Minutes_2019_11_19

Numba Meeting: 2019-11-19

Attendees: Stuart, Aaron, Val, Todd, Siu, Stan, James

0. Feature Discussion

• Python3.8 status
  • #4755 near complete
  • #4854 the looplifting
  • updating tuple hash still TBD
• boundchecking
  • https://github.com/numba/numba/pull/4432
  • making sure CI passes
  • disable boundschecking on CUDA
  • make cuda.jit(debug=True) not turn on boundscheck
  • add internal CI config for NUMBA_BOUNDSCHECK=1
• thread masking PR
  • lean toward deoptimizing hard nested parallelism cases to ensure safety
  • Stuart will help finish out PR for workqueue backend
• mixed-type tuple iteration
  • introduced "partial" type inference
    • "partial" just means it doesn't raise
  • introduced typed_getitem
  • canonicaliz(s?)e loop pass transforms

    for x in sequence:
        body(x)

    to

    for i in range(len(sequence)):
        x = sequence[i]
        body(x)

  • may make **kwargs for mixed-type typed dict possible
• need to consolidate debug printing
  • maybe logging
  • maybe more debug level

1. New issues

• #4859 - Question regarding the performance
  • performance issue.
  • involved question
• #4857 - Cannot determine Numba type of <class 'phe.paillier.PaillierPublicKey'>
• #4853 - Cannot type types.Tuple
• #4848 - AssertionError: Failed in object mode pipeline (step: object mode frontend)
  • looplifting related
• #4846 - Error when using namedtuple in parallel loop
  • likely tuple into gufunc problem
• #4833 - CUDA malloc. Fatal Python error: Aborted
  • less likely to be numba bug at this point
• #4832 - device_array_like maintains strides?
  • half bug
• #4831 - Implement .ravel() on non-contiguous data
• #4830 - cache=True for jitclasses
  • feature request
• #4829 - Allow using .view() when itemsize does not match
  • feature request. np allows it
• #4828 - Implement .copy() for DeviceNDArray
  • feature request
• #4827 - Trouble using .ravel("A") on contiguous arrays
  • done in PR 4838
• #4820 - Default Arguments jitclass in an njit method
  • works in interpreter not in jit

Already Closed

• #4856 - it doesn't work in my RaspberryPi3
• #4822 - float16
• #4819 - Does numba's implementation of np.diff have a bug?

2. New Open PRs

• String PRs:

  • #4867 - Functionality extension str.startswith() based on CPython
  • #4866 - Support params start/end for str.find()
  • #4865 - Implement str.replace
  • #4864 - Add stub of str.format()
    • *args problem in lowering
  • #4863 - Add str.encode() only for kind 1 and utf-8
  • #4861 - Implement str.rindex() based on CPython
  • #4860 - Implement str.index() based on CPython
  • #4858 - Implement str.expandtabs() based on CPython
  • #4850 - Implement str.maketrans() based on CPython
    • problem with classmethod
  • #4849 - Implement str.splitlines() based on CPython
  • #4847 - Implement str.isascii() based on CPython
  • #4845 - Implement str.partition() based on CPython
  • #4844 - Implement str.isnumeric
  • #4843 - Implement str.isdigit
  • #4842 - Implement str.isdecimal
  • #4841 - Implement str.rpartition() based on CPython
  • #4840 - Implement str.isalpha() based on CPython
  • #4839 - Implement str.isalnum() based on CPython
  • #4837 - Implement str.isidentifier() based on CPython
  • #4836 - Implement str.isprintable() based on CPython
  • #4835 - Implement str.isspace() based on CPython
  • #4834 - Implement str.rsplit() based on CPython
  • #4825 - Implement str.swapcase() based on CPython
  • #4824 - Implement str.casefold() based on CPython
  • #4823 - Implement str.capitalize() based on CPython

• #4862 - Addition of PyCon India 2019 talk on Numba

  • ready to merge

• #4855 - support for math functions(special.erfinv, special.erfcinv, math.frex…

  • can't have scipy dependency in numba

• #4854 - Python3.8 looplifting

  • wait til python3.8 patch gets merged

• #4852 - windowing test should check equality only up to double precision errors

  • ready to merge

• #4838 - Permit ravel('A') for contig device arrays in CUDA target

• #4821 - WIP Argsort improvement

• **** #4818 - Typed list faster copy

• **** #4817 - Typed list implement and expose allocation

Already merged/closed

• #4826 - [DO NOT MERGE] Enh/py38 v2 ci
• #4851 - Add graphviz output for FunctionIR

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