Evgeni Burovski

Thank you for the list! > Also LinearNDInterpolator is much faster https://github.com/cupy/cupy/pull/7864 should bring a perf improvement (at least, way better scaling with the dataset size).

I'm going to work towards this, starting with the python api, NEP 52.

Is this still relevant @matthew-brett ? I need to make DoctestParser pluggable in scipy-doctest, and started with OutputChecker since it's easier: https://github.com/ev-br/scpdt/pull/141

Not sure how to do it with pytest-doctestplus either. Do you feel like giving an alternative a go? If you do, we can take it to the tracker https://github.com/ev-br/scpdt ---...

In this specific case, rerunning the full set of calculations might not be very practical, since it's going to require a non-negligible cpu time on a cluster. Not sure what...

The largest runs in the supplement repository are some 120 CPU hours on 24 cores. Multiply it by about 1.5-2 for thermalization. Smaller system sizes are *much* faster, some 4-10...

Anything needed from me at this stage?

It certainly does. https://github.com/ev-br/10yr_repro_challenge_35/commit/837038f2461ee5e17da895388bda935749918e0c is the relevant makefile. I then simply commented it out when transferring to cluster/intel compiler instead of adding platform detection (more brittle stuff to debug ten...

OK, this starts making sense. Locally I'm down to a handful of failures in jit compilation and fusion, and a first sanity review would be great. Also might be a...

Progress: `@fuse` reuses the ufunc machinery for type promotion. `core_tests/fusion_tests` all pass, at least locally.