Minutes_2019_05_07
Siu Kwan Lam edited this page May 20, 2019
·
1 revision
Attendees: Stuart, Val, Todd
- LLVM status
- split approach: LLVM 8 for all except ppc64le
- pending decision end of this week
- nothing new on LLVM issue tracker, follow up?
- Deprecation warning status
- DeprecationWarning vs. PendingDeprecationWarning
- Anything else to deprecate?
- Numba integration test status (ready to announce to world?)
- announced today via ML list + twitter
- unittests to run on itself
- Iterator API public/backward compatible changes.
- https://github.com/numba/numba-integration-testing/pull/2#issuecomment-489128791
- https://github.com/numba/numba-integration-testing/pull/5#issuecomment-489148626
- Propose lowlevel.extending API, and keep new interface.
-
#3908 - Duplicate code with a dynamic aliasing check
- Consider option "alias={always, maybe, never}" to do {pessimistic branch, both branches, fast branch}
- Emit branch taken with print statment at runtime
- #4068 - Mention TBB on install page
- #4064 - Annotate HTML alignment broken in JupyterLab
- #4062 - Result caching in Numba
- **** #4058 - "Fatal Python error: deallocating None" at run time
- #4057 - Misleading in numba.cuda.const.array_like
- #4056 - LoweringError: Fail in nopython mode pipeline (raise/assert related, caching issue)?
- #4055 - methods significantly slower than functions
-
#4053 - np.transpose crashes when passed a tuple bug
- unusual use case, ragged arrays?
- fix overload for transpose to check types + add asarray on input value
- #4052 - Faulty Loop Unrolling for CUDA device functions
- #4051 - Improve the efficiency of generating exponential random variables.
- #4050 - cudnn interface or raw kernel CUDA
- #4046 - Formalise optional dependencies doc
- **** #4045 - Calling vectorize with parallel target inside njit
- danger of conflating parfors and
vectorize(target=parallel)
- PR from ages back to unify and extract kernel body
- danger of conflating parfors and
- #4040 - Numba namespace package in ipython bug easy
- #4038 - Expose numba -s via python -m and in Jupyter
- #4037 - Add Dataclasses support
- **** #4036 - Issue with CFFI
from_buffer
, reference to buffer data deleted too early- need to sort out refcounts
- speak to Siu
-
#4035 - Fancy Indexing for
DeviceNDArray
CUDA
- **** #4042 - prange with creation of jitclasses ParallelAccelerator
- What machine.
- Can it be reproduced for performance testing.
- #4032 - readthedocs build failure: cannot cache function: no locator available for file caching
- #4054 CUDA: Use min/max PTX Instructions
- **** #4049 [WIP] Support for string lower() method
- #4048 strip, lstrip and rstrip operations implementation
-
#4047 Ambiguous Documentation fix for guvectorize
- Ready to merge
-
#4044 rjust and center operations implementation
- Ready to merge
-
#4043 fix typo
- Ready to merge
- #4041 WIP: Deprecation warnings
- #4034 Make type.Optional str more representative
- #4039 enable njit for count method on unicode type
- LLVM 8
- fixes twine bug
- fixes common symbol bug
- Type-inferred Dict
- Caching improvements requirement gathering
- transitive dependency
- bug related to storing runtime-only info
- may punt on jitclasses
- may punt on issues of (g)ufunc
- future: pre-filled/package-able cache
- open questions:
- numba version pinning(?)
- regen cache for different numba versions
- regen at post-install
- open questions:
- Add deprecation warnings
- objmode
- list/set reflection