Minutes_2021_10_26

Numba Meeting: 2021-10-26

Attendees: Siu Kwan Lam, Val, Graham Markall, stuart, brandon willard, Guilherme Leobas, Jim Pivarski, KAUSTUBH 460, Luk, Nick Riasanovsky, Todd Anderson

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

0. Feature Discussion

• GPU CI discussion
  • Val & Graham will look into setting up GPUCI (Setting up webhook and merging pull-request).
• PR #7177: GDB improvements
  • https://github.com/numba/numba/pull/7177
  • debug experience is improved
    • can show type and memory layout
    • can repr value in its native form
    • better stepping and break on symbol; no more jump around
  • PR is almost ready; pending fixups for CUDA debuginfo due to Numba internal API drift
• Python 3.10
  • reworking with-handling analysis and transform
  • stuck on code like:

    try:
        with something:
            ...
    except ...:
        ...

1. New Issues

• #7492 - partial type inference stores the wrong type when a PHI node is involved
• #7494 - CUDA Simulator IndexError when using @jit instead of @cuda.jit(device=True) functions inside of kernels
• #7495 - nb.prange returns int64 in sequential mode but uint64 in parallel mode
  • Discussion: how can we make it more consistent?
  • Potential fix: add fields to the integer type to track positive domain and also other infos.
• #7501 - aot omitted argument
• #7503 - cannot determine Numba type of <class 'dict'>
• #7505 - cudasim breaks ufunc. missing is_cuda_array
• #7509 - astype not support datetime64
• #7510 - Nooby needs help

Closed Issues

• #7497 - macOS: Numba needs NumPy 1.20 or less
• #7502 - cuda.syncthreads() not working?
• #7504 - Numba
• #7507 - Numba is raising a NotImplementedError instead of the NumbaNotImplementedError equivalent

2. New PRs

• #7496 - CUDA: Use a single dispatcher class for all kinds of functions
• #7498 - refactor with-detection logic
• #7499 - Add build scripts for CUDA testing on gpuCI
• #7500 - Update parallel.rst
• #7506 - Enhance Flags mangling/demangling
• #7511 - [CI test only] Remove call to omp_set_nested

Closed PRs

• #7493 - Consolidate TLS stacks in target configuration
• #7508 - Replace NotImplementedError by Numba equivalent

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