Hi Kari,
I have time to spare currently and I'd like to make myself useful. Anything I could do to help with the ROCm 5.7 packaging goal? Or even try to help with reaching 6.2 in time for Trixie. I checked the wiki and the help needed page listed some missing man pages which would certainly be nice but not the thing if there's a time pressure. The release plan for 5.7 says that rocm-hipamd's debug symbols issue has been deferred. Is it still a blocker for something regardless?
We could certainly use your help. The ROCm 5.7 update has been completed, so I suppose that wiki page should be archived somehow. We never did figure out why rocm-hipamd is missing debug symbols, but it's perhaps not the highest priority. The items listed on the Help Needed wiki page [2] were added in part because I thought they'd be good for new contributors and were therefore deliberately not time sensitive.
To contribute to the migration to ROCm 6.1, I would suggest (in no particular order):
1. Test with rocm-llvm 6.1.2+dfsg-1~exp1 and migrate to unstable
if appropriate.
2. Update rocthrust and hipcub from ROCm 5.7 to ROCm 6.1 or later.
3. Review and attempt to fix the RC bugs filed on rocsparse.
4. Update rocsparse from ROCm 5.7 to ROCm 6.1 or later. This will
involve a SOVERSION bump.
5. Package the rocsparse benchmark client. You can follow the same
pattern as was used for librocblas0-bench.
6. Review Xuanteng's update to rocm-smi [3] and upload if
appropriate.
To contribute to the packaging of pytorch-rocm, I would suggest:
1. Complete the packaging of hipblaslt (or summarize the
remaining work so I can pick it up).
2. Test miopen and migrate it and half to unstable if appropriate.
This is obviously a lot of work. Feel free to pick and choose
whatever you have time for.
I saw roctracer's RFS and uploaded it. I checked the sources and packaging and saw that it compiled but I didn't test it further at this time.
Thank you very much.
I have gfx1032 and gfx1036 available (not expecting everyone to remember about my Dimgrey Cavefish).
There is also a gfx90a system hosted by the Oregon Advanced Computing Institute for Science and Society that is available to you. Feel free to reach out to me privately if you need access (e.g., for testing hipblaslt) and do not yet have an account.
I'm also a bit curious if anything works on gfx1036 on Linux 6.10
or later. If you want to run the tests from rocrand1-tests or
hipsolver0-tests, I'd be curious to see the results.
Sincerely,
Cory Bloor
[2]:
https://salsa.debian.org/rocm-team/community/team-project/-/wikis/Help-needed
[3]:
https://salsa.debian.org/rocm-team/rocm-smi-lib/-/merge_requests/5