stuartarchibald

> > In the interests of making sure that literal_unroll is doing what is intended, I'd probably write the above referenced code like: > > With this code change we...

> Update: I now understand that there are three work-arounds required > > * one casting True to int > > * changing the import > > * moving the...

> @stuartarchibald We'll go through things with a debugger next week, hopefully we can locate the problem a bit better. Great, thanks. That the reduced MWR wouldn't even compile is...

https://github.com/numba/numba/pull/8219 fixes the compiler pass that handles `literal_unroll` to now also understand `numba.literal_unroll`.

Thanks for taking a look at this @apmasell. This seems a bit strange in general, am wondering if there's some undefined behaviour in LLVM that is being bypassed in the...

Thanks for the report. This is known behaviour, in translating from a dynamic to a static representation of the function Numba creates "slots" to store all variables and zero intialises...

I've manually verified this patch on real hardware, including testing against the original reproducer in #8309.

Thanks for running this through the farm @esc, am wondering if it ought to also have CPU run too as this patch impacts both CPU and CUDA?

> > Thanks for running this through the farm @esc, am wondering if it ought to also have CPU run too as this patch impacts both CPU and CUDA? >...

Buildfarm ID: `numba_smoketest_cpu_yaml_122`.