Minutes_2021_05_25
Valentin Haenel edited this page May 26, 2021
·
1 revision
Attendees: Siu, Nick, stuart, Todd A, Graham, brandon, Val
NOTE: All communication is subject to the Numba Code of Conduct.
- 0.54 update
- llvm 11
- other things
- numba-scipy: contrib packages
- Numba MinPack: https://github.com/Nicholaswogan/NumbaMinpack
- TODO create CI build template for C or Fortran code to be used with Numba
- WIP overload_classmethod (PR #7050) and Array._allocate API
@overload_classmethod(types.Array, "_allocate") def _ol_array_allocate(cls, allocsize, dtype): # allocsize [int]: bytes to allocate # dtype [type]: data type; mostly def impl(cls, allocsize, dtype): ... return impl
- Stuart suggested to use item-count instead of num-of-bytes
- #7053 - ImportError: Numba could not be imported
- ** #7051 - conditional is not evaluated correctly if parallel=True
-
CovertLoopPass
miscompilingelse
-
- #7048 - Update contribute guide to use python 3.7 instead of python 3.6
-
#7045 - Silent variable creation
- TODO: make new issue for runtime checks in future pedantic mode
- #7043 - register_jitable -ed function not recognized when used as arg to njit function
- #7042 - wrong result with numpy dot()
- #7041 - Trouble transferring array with string values on cuda
- #7039 - Rename master to main
-
#7035 -
DEBUG_JIT
doesn't work for CUDA- MegaIng working on PR #7036 to implement this
- #7054 - TypingError: movemen is not defined
-
#7038 -
Cannot unify array(float64, 1d, C) and list(float64)<iv=None>
-
#7037 - Can't call GUfunc with another GUfunc defined with @guvectorize
- Further discussion taken place on Gitter, resolved
- #7050 - Add overload_classmethod
- #7049 - Make NumPy random module use @overload_glue
- #7047 - Support hash for numpy.datetime64
- #7046 - Bf test/7044
- #7044 - Fixes for LLVM 11
- #7040 - Add Github action to mark issues as stale
-
#7036 - Add support for debug_print on cuda
- Awaiting addition of tests by author
- Follow-up PR planned to restrict the output to specific threads
-
#7052 - Fix string support in CUDA target
- Intention / ambition to make CUDA stable on the BuildFarm again