Minutes_2018_07_19

Numba Meeting: 2018-07-19

Attendees: Siu, Stuart, Todd, Ehsan

1. New issues

• #3131: LoweringError: Failed at object (object mode frontend)
  • Stuart has a patch
• #3130: Getting Thread IDs in prange
  • Need discussion
  • May not make sense to our thread-scheduling
• #3129: Numpy array method "sum" with "axis" argument not supported in parallel mode.
  • Patch #3127 will make this possible
• #3123: np.nditer erroneously works on 0 size arrays bug
  • in family of issue of 0-size array
• #3121: Feature request np.fill_diagonal
• #3120: ndarray.flat does not support setitem bug
• #3119: min/max behaves incorrectly on 0-size array bug
  • in family of issue of 0-size array
• #3118: [Question]List comprehension compile failed
  • Similar to #3131
• #3117: Strange behavior of flat/ravel when combined with nopython/parallel
  • many related issues in other tickets
• #3115: np.broadcast support
  • feature request
• #3114: Add np.asarray support
  • feature request
• #3113: function defined with @guvectorize target='cuda' but not used (and no NVIDIA card) gives error
• #3112: Bug: numba does not preserve dtype in scalar multiplications
  • Issue w/ promotion rules.
  • sklam should take a look
• #3111: Feature request. Iterating by dimension over ND array
• #3110: Select @vectorize and @guvectorize target at call time, not decoration time
  • feature request. need deep refactoring in ufunc machinery
• #3107: List supported architectures / hardware / OS in docs
  • doc change
• #3106: Accommodate proposed changes in CUDA toolkit for windows
  • package issue
• #3105: NUMBA_DUMP_ Environment Variables not Respected
  • need to review (used to work?)
  • check on x86
• #3103: parfors lowering error ParallelAccelerator
  • can we return a better error message?
  • see PR #3133
• #3101: numba vectorize returning list/array
  • NumPy vectorize allows returning lists
  • We should not do that in Numba vectorize
• #3100: test_deadlock_on_exception failing on CUDA testers
  • Resolved by PR
• #3099: error converting array of floats to integers
  • solvable
  • Stuart runs diagnostic and tells us what is wrong
• #3098: test_optional_unpack Heisenbug
  • very hard to reproduce (only Travis-CI)
  • Need to learn how to open live shell to Travis-CI
• #3095: Why does numba not work with this nested function?
  • Error message fixed on master
• #3092: ParallelAccelerator without threads
  • Want some kind of optimization level independent of parallelization
  • Experiment with removing gufunc, caching fails
• #3090: Issue with decorators
  • This is going to be very hard to support
• #3087: unable to use parallel for bytearray inputs
  • Type system creates some confusion
  • Siu will take a look
• #3086: support for scipy.special functions : feature request
  • Looks like someone is trying to do this as a pure Python library
  • Could also call C pointers

2. Open PRs

New

• 3134 [WIP] Cfunc x86 abi
  • Great PR, need to isolate ABI
• 3132 Adds an ~5 minute guide to Numba.
  • Stan will review
• 3128 WIP: Fix recipe for jetson tx2/ARM
  • Will merge when ready
• 3127 Support for reductions on arrays.
  • First pass review, waiting for changes
• 3124 Fix 3119, raise for 0d arrays in reductions
  • Stuart will look at review comments
• 3122 WIP: Add inliner to object mode pipeline
  • Fix closure inlining in object mode
  • Needs test and review
• 3093 [WIP] Singledispatch overload support for cuda array interface.
  • Needs review

Old

• 3046 Pairwise sum implementation.
• 3017 Add facility to support with-contexts
• #2999 Support LowLevelCallable
• #2983 [WIP] invert mapping b/w binop operators and the operator module
• #2977 Added docstring coding convention and specified line terminator to developer docs
• #2976 [WIP] Expansion of developer documentation.
• #2959 np.unique return_counts support
• #2950 Fix dispatcher to only consider contiguous-ness.
• #2942 Fix linkage nature (declspec(dllexport)) of some test functions
• #2894: [WIP] Implement jitclass default constructor arguments.
  • Need to give some feedback to contributor on approach
• #2817: [WIP] Emit LLVM optimization remarks
  • Want to refactor to not require temp file on disk

===========================

3. Feature Discussion

4. Next Release: Version 0.40, RC=Sept 3, 2018, Final=Sept 10, 2018

• Experimental python mode blocks
• Refactored threadpool interface
• AMD GPU backend
• Parallel diagnostics
• Usual collection of bug fixes