Minutes_2022_08_02

Numba Meeting: 2022-08-02

Attendees: Alexander Kalistratov, Graham Markall, Sebastian Ernst, Todd A. Anderson, Siu Kwan Lam, Val, Benjamin Graham, Andre Masella, LI Da, Nicholas Riasanovsky, Shannon Quinn, Guilherme Leobas

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

Please refer to this calendar for the next meeting date.

0. Discussion

• Python 3.11 frontend rework (Val)
  • https://github.com/numba/numba-hatchery
  • https://github.com/numba/numba/issues/8304
  • jittery live demo
  • Link to paper being implemented: https://dl.acm.org/doi/pdf/10.1145/2693261
• Dropping support for old NumPy versions: https://github.com/numba/numba/pull/8283#issuecomment-1197505150
  • Only 1.19 and upwards?
  • Need to check on the platform support
• https://numba.discourse.group/t/returning-an-array-of-3d-or-6d-vectors-from-guvectorize-on-cuda-cross-compiling-for-cpu-cuda/1473 (Sebastian)
• (suggest topic) - (name)

1. New Issues

• #8286 - Bytes type fails len when parallel
• #8291 - parallel=True in nopython mode seems to make memory leak
• #8296 - Add citation support
• #8298 - Function compilation not being updated for specific types on 32-bit systems
  • Can potentially be fixed easily at the typeof level.
• #8300 - No implementation of CUDA shared.array error
• #8301 - Worst case performance of njitted any vs. plain numpy any
  • Numba is faster when the input is not just zeros
  • np.any version is vectorized?
  • Check the assembly of both functions
• #8303 - Allow multiple outputs for guvectorize on CUDA target
• #8304 - Python 3.11

llvmlite:

• #868 - Apple M1, poetry and tensorflow

Closed Issues

• #8288 - main failing in parfors testing as of 6f70229
• #8293 - Numba appears ~10X slower running in docker versus native on OSX

2. New PRs

• #8287 - Drop CUDA 10.2
• #8290 - CUDA: Replace use of deprecated NVVM IR features, questionable constructs
• #8294 - CUDA: Add trig ufunc support
• #8295 - Add get_const_mem_size method to Dispatcher
• #8297 - Add __name__ attribute to CUDAUFuncDispatcher and test case
• #8299 - Fix build for mingw toolchain
• #8302 - CUDA: Revert numba_nvvm intrinsic name workaround

llvmlite:

• #869 - bump max Python version to 3.11

Closed PRs

• merged - #8289 - Revert #8265.
• merged - #8292 - update checklist

llvmlite:

• merged - #866 - Cherry pick for 0.39 release
• merged - #867 - Update CHANGE_LOG 0.39.0 final.

3. Next Release: Version 0.57.0/0.40.0, RC Jan 2023