PR #23136 and #22889 created fast paths through ufunc.at for 1d aligned arrays with no casting. I closed issues #11156, #7998, #5922, and #8495, as mentioned in this comment and opened this to continue the discussion around additional speedups and concerns:

• 2d and more contiguous arrays could be sped up by calling the indexed loop on the fastest dimension. Currently the indexed loop will only be called for 1d arrays
• Add documentation on how user-defined loops can utilize this. The PR added a test that uses the experimental DType API to create a ufunc type with both a strided loop and an indexed loop. Is that the desired API?
• https://github.com/ml31415/numpy-groupies has a number of specific ufunc-at loops. Is there more we could integrate from there? @nschloe thoughts?
• Are there still cases where bincount is faster than ufunc.add.at?
• Feature request: Incremental bincount and axis argument #9397 asks for an axis argument to bincount. ufunc.at does not support the axis kwarg, so I left that issue open.