Minutes_2024_03_19

Numba Meeting: 2024-03-19

Attendees: FPOC (last week): Val FPOC (incoming): Kaustubh

NOTE: All communication is subject to the Numba Code of Conduct.

Please refer to this calendar for the next meeting date.

Numba user survey is now closed. We are evaluating the data and working on an interpretation. Hoping to get all of these published by Friday.
NumPy 2.0 :: CUDA array construction: Is copy=None a good solution to the changed NumPy behaviour.
0.59.1
- conda packages done
- wheel stuck with changes on pypi.org
NumPy 2.0 current status
- Binary compat against beta release
- Only failing on CUDA copy behaviour, fix is in place but needs to be verified
- Baseline for Numba 0.60 will be NumPy 2.x -- but should work on NumPy 1.x

numba#9492 - [feature request] Support arr[:, [i]]
- Fancy Indexing
numba#9493 - random.Generator.binomial infinite loop
- Seems like a precision error that causes the loop to hang
numba#9494 - [feature request] Broadcasting for np.full with array argument
numba#9496 - Incorrect rounding with uint8 and float32
- May be valid, may not be, needs debugging to be certain
- Unlikely to be float64, since that would upcast the final result
- Best guess: LLVM re-ordering leading to slightly different float results
- May be an arm64 issue, with fused add or multiply
numba#9499 - support call in jitclass
- May be a duplicate
numba#9500 - Numba.cloudpickle error, although numba installed
- Probably some issues with pyinstaller, path issues and stuff not being found
numba#9501 - Bad interaction (hang) involving TBB and subprocess fork and maybe MKL OMP

numba#9495 - Schedule guvectorize'd functions over dask.distributed
- Adding to @guilhermeleobas review queue
numba#9498 - median raises assertion error for an empty array
- @kc611 will review
llvmlite#1035 - Support building against llvm15
- @sklam will look into this