Siu Kwan Lam
Siu Kwan Lam
Thanks for summarizing the issue with array layout. I wanted to give it a little more context. The issue here shows in **typing** that Numba does not preserve contiguousness information....
Note: we'll need to fix the array layout to handle 1D arrays that are C and F ordered at the same time.
Here's the traceback on the second time `.get_call_template()` is called: ```pytb /path/to/numba/core/typeinfer.py(589)resolve() -> sig = typeinfer.resolve_call(fnty, pos_args, kw_args) /path/to/numba/core/typeinfer.py(1541)resolve_call() -> return self.context.resolve_function_type(fnty, pos_args, kw_args) /path/to/numba/core/typing/context.py(195)resolve_function_type() -> res = self._resolve_user_function_type(func, args,...
The solution is likely requiring https://github.com/numba/numba/blob/ab1a955836b63ed2695e1d16ebc846afb16364f6/numba/core/dispatcher.py#L377 to consider the flags in `ConfigStack` and compare that against `self.targetoptions`. If they are different, it needs to compile a new version. The new...
Relates to https://github.com/numba/numba/pull/9403. We need to fix dispatcher to respect compiler flags.
#6977 would just give a better error: ``` Failed in nopython mode pipeline (step: native lowering) pickling of is disabled ```
When I read the description, my first thought is that it might increase the number of unnecessary cast to `float64` because of numpy semantic on mixing signed and unsigned 64-bit...
BFID: `llvmlite_275`
`llvmlite_275` passed
Also passed buildfarm test against numba main (`numba_yaml_448`)