Minutes_2021_01_05

Numba Meeting: 2021-01-05

Attendees: Alex, Graham, Guilherme, Ivan, Jim, Luk, Siu, Stu, Val, Dale, Todd

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

Python 3.9 update
- mostly working now
- stencil test skipped on 3.9
Siu discuss bytecode changes in recent Python versions
Actions:
- Ask Python core devs for early warnings?
- Write PEP/start conversation about the impact of bytecode changes.
- Ask for switches:
  - AST on function
  - Simple bytecode output
- PEP 638, macro pep: https://www.python.org/dev/peps/pep-0638/
0.53 RC schedule
- RC1 end of Jan
- Other things to get in too:
  - profiler patches
  - cuda dispatcher patches
0.54 (May-ish)
- Plan to do a bunch of refactoring (module level, not repo level)
- Idea is to help with multiple hardware target support and make the current extension API work for multiple targets.
LLVM 11 timeline
- IBM folks fixed issues with ELF TOC
- attempt 0.54
- ARM bug in instruction selection won't be fixed. Still require old LLVM on ARM https://bugs.llvm.org/show_bug.cgi?id=46996
- ACTION: stuart, update tickets about it.
- Ivan: SVML patch is already done, it's up on conda-forge.
Discussion topic(s) for next week?
- Generic jit classes (jit class extensions, bit more like C++ templates and using Python typing annotations).
- Changes for target-specific overload

week 3

#6615 - FakeCUDAKernel doesn't take "debug"
- looks like sth wrong with the import logic
#6613 - Wrong result when call numba function which returns structured array inside of other numba function
- An old problem, that has not been reported yet.
#6611 - Feature Request: Support ddof argument in np.nan[std | var]
- Good first issue
#6607 - <unnamed> (81, 19): parse expected comma after load's type
**** #6606 - [ppc64le] RuntimeDyldELF/TOC problem (segfaults)
#6605 - Jetson TX2 numba installing

week 2

#6604 - Segmentation fault on ubuntu 20
- Closing as no new information has been provided
#6603 - Parallel loop unexpected overhead
#6601 - get_numba_envvar doesn't seem to return the right value
#6600 - Numba accepts floats as array index
#6597 - Parallel loop prange gives wrong outcomes
- fixed by #6602
#6596 - error during np_append which is implemented in numba.np.arrayobj.np_append
#6595 - LoweringError: Failed in nopython mode pipeline (step: nopython mode backend)
#6592 - Fix python -We support wrt the threading backend loading sequence.

week 1

#6616 - Use of unsupported opcode (IS_OP) found on Python 3.9.1
#6614 - Use of unsupported opcode (LIST_EXTEND) found
#6612 - numba.errors.TypingError: Failed in nopython mode pipeline
#6610 - Feature Request: axis argument in np.nan[sum | mean | std | var | max | min | median]
#6594 - LoweringError: Failed in nopython mode pipeline (step: nopython mode backend)
#6590 - test_inline_asm failing on linux-32
#6585 - Unexpected behaviour of and in guvectorize returned from factory function (bad specialisation of bool as int64?)
#6583 - from numba import * fails due to jitclass still being in __all__

week 3

#6609 - Recommend miniforge instead of c4aarch64
- Val might deploy this to farm this week
#6608 - Support f-strings for common cases

week 2

week 1