Minutes_2021_07_06
Valentin Haenel edited this page Jul 7, 2021
·
1 revision
Attendees: Siu Kwan Lam, Valentin Haenel, stuart, Graham Markall, Todd A. Anderson, Brandon Willard, Guilherme
NOTE: All communication is subject to the Numba Code of Conduct.
- Numba 0.54.0rc1/llvmlite 0.37.0rc1 status
- conda artifacts went fine
- release announced
- no issue reported yet
- wheels WIP
- GCC manylinux compatability issue
- Todd:
isinstance
support- probably doable nowadays
- needs stuart's dead-branch-prune
- difficulty: "not a research project"
- Graham: make progress on https://github.com/numba/numba/pulls/asodeur
- a few of these PRs are dealing with complicated changes.
- Stuart: porting datamodel into debuginfo to improve GDB debug experience
- e.g.
print(array)
in GDB will show up as a readable struct
- e.g.
- Todd: Python 3.10
- pattern-matching
case
statements - will need early testing to see how Numba is broken
- pattern-matching
- #7164 - Numba 0.54.0 rc1 llvmlite 0.37.0 RC1 Checklist
- #7166 - LoweringError with tuple_setitem and slice
- *** #7168 - Impact of flags in function selection/recompilation
- we agreed this is a critical bug for RC2
-
#7169 - CUDA:
debug=True
conflates checking for exceptions and generating debug info - #7170 - Parallel optimization breaks results in a very bad way
- #7172 - Prune 0.54.0 RC1 Changelog
- #7174 - operator.eq with Bytes selects Numpy_rules_ufunc template's signature
- #7175 - Ufunc method incompatibility with Numpy 1.21
- #7176 - NumPy 1.21 support tracker
- #7178 - CUDA: IPC tests run and fail when IPC is unsupported
- #7179 - Support jit decorator flags in cc.export (e.g. fastmath)
- #7180 - CUDA: Managed alloc test fail on Jetson AGX Xavier
- #7183 - error_model target option not working with cfunc (and maybe other target options?)
- #7184 - Bug with multidimensional numpy index