Optimize ndarray indexing #2
Description
It currently recomputes lots of stuff for very similar indices, i.e. they're all computed independently. It'd be sorta straightforward, just a little messy to do ad hoc optimization for this, just to avoid some trivially/obviously repeated math.