stuartarchibald
stuartarchibald
Buildfarm ID: `numba_smoketest_cuda_yaml_148`.
> Buildfarm ID: `numba_smoketest_cuda_yaml_148`. Passed.
> Buildfarm ID: `numba_smoketest_cpu_yaml_122`. This "passed", the quotes are because `numpy=1.23` tests failed (expected, it's not supported yet) and there was a single test failure in `numba.tests.test_gdb_dwarf` which is known...
Thanks for the report. This stopped working sometime between 0.53 and 0.54.
`git bisect` suggests: 98f733f0a4b33bfa87671b6e04585f9df8ca6b3f is the first bad commit.
Given the contents of 98f733f, I wonder if WAP relied on some import ordering/side effect.
This "fixes" it: ```diff diff --git a/numba/core/cpu.py b/numba/core/cpu.py --- a/numba/core/cpu.py +++ b/numba/core/cpu.py @@ -58,6 +58,17 @@ class CPUContext(BaseContext): # Initialize NRT runtime rtsys.initialize(self) + from numba.core.types import FunctionType, WrapperAddressProtocol +...
Erratic warnings (NumbaPerformanceWarning: Grid size) when using `@numba.vectorize(target = 'cuda')`
Thanks for the report. I suspect that the "erratic" location of warnings etc is related to synchronisation and buffered/interleaved std streams. @gmarkall please could you take a look at this?...
Thanks for opening the request @marioroy, I think this is possible for the OpenMP layer.