Minutes_2021_01_12
Valentin Haenel edited this page Jan 13, 2021
·
2 revisions
Attendees: Siu, Stu, Jim P, Luk, Ivan, P. Ortmann, Alex, Siu Chi, Angelina, Keith, Val, Dipto, Guilherme.
NOTE: All communication is subject to the Numba Code of Conduct.
-
https://numba.discourse.group/t/public-meeting-tuesday-january-12-2021/423
-
Generic jitclasses
- https://github.com/numba/numba-extras/pull/3
- will resume type-annotation discussion next week
-
Target overload:
- https://github.com/numba/numba/pull/6598
- https://gist.github.com/stuartarchibald/7051191e0f0cb09ba9c2d42c7c21ced6
- will resume discussion next week
-
https://github.com/numba/llvmlite/issues/686
- Remove Wiki --> a possibility?
- Adding Milestone --> perhaps checkboxes on an issue?
- Remove projects --> a possibility?
- Adding PR template: possibly reference https://numba.readthedocs.io/en/latest/developer/contributing.html#why-is-my-pull-request-issue-seemingly-being-ignored
- By the time one creates a pull-request it may be too late to advise about this?
- Maybe it's something that can into the llvmlite FAQ of contributing.md?
- State that LLVM upgrades or modifications will take a long time to test and this will necessitate a rebuild and test of Numba and llvmlite is tightly bound to that and llvmlite and Numba compatability must be maintained.
- May need an FAQ entry about how and why llvmlite upgrades LLVM support.
- Process old PRs --> as time allows, would move to different issue to increase the chances of closing this one soonish
- Process old Issues --> as above
- Improve README and docs --> pull-requests welcome
- #6629 - [PROJECT] Provide CI for all Supported Architectures
- #6628 - Consider Using Bounties
- #6627 - Passing elements by reference to CUDA device kernel
-
#6625 - Performance issue: jitclass+typedlist more than 10 x slower than Julia
- most of the slowdown seem to come from
jitclass
- most of the slowdown seem to come from
-
#6624 - np.where fails in @vectorize
- may need another peephole optimizer (joke?)
-
#6621 - Issue with casting? of types in
set.add()
. -
#6617 - Add support for more complex signatures to guvectorize
- will need to go through NumPy, Siu will answer
- #6620 - CUDA_ERROR_UNKNOWN
-
#6626 - A simple event system to broadcast compiler events
- needs a review, then a corresponding setup for llvmlite
- #6623 - Fix #6556 extend variable lifetime to block end.
- #6622 - Fix 6534, handle unpack of assign-like tuples.
- #6619 - Improved fastmath code generation for trig, log, and exp/pow.
- #6618 - Wip/pr6579 withlift
- Relase branch: mid-Jan