Minutes_2021_02_23

Numba Meeting: 2021-02-23

Attendees: Jim, Ehsan, Todd, Leo, Graham, Siu, Stu, Val

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

RC status: currently reviewing rest of patches.
- Siu and Stuart will look at the inliner/SSA patch.
- Cut RC2 ASAP!
- Numba:
  - RC2: milestone: https://github.com/numba/numba/milestone/53
- llvmlite:
  - RC2 milestone: https://github.com/numba/llvmlite/milestone/11
  - [Siu TODO] make dev tag (https://github.com/numba/numba/issues/6684)
Should thread_id be exposed?
- list design notes
  - Ehsan: wants no performance impact for serial case
  - Stuart: wants automatic use of parallel-safe append
- thread_id enables lockless approaches
- properties of thread_id?
  - integer value [0, threadmask)
IPA spelling of Numba. Does not rhyme with roomba.
In upcoming weeks, Siu will propose changes to project roles (currently just "core devs") and adding more roles; e.g. release managers, triagers, architects, code-owners.
Is the Q1 refactoring going to break anyone's projects?! (esp. those using direct lowering)
- Awkward should be ok on inspection.

#6740 - Support for advanced/fancy indexing support in the cuda target
#6739 - np.linalg.solve does not work with complex numbers in Python 3.9 + numba 0.51.2
#6738 - Objmode guard for multiply defined variables
#6737 - Docs: redefinition of stdlib functions causing invalid cross references.
#6734 - Development packages/RCs for Apple M1 (osx-arm64)

#6731 - Add CUDA-specific pipeline
#6730 - Fix Typeguard error
#6728 - [WIP] Implement IR-only pipline for CUDA (was "Pull NativeLowering pass out into pipeline")
#6727 - Avoid errors in array analysis for global tuples with non-int