etseidl commented on PR #11524: URL: https://github.com/apache/arrow-rs/pull/11524#issuecomment-6489309278
Sorry for the delay @rluvaton, I just wanted to make sure the benefit was worth the extra complexity. I had a long conversation with my agent to explore the odd benchmark results. TL;DR is - the hash reuse is greatly beneficial in growth heavy workloads that don't fallback - there is a slight regression for strings with len <= 7 with low cardinality - the existing benches don't really show either of these There was a suggestion to replace the `copy_from_slice` in `new_inlined` (which leads to a memcpy) with a match on string len with arms: - Length 0: zero payload. - Length 1: byte load. - Lengths 2–3: first/last two-byte loads, shifted and OR-ed. - Lengths 4–7: first/last four-byte loads, shifted and OR-ed. But that's even more complexity, and winds up causing a different regression, so likely not worth it. -- 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]
