Skip to content
Stan Seibert edited this page Feb 1, 2018 · 1 revision

Numba Meeting: 2018-01-25

Attendees: Ehsan, Stan, Todd, Siu

1. New/outstanding issues

  • numba#2689: Numba 0.36.2 tests fail on npy114, py27, win64
    • Sporadic error. Unable to reproduce w/ Anaconda packages.
  • numba#2699: prange fails when temporary numpy array object created in the loops
    • Looks like overhead associated with array allocation and atomic reference counts inside of loops
    • Any mitigation? (User should use random module for random scalars that don't require array allocation?)
  • Python 3.6 / Win32&Win64 failures on test_matmul_mm disappeared after numpy 1.14.0 build update from 0 -> 1?
    • Seems to have been fixed by changes to NumPy 1.14 build on Windows to correct some MKL issues.
  • New question on numba#2627: Can Numba 0.36.2 be used with llvmlite 0.22 (dev)? Will there be deadlocks, or just an unnecessary amount of extra locking?

3. Open PRs

  • numba#2703: Test runner feature: failed-first and last-failed
    • Make it easier to track down problems in big test suite
  • numba#2702: Make DummyArray do High Dimensional Reshapes
  • numba#2694: Avoid type inference failure due to a typing template rejection
    • @ehsantn to review
  • numba#2681: Allow Assigning to DeviceNDArrays
    • @sklam to re-review now that blocker has been merged
  • numba#2672: Add ascontinguousarray
    • Blocker is merged. Just need to fix conflict.
  • numba#2668: Fix empty_like from readonly array
    • Contributor may have disappeared? We should add the tests.
  • numba#2660: Support bools from cffi in nopython
    • Contributor working on improving tests.

4. Feature Discussion

  • Gufunc rewrite (aiming to start landing improvements in 0.38)
    • Next steps?
  • Allow pipeline manipulation by user #2398
    • Discuss

5. Next Release: Version 0.37, RC=Feb 7, 2018, Final=Feb 14, 2018

  • Focus on bug fixes / refactoring
  • Small features
  • Prep for more feature development in Version 0.38

6. Action Items

  • Reach out to the community about the Gufunc rewrite.
  • Pipeline refactor question.
Clone this wiki locally