stuartarchibald

> The before and after measurements are done with the same dependency versions, and are after numba and sparse have already been imported - it seems that the compilation time...

It looks like this might be Numba caching related. Is the number threads in use changing between the first compilation and the cache replay?

xref: https://github.com/numba/numba/pull/7522

The specific issue to do with changing the number of threads in use between first compilation and cache replay has been fixed in https://github.com/numba/numba/pull/7625 and is in the 0.56.x release...

Thanks for the report. `llvmlite` as of version `v0.23` needs LLVM 6.0.x, which is the root cause of the runtime error reported: ``` RuntimeError: Building llvmlite requires LLVM 6.0.x. Be...

As noted by @Cat-31 manual install should be fine. FWIW I think we are only a minor bug in the Jetconda builds away from being able to ship a conda...

I don't think that there's an API exposed in `llvmlite` to obtain the CFG as a structure for further parsing/manipulation, though it's possible to view the CFG like this: 1....

xref: https://github.com/numba/numba/issues/4380 and https://github.com/numba/numba/issues/4444

Does turning down the optimisation level "fix" it?

hmmmm, strange, do the command line tools do the same (just trying to eliminate llvmlite)?