stuartarchibald

xref: https://github.com/conda-forge/numba-feedstock/pull/87#discussion_r784692198 when this patch is tested etc against 2021/2022, check back with CF to see if the repo metadata needs a fix based on any potential regressions/build changes etc.

Debugging with Numba modified to use the new remarks capturing interface... ``` --- !Analysis Pass: loop-vectorize Name: NonReductionValueUsedOutsideLoop DebugLoc: { File: bug.py, Line: 13, Column: 1 } Function: _ZN8__main__3foo Args:...

Thanks for the report, the above reproduces. The issue is in here: https://github.com/numba/numba/blob/88fef5b350f5b9b26b26ec963147c9dbaf3d0998/numba/np/arrayobj.py#L5642-L5653 I think the negative axis should be wrapped around before handing it to `tuple_setitem`, something like: ```diff...

> Ah, maybe this one looks right? https://github.com/numba/numba/blob/main/numba/np/arrayobj.py#L4990 > > I'm not really sure about how this should be used in the split implementation. It'll need wrapping in a `numba.extending.intrinsic`...

The test suite is currently failing. An example that seems to trigger the problem is: ```python from numba import njit import numpy as np @njit def a(): return 2 @njit...

> > Adding in SSA would probably "fix" it. > > So, DeadLoopElimination needs to run after SSA. This is because (partial) type inference needs SSA?

> > > > Adding in SSA would probably "fix" it. > > > > > > > > > So, DeadLoopElimination needs to run after SSA. > > >...

A couple of suggestions: 1. The environment variable `NUMBA_DEBUG_CACHE=1` can be set to help debug Numba's caching behaviour. When set it should show what is being stored and loaded, and...

> 1. Adding `NUMBA_DEBUG_CACHE=1` does not affect anything, not any debug output appeared > > 2. Good point, I made a basic test and it works correctly in my case....

> Hmm, seems ~1.4x slower to startup: > > With this patch > > ``` > In [1]: import clifford > > In [2]: %timeit clifford.Cl(5) > 813 ms ±...