Skip to content
Siu Kwan Lam edited this page Feb 19, 2019 · 1 revision

Attendees: Stan, Siu, Stuart, Ehsan, Todd

0. Feature Discussion

  • statsmodels

    • should a user want to use Numba to accelerate parts of some package e.g. statsmodels
    • Do we need to create a page on how to take Numba as a dependency?
      • numba @jit as an optional dependency that aliases to a nop
        • njit function bodies
      • overload packages
        • Valid: Project exits to overload function for package X OR
        • Valid: Project overloads its own functions
        • Not valid: overloading functions for some other package that is not owned
  • Developer onboarding:

    • Document how new targets are added
      • Two levels:
        • How user interact with the new target? Most of the effort is here and it is not really dependent on Numba.
        • How target code is lowered? Some changes to customize codegen and calling convention.
  • Need to start an issue talking about how to integrate with C++ (and what that even means)

1. New issues

  • #3756 - can not install 0.42.1 on Ubuntu 18.04
  • #3751 - TypeError: 'ByteCodeInst' object is not iterable
  • #3750 - Parallel Accelerator: automatically dispatch optimization for non-aliasing arrays
  • #3748 - parfors gufunc lowering of tuple as kernel arg (could be a general gufunc problem)
  • #3747 - Support for unicode/string in cfuncs
  • #3746 - Documentation clarification: Does prange produce "parallel code" if parallel=False in jit/njit?
  • #3744 - Allow callee to update imprecise type in the caller
  • #3742 - inline_closure_call._fix_nested_array can fail if array has multiple definitions
  • #3741 - Add support for Numpy 1.16
  • #3739 - Exceptions at shutdown
  • #3738 - Support using system CUDA on Ubuntu
  • #3736 - UnicodeView type
  • #3735 - docs request: How to add optional jit support to a package

Already Closed

  • #3752 - Numpy isnan failing when used multiple times in prange
  • #3745 - Low performance of jitted numpy-style code versus loop-style
  • #3737 - Numba wheel doesn't install six

2. Open PRs

  • #3757 - [WIP] better analysis based on variable dependency
    • Looking for suggestions of where this analysis could be useful in the code base
  • #3755 - Make cuda.to_device accept readonly host array
    • ready to merge
  • #3754 - Updates for llvmlite 0.28
    • need to update anaconda.org labels on llvmlite
  • #3753 - Add support for cffi structs, pointers and array
  • #3749 - DOC: Clarify when prange is different from range
  • #3743 - R&D inlining.
  • #3740 - add uintp as a valid type to the tuple operator.getitem
    • ready to merge

Old Active

See https://github.com/numba/numba/projects/10

===========================

4. Next Release: Version 0.43, RC=Feb 26?, Final=March 5?

  • Initial dictionary support
  • ???
Clone this wiki locally