tustvold commented on PR #4701: URL: https://github.com/apache/arrow-rs/pull/4701#issuecomment-1680337340
For the case of dictionaries it is likely because the process of normalizing the keys allows for unchecked indexing, which in turn helps LLVM optimise the code properly as there isn't a branch in the main loop. For the case of primitives I don't have a very good intuition, however, these kernels historically have been extremely sensitive to inlining. It is possible that collocating more of the code allows LLVM to better optimise the code... I haven't looked in too much detail -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
