.] +1. Luc: [Created new git branch to save the change of using Kokkos 5.2.0 + instead of 5.0.0]. +1. Christian: Auto-detection only works when you build on a compute node. Can + remove it. +1. Pariksheet: I see it only says if autodetection fails. +1. Christian: Can keep it then but probably should say e.g. and/or update to be + more future proof by updating to Hopper 90. V100 Kokkos will be dropped in + Kokkos 6 because CUDA does not support Volta anymore. Don't show an + example that doesn't actually work with the current CUDA. +1. Pariksheet: Slide 64 of building Kokkos from source uses an old version + 4.6.01. +1. Luc: Slide 51 has Kokkos 5.1. Git commit version is 4.5.1 which is also + old. Don't know if we should update. +1. Pariksheet: These are magic strings that should be consistent across slides. +1. Christian: Not a bad idea. +1. Luc: We have 3 different versions here which is arguably not great. +1. Pariksheet: Keep version the same for building from source optional + exercise. +1. Christian: Slide 67 gcc@8.3 is not supported anymore! We are on [gcc@]10 + something for OpenMP. Need to update that. Might be worthwhile for version + numbers for that as well. CUDA, Kokkos, GCC, version numbers should be + updateable in a macro we can use throughout. +1. Luc: Don't think tutorial should repeat everything in the documentation. + Fine to point people to the documentation on how to build. +1. Pariksheet: Can also print the hash from your lua code of LuaLatex :) +1. Luc: Can complain in the PR :) +1. Christian: We have master LaTeX files somewhere. +1. Luc: As long as we have version defined correctly should be correct because + of our convention for naming things. +1. Christian: For SC still waiting to know actual date that our tutorial is on. +1. Luc: Checking website; they're good about updating their program. I don't + see anything specific. +1. Christian: I think it's not there yet. + +## Tasks for next meeting + +- [ ] Continue creating PRs for Module 1, highlighted in [issue + #162][issue-162]. +- [x] In module 1, don't show an example older GPU that doesn't actually work + with the current CUDA. (Christian) + - Added new item to [issue #162][issue-162]. +- [x] Across all modules, replace magic strings of Kokkos, CUDA, GCC versions, + etc. to be consistent across slides using LaTeX macros. (Christian) + - Added new item to [issue #162][issue-162]. +- [x] In module 1, update Kokkos [sha1] hash. (Luc, Pariksheet) + - Added new item to [issue #162][issue-162]. +- [ ] Find out the Kokkos tutorial date at the SuperComputing conference. + (Christian) + +## Next meeting + +2026-08-18 + +## Attendees + +- Christian Trott +- Luc Berger-Vergiat +- Pariksheet Nanda + + +[issue-62]: +[kokkos-tutorials]: +[issue-162]: diff --git a/meeting_notes/wg-education-training/minutes/2026_08_18.md b/meeting_notes/wg-education-training/minutes/2026_08_18.md new file mode 100644 index 0000000..c54a680 --- /dev/null +++ b/meeting_notes/wg-education-training/minutes/2026_08_18.md @@ -0,0 +1,103 @@ +# Education & Training working group meeting (2026-08-18) + +## Minutes + +1. Christian: Date set for SC materials being read? Found slides are due + Oct 26. No announcement for actual date [when the tutorial is being run]. + Could create event page [on Kokkos website?] and say date TBD. +1. Damien: Do it under events. Find all the events we want on the blog. There + is an events boolean you can set. For this very thing. +1. Christian: [Editing `kokkos.github.io` repository + `content/community/events/2026-11-15-sc26-tutorial.md`]. +1. Damien: If you put it under events you don't need to do anything. Wrote + short code to take everything in events and look in blogs for boolean. We + use the date when the event [...]. The name of the [Markdown] file is + arbitrary to sort the [Markdown] file in the directory. In theory we could + differentiate the date posted with the date of the event, but I don't see a + huge value in it. +1. Christian: [Continued writing blog post for SC26 including a schedule with + content]. +1. Pariksheet: Direct messaged Patrick on [Slack Kokkos][slack-kokkos] + to see if he can tell us anything about the SC26 tutorial planning for the + blog post. I remember Patrick saying we would do modules 1--3. +1. Christian: That's about right. +1. Pariksheet: Mention Nvidia Brev cloud accounts for everyone attending and + signing up? +1. Christian: Absolutely. +1. Christian: Can I see the SC submission? On `submission.supercomputing.org` +1. Damien: Check that whether you received one or two e-mails for the + submissions; I don't remember if we were listed on the tutorial. +1. Christian: We might not have been. +1. Luc: Paul wrote the proposal more or less alone, but he's not going to + attend. Me, Patrick, you, Hari are running the tutorial. +1. Christian: I don't have access to the proposal! +1. Luc: Yes, you don't! I can forward you stuff. +1. Christian: Trying to match the blog post to the what we actually submitted. +1. Pariksheet: Patrick replied and will be out for the next 2 weeks hosting a + visitor. Can provide details later today. But Luc said he can provide + details in the mean time. +1. Luc: [Sent details to Christian.] +1. Christian: We have the whole day, right? +1. Luc: Yes +1. Christian: That's what I hoped for. +1. Luc: Hopefully people understand the schedule depends on how fast we can go + and how people keep up. +1. Pariksheet: Does blog post alert users with an RSS feed? +1. Damien: It does not. +1. Luc: Could post on the mailing list. +1. Christian: Do we need a picture [forthe blog post]. +1. Damien: Generally good to have. +1. Pariksheet: Use the SC26 logo? +1. Christian: [Looked in `../../../assets/img/blog/`]. +1. Luc: Could copy the image into our thing [repository]. +1. Damien: For now, put the file in the PR and link in the description to see + if we can point to the SVG. I'll try to see if we can make it work; doesn't + seem like a very good use of your time. Also different support for the + newer version (or maybe the same?). +1. Christian: [Push git branch `sc26-tutorial-event`]. +1. Damien: [Advised details of formatting the markdown header, etc. for the + blog post]. I'll fiddle with it. +1. Christian: [Pushed branch and opened [PR# 285][pr-285].] +1. Pariksheet: Should we prioritize what we should do to get done in time for + SC? +1. Christian: Absolutely. Was just looking at that. +1. Christian: [Looked at Module 1] +1. Pariksheet: Will do cost of coding after applying your comments to my + lambdas PR. +1. Christian: [Looked at Module 2]. Will do slides 4/60, 8/60. Will start on + Module 2. Will try to get a lot next week. Let's get through Module 1 and + 2 ASAP. Then Module 3. We can setup the second top-level file with the + correct subsections. +1. Christian: Create new top-level issue for the SC26 tutorial. +1. Christian: If you could work on Module 1 I can work on Module 2. Sound fair? +1. Luc: Yes. +1. Pariksheet: Any preferred reviewer for Module 1 PRs? +1. Christian: Doesn't matter because we will do another walk-through [looking + at the PDF slides]. See you guys! + +## Tasks for next meeting + +- [ ] Continue creating PRs for Module 1, highlighted in [issue + #162][issue-162]. +- [x] Added Christian's self-assignments for Module 2, highlighted in [issue + #166][issue-166]. +- [ ] Create new top-level kokkos/kokkos-tutorials issue for priority work for + SC26 tutorial that needs to be finished and submitted by 2026-10-26. + (Christian) + +## Next meeting + +2026-08-25 + +## Attendees + +- Christian Trott +- Damien Lebrun-Grandie +- Luc Berger-Vergiat +- Pariksheet Nanda + + +[slack-kokkos]: +[pr-285]: +[issue-162]: +[issue-166]: diff --git a/meeting_notes/wg-education-training/minutes/2026_08_25.md b/meeting_notes/wg-education-training/minutes/2026_08_25.md new file mode 100644 index 0000000..71f01df --- /dev/null +++ b/meeting_notes/wg-education-training/minutes/2026_08_25.md @@ -0,0 +1,32 @@ +# Education & Training working group meeting (2026-08-25) + +## Minutes + +1. Pariksheet: [Continued working on [PR# 174][pr-174] to incorporate + Christian's detailed feedback from our [working group meeting on + 2026-07-28][meeting-2026-07-28]]. + +## Tasks for next meeting + +- [ ] Continue creating PRs for Modules 1–2 (highlighted in issues + [#162][issue-162] and [#166][issue-166], respectively). +- [ ] Create new top-level kokkos/kokkos-tutorials issue for priority work for + SC26 tutorial that needs to be finished and submitted by 2026-10-26. + (Christian) + - [ ] Also need to create an issue for updating Module 3. (Pariksheet) +- [ ] Find out the Kokkos tutorial date at the SuperComputing + conference. (Luc) + +## Next meeting + +2026-09-01 + +## Attendees + +- Pariksheet Nanda + + +[pr-174]: +[meeting-2026-07-28]: +[issue-162]: +[issue-166]: diff --git a/meeting_notes/wg-education-training/minutes/2026_09_01.md b/meeting_notes/wg-education-training/minutes/2026_09_01.md new file mode 100644 index 0000000..551ed30 --- /dev/null +++ b/meeting_notes/wg-education-training/minutes/2026_09_01.md @@ -0,0 +1,55 @@ +# Education & Training working group meeting (2026-09-01) + +## Minutes + +1. Damien: Christian [is] on vacation starting today per [the] development + notes. Luc [may be] on vacation? The link to people on vacation is in the + [kokkos-core meeting community notes][vacation-community]. +1. Rahul: What was the slide number for updating performance plots I was asked + to work on? +1. Patrick: It's the "Amdahl's Law" slide in [issue #162][issue-162]. If you + could do that, it would be super helpful because we would want to have that + for SC. +1. Rahul: I'll create that PR this week. +1. Patrick: Next week can finish open next week. Can meet Damien this week. + For now, we can focus on our assigned PRs. +1. Pariksheet: Possible to help at SC for the Kokkos Tutorial? I won't be able + to make HPSF this year because it's in Canada. +1. Patrick: Postdocs don't qualify as a student member. Students can attend + tutorials only for $200. With tutorials and week attendance it's $1600. Not + worth just for the 1 day tutorial; SC more useful for networking and jobs. + We could add you for next year's tutorial. +1. Pariksheet: Checking to see any plans to update module 3. No, Luc said in + our [2026-07-14 WG meeting][meeting-2026-07-14] that would be too ambitious. +1. Pariksheet: Working on revising [PR #174][pr-174] with Christian's feedback. +1. Patrick: Finished updating [PR #175][pr-175]. +1. Pariksheet: [Glances at [PR #175][pr-175]]. AWS and HPE still showing up as + premier members when they shouldn't in the GitHub PR image diff of commit + 19fd11a10de87b7342c8247ffcc5f670933955ba. +1. Patrick: I will fix it, thank you. + +## Tasks for next meeting + +- [ ] Patrick and Damien to meet separately this week about the priority list of + items of modules 1 and 2 to complete ahead of the SC26 Kokkos Tutorial; + Christian mentioned at last meeting he wanted this SC26 priority list in + another issue separate from the otutorial long term plan [issue + #164][issue-164]. + +## Next meeting + +2026-09-08 + +## Attendees + +- Damien Lebrun-Grandie +- Rahul Gayatri +- Patrick Diehl +- Pariksheet Nanda + +[vacation-community]: +[issue-162]: +[meeting-2026-07-14]: +[pr-174]: +[pr-175]: +[issue-164]: