Some matrix field broadcasting results in `Failed to compile PTX code (ptxas received signal 9)` errors
- Dominant language
- Julia
- Stars
- 117
- Forks
- 19
- Avg merge
- 3d 4h
- Merged PRs (30d)
- 41
Description
For example: https://buildkite.com/clima/climacore-ci/builds/4458#01924384-ee2a-4328-a106-c6501a46657e.
Per Tim Besard on slack:
> signal 9 comes from outside; OOM killer maybe?
I confirmed with him, this is OOM during compilation.
I'm not sure how we can fix this. I think we'll need to profile some simpler cases to see where the compiler is mostly allocating.
Contributor guide
No contributing guide indexed for this repository
Research direction
Start with the linked Buildkite failure and reproduce the matrix field broadcasting case that produces the PTX compilation error. Profile simpler broadcasting cases as suggested in the issue to identify where compilation memory grows. Done means the affected compilation no longer fails from an out-of-memory condition, with the relevant behavior verified on a representative case.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- julia
- Domain
- compilers, hpc, performance
- Issue type
- Bug
- Difficulty
- 5/5
- Estimated time
- Over a week
- Activity status
- Stale
- Clarity
- Needs clarification
- Newbie friendliness
- 35/100