Minutes_2021_02_16
Valentin Haenel edited this page Feb 18, 2021
·
1 revision
Attendees:
NOTE: All communication is subject to the Numba Code of Conduct.
- 0.53.0RC... updates
- 2 bugs related to temporary assignment pass (#6715, #6721)
- TODO: depending how the fixes will be, we will determine if we have enough confidence about the pass.
- 1 bug for objmode #6668?
- 1 bug for parfors (#6726)
- 1 bug for cuda (#6719)
- Aim for RC2 next week on Thurs
- Get patches merged by Wed
- Request for testers or wait for RC2?
- 2 bugs related to temporary assignment pass (#6715, #6721)
- Performance linter (?)
- #6714
- Most folks think numba we shouldn't be a strict perf linter
- Relates to error messages, which should not be too complicated.
- JimP compares to PyPy.
- Trade-off for users of 'needing performance' vs 'wanting to understand more details (i.e. error messages)'
- Luk suggest hiding or highlighting the python stack vs numba stack
- Stuart reminds us about decl-typing
- line numbers! it's there but hard to see
- Todd ask if the execution environment is in a GUI (i.e. jupyter) or on cmdline. In GUI, it possible to customize the layout more.
- Should
thread_idbe exposed?- (out of time. defer to next week)
-
#6723 - unable to install llvm
- rpi3
-
#6721 - Regression in 0.53.0rc1 for boolean conversion
- triggered by temporary assignment change
- Revert for 0.53 branch
- Patch on mainline if there appears to be a fix with confidence
- #6719 - Compiling to PTX for a device function with debug fails on CUDA 11.2
-
#6717 - segmentation fault when importing local module that imports numba
- Wait couple of days, close if no more info
- #6715 - Regression in 0.53.0rc1 due to issue in temp variable assignment removal
- *** #6714 - support np.dot() for integer types
- What would users prefer? Things to just work, or performance linting?
- #6713 - caching + (safe) casting + recursive function -> segmentation fault
- #6712 - Option/configuration/extension for smaller exception tracebacks
- #6711 - Make Buildscripts Architecture-Independent
- #6708 - Issues with tbb and tbb-devel versions
- #6707 - Migrate from MCJIT to OrcV2
- #6706 - It seems that stencil doesn't cache the compiled result.
- #6705 - Measuring and Reducing Compilation Time
- #6704 - Cannot lower setitem on a record in a parallel loop
- #6725 - CUDA: Fix compile to PTX with debug for CUDA 11.2
- #6724 - allow use on other architectures, re #6711
- #6722 - Delete manager.py
- #6720 - Faster compile
- #6718 - doc: fix typo in rewrites.rst (extra iterates)
-
#6716 - Consider assignment lhs live if used in rhs (Fixes #6715)
- Valid fix independent of temp assignment removal patch - should go into master anyway
-
#6710 - remove hard-coded branch
- Potentially close in a few days - fix needs to be applied in downstream repos, rather than Numba repo
- #6709 - Block matplotlib in test examples