Minutes_2021_02_09
Valentin Haenel edited this page Feb 9, 2021
·
1 revision
Attendees: Siu, Stu, Farah, Nick, Hadia, Val, Todd, Guilherme, Ehsan, Ryan, Graham, Luk, Ivan, Juan
NOTE: All communication is subject to the Numba Code of Conduct.
-
0.53.0rc1 release status
- trouble with wheels
-
RC2
- bugs regarding passing type to
with objmode()
- bugs regarding passing type to
-
Python 3.10
- bytecode changes on
MAKE_FUNCTION
- pattern matching PEP
- bytecode changes on
-
0.54.0 "risky" items/proposed work
- A lot of work on target overloads/offload
- Making it easier for GPU/other hardware targets to be added as extensions. Synthetic targets (e.g. remote process calls via 0mq) would work under the same ideas.
@overload_glue(np.take).wrap_typing(<>)
@overload_glue(np.take).wrap_impl(<>)
@overload_glue(np.take)......
- Val: Going to try LLVM 11. See if we can move everything to 11.
- Ehsan:
- Working on compilation time and monitoring
- Error checking
- Graham: CUDA dispatcher as a compiled object.
- Luk: adding typing to typed.Dict.
- A lot of work on target overloads/offload
- #6703 - CUDA parallel testing sometimes has an issue with the driver and/or runtime API
- #6702 - CUDA: Support compare and swap on array elements other than the first
- #6701 - Add support for CUDA device selection by UUID
- #6699 - Error Profiling Code with Numba's njit decorator in Spyder
-
#6697 - Wrong dtype when using np.asarray on DeviceNDArray
- has PR to fix it
- #6690 - 'Bad token in signature' with @guvectorize
-
#6687 - setup.py does not support multiple command execution if an info command (e.g. egg_info) is one of them
- has PR to fix
- #6686 - LoweringError when using prange with types.Optional array
- #6684 - Numba 0.53.0RC1 / llvmlite 0.36.0RC1 checklist
- #6679 - with.. as construct breaks bytecode analysis and leads to an uncontrolled failure
-
#6678 - numba ufunc without explicit signature vs. xarray
- Links into discussion about
ArrayLike
- Links into discussion about
- #6677 - Guiding users and search engines to latest docs
- #6675 - Parallelize (or fastmath) some methods of a jitclass
- #6692 - Numba Installation Failing on Python 3.9.1
- #6691 - Numba installation error with Windows 10; Python 3.9.1
- #6700 - Add UUID to CUDA devices
- #6698 - Fix #6697: Wrong dtype when using np.asarray on DeviceNDArray
- #6696 - flake8: made more files flake8 compliant
- #6695 - Enable negative indexing for cuda atomic operations
- #6694 - WIP Fix #6690 'Bad token in signature' with @guvectorize where bad token is NUMBER
- #6693 - [WIP] patch to compile _devicearray.cpp with c++11
- #6689 - Fix setup.py build command detection
-
#6681 - Explicitly catch
with..as
and raise error. - #6680 - Allow getattr to work in objmode output type spec