Skip to content
esc edited this page Aug 9, 2023 · 1 revision

Numba Meeting: 2023-08-08

Attendees: Siu Kwan Lam, Da Li, Graham Markall, Guilherme, Jim Pivarski, Kristian, Todd A. Anderson, Val, stuart, brandon willard, Kaustubh FPOC (last week): Graham FPOC (incoming): Val

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

Please refer to this calendar for the next meeting date.

0. Discussion

  • Numba 0.58 / llvmlite 0.42 -- releases
  • NumPy 2.0
    • There's an umbrella-issue https://github.com/numpy/numpy/issues/24300 which is going to be used for announcements etc that folks can subscribe to through GH.
    • NumPy 2.0 nightly builds are available. Proposal... Numba main fork gets a feature/numpy2.0 branch and we set up CI to use the nightlies on there. Incrementally tracking ABI-level updates will make it easier to develop the rest of the 2.0 related changes as the Numba code will always compile. It will also make it possible to feedback issues to NumPy folks to potentially help them out with their efforts.
  • LPython dict performance etc. Should Numba use PIXIE to build its dict and list impls, LLVM can just use the bitcode and "see" through what is currently coming from symbols in DSOs.
    • When the time is suitable, this is a worthwhile experiment to try
  • Why do we need approvals for the release notes check for first time contributors?
    • Maybe as a result of Github Actions security thing?
  • Review / milestone assignment for #9089
  • Moving Issue #8989 / PR #8990 forward
    • Kristian: Discuss implementation of np.diag_indices() (regarding numba#9115)
      • Kristian is the Quansight intern working on Numba, welcome!!
      • Can not deduce the output type since it depends on input value
      • Graham Need the paramater to be a literal?
  • feature request: mark a branch likely

Suggestion on branch likliness from Graham:


# Idea: overload the _is_likely function with an intrinsic that branches
# with likeliness based on the condition

def _is_likely(cond):
    pass

@intrinsic
def likely_intrin(typingctx, sig):
    with cgutils.if_likely as (if_branch, else_branch):
        with if_branch:
            return contrxt.get_constant(1, bool)
        with else_branch:
            return context.get_constant(0, bool)
        
    

@overload(_is_likely)
def is_likely_ol(cond):
    def impl(cond):
        return likely_intrin(cond)
    return impl

New "Ready for Review" PRs

1. New Issues

  • numba#9116 - numpy.diff fails on datetime64[ns] array
    • Query with Sebastian on the behaviour - fix for NumPy 2.0?
  • numba#9121 - if condition test on variadic parameter tuple failed
  • numba#9122 - Numba 0.58.0-rc1 Checklist
  • llvmlite#977 - llvmlite 0.41.0-rc1 Checklist

Closed Issues

  • llvmlite#974 - Release llvmlite >= 0.41.0 via pip?
  • llvmlite#975 - numba RuntimeError: No available targets are compatible with triple "arm-unknown-linux-gnu"

2. New PRs

  • numba#9115 - Add support for np.diag_indices() and np.diag_indices_from()
  • numba#9117 - Fixed towncrier template rendering
  • numba#9118 - Add support for np.resize()
  • numba#9119 - memcpy static buffer content into newly allocated buffer
  • numba#9120 - Update conda-recipe for numba-rvsdg
  • numba#9123 - Implement most ufunc attributes and ufunc.reduce
  • llvmlite#973 - Bindings: Query incoming blocks of a phi instruction
  • llvmlite#976 - Statically link compiler-rt
    • Split into llvmdev changes and rest of the PR?
  • llvmlite#978 - Bindings: Query type kinds, derived types, and elements

Closed PRs

3. Short-term Roadmap

gantt: https://github.com/numba/numba/issues/8971

Clone this wiki locally