Minutes_2021_10_19

Numba Meeting: 2021-10-19

Attendees: Guilherme Leobas, Graham Markall, Jim Pivarski, stuart, Todd Anderson, Siu Kwan Lam, Valentin Haenel, Luk, Nick Riasanovsky, brandon willard

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

0. Feature Discussion

• preparing for llvmlite 0.37.1 release

  • include py3.10 support
  • blocked by problems pinning conda compiler toolchain
  • more info on https://github.com/numba/llvmlite/pull/769

• High effort CUDA PRs:

  • #7461: Support NVIDIA's CUDA Python bindings
  • #7496: CUDA: Use a single dispatcher class for all kinds of functions
  • #7255: CUDA: Support CUDA Toolkit conda packages from NVIDIA
    • Graham looking at CI setup

• Lower effort CUDA PRs:

  • #7460: Testhound/fp16 support
  • #7479: CUDA: Print format string and warn for > 32 print() args
  • #7057: Fix #7041: Add charseq registry to CUDA target
  • #6515: CUDA: Add support for the jit_module function

• Reminder for folks to early test the "hard error" feature in https://github.com/numba/numba/pull/7397

• Aesara update

  • missing feature: https://numpy.org/doc/stable/reference/generated/numpy.ufunc.at.html#numpy.ufunc.at
  • advance indexing help
  • Jim's idea: calling into numpy C code instead of lowering in Numba

1. New Issues

• #7480 - Need common util to handle Optional types in comparator
• #7482 - Failed in nopython mode pipeline len(none)
• #7485 - Validate values on assignment in numba.config
• #7486 - Error importing numba
• #7487 - Random shuffle error in structured array
• #7488 - A case where type inference does not halt.
• #7491 - support for set and arrays of str type

Closed Issues

2. New PRs

• #7479 - CUDA: Print format string and warn for > 32 print() args
• #7481 - [CI testing only] CUDA FP16 changes related to #7460
• #7483 - NumPy 1.21 support
• #7490 - CUDA: Make the device function dispatcher a _dispatcher.Dispatcher subclass

Closed PRs

• #7484 - Fixed outgoing link to nvidia documentation.
• #7489 - CUDA: Type device functions as dispatchers

3. Next Release: Version 0.55.0/0.38.0, RC=Early Nov

• Request for 0.55
  • broadcast_to https://github.com/numba/numba/pull/7119
    • driven by need for vectorize support in aesara
    • Now merged