Minutes_2022_11_08
esc edited this page Nov 8, 2022
·
1 revision
Attendees: FPOC (last week): Graham FPOC (incoming): Andre
NOTE: All communication is subject to the Numba Code of Conduct.
Please refer to this calendar for the next meeting date.
- 0.56.4 is out now! (Val)
- available in the usual places for conda and pip
-
view
in arrays updated for CUDA - 0.57 planned for early January
- Ctrlflgrdiff presentation (Andre)
- Question on templates (Luk)
- templates are for classes, but classes are created dynamically
- class are for each function and each instantiation is for a context
- adding to the base class is fine
- PR #8575 (Guilherme)
- cmath has change in behaviour, specific vecotrization, under complex contextual differences
- might be due to fastmath propagation? (Siu)
- may be due to LLVM inling changes to fastmath? (Stu)
- seems to be a pass manager configuration issue -- Stu will debug a reproducer
- #8563 - workqueue threading layer "safety" mechanism aliases nesting and concurrent access
-
#8564 - Make Numba's
captured_errors="new_style"
by default. - #8565 - Autodiff support in Numba
- #8567 - M1 LLVM Runtimedyld Invalid page reloc value assertion error
-
#8574 -
np.fmod
doesn't work with negative inputs - #8578 - Segfault when calling nested parallel function with caching
- #8581 - CUDA: Lineinfo doesn't work for device functions
- #8582 - CUDA: Enabling lineinfo causes many redundant moves in SASS
-
llvmlite#888 - Classes should use
__slots__
in order to reduce memory footprint. - llvmlite#889 - Valgrind reports use of uninitialized value
- #8568 - The 0.56.3 release broke poetry dependency resolution
- #8576 - 1d array, column of fortran array, with F and C contiguity, raises error for not being C-contiguous.
- #8566 - Update workqueue abort message on concurrent access.
- *** #8575 - Migrate
cmath.*
to use overload - #8577 - Document that AOT compilation of recursive functions is unsupported
- #8579 - CUDA: Add CUDA 11.8 / Hopper support and required fixes
- #8583 - Skip tests that contribute to M1 RuntimeDyLd Assertion error
- #8569 - Empty release0.56 PR for testing on gpuCI
- merged - #8570 - Release 0.56 branch: Fix overloads with
target="generic"
for CUDA - merged - #8571 - Additional update to CHANGE_LOG for 0.56.4
- merged - #8572 - CUDA: Reduce memory pressure from local memory tests
- merged - #8573 - Additional cherry-picks for 0.56.4
- merged - #8580 - adding note about doing a wheel test build prior to tagging