Skip to content
Valentin Haenel edited this page Mar 30, 2020 · 1 revision

Numba Meeting: 2019-12-10

Attendees: Graham, Pearu, Ehsan, Todd, Stuart, Siu, Stan, Val, Aaron, James

0. Feature Discussion

  • from future import disable reflected list PR
    • ready for review, PRS: #4883 #4937 and #4941
  • Functions as first class types progress
    • Added FunctionType
    • Can't call them yet.
    • working on lowering of calls
  • #4903: Multiple reductions
    • Todd will be opening a PR to fix this
  • Should we have an anti-aliasing check in testing to verify that objects are not being reused in the IR.
    • Aliasing being introduced in parfor creation. (new var name = expression for numpy exp, and expression was being reused)
    • Make IR nodes immutable?
    • Stick aliasing check in the legalizer?
    • This will uncover a lot of potential issues
  • what are the requirements for refcounting in boxing functions?
    • not sure, also should add to dev docs
  • review what PRs are in and out of release

1. New issues

  • #4939 - comparing ndarray with boolean fails in lowering error
  • #4933 - Discussion: should __cuda_array_interface__ include a stream pointer?
    • out of scope, IMO
  • #4930 - Cannot create jitclass with python list
    • type spelling not obvious
  • #4928 - min/max on typed lists extremely slower than on python lists
    • user education on the cost of boxing/unboxing from interpreter in typed list. Seems to be resolved
  • #4927 - Importing numba is slow
    • easy task
  • #4926 - heapq doesn't work with numba typed lists
    • due to hardcoded C impl in CPython's heapq
    • need to fix Numba's heapq w/ typed list
  • #4925 - Make the "not" supported part of the python language in the docs more clear.
    • Siu will fix formatting to be more shouty when he updates docs for try/except support.
  • #4923 - Cannot use 1d array containing a single value as a predicate.
    • TODO:
      • allow typing to pass
      • but raise in runtime
  • #4922 - prange reduction incorrect when operation happens inside a conditional block
    • PR already opened
  • #4919 - object of type Set has no len() when using it in condition
    • See previous predicate discussion

Already Closed

  • #4936 - Numba 0.47 Python 3.8 on pypi and anaconda cloud
  • #4924 - @jit
  • #4917 - reshape() supports contiguous array only
  • #4915 - The value of the variable does not change in the while True: loops

2. New Open PRs

  • #4938 - [WIP] Improve errors / warnings for cuda vectorize decorator
  • #4937 - Support None as a type for the typed-list
  • **** #4935 - Better handling of prange with multiple reductions on the same variable.
  • #4934 - Fix fails in TestCudaWarpOperations
    • Need to add V100 to our test matrix
  • #4932 - Fixup AUTHORS file
    • Going to remove AUTHORS file as redundant to Github history
  • #4931 - Add physical limits for CC 7.0 / 7.5 to autotune
  • #4929 - [Experiment] Using env_manager.read_const(env_manager.add_const(...)) in boxers
  • #4921 - fix creating dict with JIT disabled
  • #4920 - fix creating list with JIT disabled

Already merged/closed

  • #4916 - fix Azure test failure detection on windows
  • #4918 - Release0.46
    • Closed as seems like a random PR

4. Next Release: Version 0.47.0, RC=December 19

  • CPython 3.8

  • Requests for 0.47 (last release for the year) - jitclass performance issues - llvm 9 trial - CTK libcudadevrt.a - CI needs to take 50% of current time - Val & Stu already looking at this - also checking Azure CI config to avoid wasting compute time - Caching: - transitive dependency - other issues: i.e. function as argument, with objmode - distributing cache - Immutable list and deprecating reflected list - Switch to pytest (see above) - Using Numba to generate LLVM/NVVM IR for different targets https://github.com/numba/numba/issues/4546 - @overload for gpu

Clone this wiki locally