Siu Kwan Lam
Siu Kwan Lam
Btw, the problem is happening at unboxing. Numba needs to read into the annotations in `ParamsOptional` to recognize that the second field should be a optional type.
After discussing at today's triage meeting, we agree that that proposed technique is needed for a bigger scale to support OrcJIT. There needs to be a re-architecture of the [codegen.py](https://github.com/numba/numba/blob/main/numba/core/codegen.py).
The LLVM IR is from post-optimization and LLVM is allowed to move/remove metadata as it wish. Perhaps, we should stop using metadata but instead detect the store going to `%exeinfo`...
That is interesting. I'll need to take a deeper look. So far, it seems to be a problem for running SSA after deadbranch pruning. The first version work because deadbranch...
Conclusion from triage meeting: This happens because deadbranch pruning is running before SSA. To resolve this bug, we need to make deadbranch pruning SSA aware and then move SSA above...
Hm... assertion check that I added (#2668) does not look right
Found the problem. First the '一' character is: ```python >>> np.array(['一']).tobytes() b'\x00N\x00\x00' >>> list(map(hex, np.array(['一']).tobytes())) ['0x0', '0x4e', '0x0', '0x0'] ``` The boxer for unicodecharseq has a invalid skip on null-byte...
Sample runs of the perfsuite: ```bash % NUMBA_PERFSUITE_BENCHMARK=1 python runtests.py -k perfsuite -m=1 --junit --show-timing -v skipped CUDA tests skipped CUDA tests Parallel: 1. Serial: 0 test_find_chr_next_occurance (numba.tests.perfsuite.test_perf_range.TestFindChrNextOccurance.test_find_chr_next_occurance) ... ok...
Buildfarm failed vectorizer tests on linux: ``` ====================================================================== FAIL: test_unsigned_refusal_to_vectorize (numba.tests.test_parfors.TestParforsVectorizer.test_unsigned_refusal_to_vectorize) ---------------------------------------------------------------------- Traceback (most recent call last): File "/opt/conda/envs/testenv_8d765046/lib/python3.11/site-packages/numba/tests/support.py", line 664, in test_inner func(self) File "/opt/conda/envs/testenv_8d765046/lib/python3.11/site-packages/numba/tests/test_parfors.py", line 4793, in test_unsigned_refusal_to_vectorize...
But confirmed on py3.9 as well. It looks like a bug in the array-comprehension inliner.