Minutes_2019_11_04
Attendees: Siu, Aaron, Pearu, Todd, Val, Stuart, Ehsan, James
-
Py3.8 patch #4755
- no patch release. because too much changes
- plan to release with 0.47 end-of-year release
-
pytest/CI
- pytest changes on hold (25% test throughput penalty on 4 cores compared to current test runner)
- git-diff and sliced testing port to current numba.runtests
- faster public CI turn-around; 10ish mins
-
Y2020 Q1 Spring cleaning project board:
-
bitshifting: https://github.com/numba/numba/issues/4786
- numpy:
(np.uint64(255) << 1)
->TypeError
- again, np-integer vs py-integer casting rules
- numpy:
- #4778 - Numba v0.46.0 issue when executing umap
-
#4775 - Malloc Error when checking array flag
- bug on arr.flags
-
#4773 - jitclass member array assignments do not type promote
- probably setattr not promoting
- **** #4772 - String rjust/ljust/center slow to compile
- compiler performance
- Things to look at:
- big IR?
- compiling same thing over again?
- Discussion:
- not much benefit to on-the-fly compilation
- consider shipping pre-compiled code
- #4771 - is there any method to calculate massive regression by the power of GPU?
- #4765 - Error initialising ROC
- #4764 - Adding a transposed tensor gives unexpected (wrong) result
-
#4763 - Numba 0.46.0 depends on pkg_resources but does not declare a setuptools dependency
- conda meta.yaml fix
- #4761 - Build time/pip package. Using PyPy7.2 with Python level 3.6 doesn’t means using Windows.
-
#4759 - NUMBA_DISABLE_JIT breaks @generated_jit
- another reason for coverage support
-
#4758 - convert float to string within numba function
- not easy to do
-
#4754 - Passing a Tuple of Numpy arrays to cfunc (scipy.integrate.quad)
- Siu will take a look
- structure cast
-
#4752 - Profiling only the running time of a jit-decorated function
- save compilation time
- #4774 - Can't unify return type from the following types: array(float32, 2d, C), array(float64, 2d, C)
- #4766 - how to use the jitclass without the spec parameter
- #4757 - numba-RNG in 2D threads
- #4780 - [WIP] Implement np.gcd and math.gcd
- #4779 - Make slice constructor more similar to python.
- **** #4777 - Fix transpose output array layout
- it should not need to degrade to A layout
- Siu will look
- #4776 - Fix and enable idle tests from test_array_manipulation
-
#4770 - Update
np.arange
implementation with@overload
- Need to check compile_isolated
- #4769 - [WIP] objmode complex type spelling
- #4768 - Update typeconv error message to ask for sys.executable.
- #4760 - Make type_callable return the decorated function
- **** #4756 - Add parfor support for ndarray.fill.
- Todd will simplify it a little with find_call_name
- Further clean up and move to new style in springcleaning
- **** #4755 - Python3.8 support
- **** #4753 - [WIP] pytest
- #4767 - pin to 2.7.16 for py27 on osx
- Closed #4762 - fix Python 2.7 on MacOSX build issue (second attempt)
-
CPython 3.8
-
Requests for 0.47 (last release for the year) - jitclass performance issues - llvm 9 trial - CTK libcudadevrt.a - CI needs to take 50% of current time - Val & Stu already looking at this - also checking Azure CI config to avoid wasting compute time - Caching: - transitive dependency - other issues: i.e. function as argument,
with objmode
- distributing cache - Immutable list and deprecating reflected list - Switch to pytest (see above) - Using Numba to generate LLVM/NVVM IR for different targets https://github.com/numba/numba/issues/4546 -@overload
for gpu