bhaskargurram-ai commented on PR #11741: URL: https://github.com/apache/arrow-rs/pull/11741#issuecomment-6475631628
@adriangb thanks for suggesting the inline and for re-running. With it on top of main (a5436347), both arrow_writer runs against the merge-base 5db01b1 agree: Overall about 3% faster (0.969 and 0.973). 99 benchmarks are more than 5% faster in both runs. The dictionary and REE groups that regressed before are now about 11% faster than base, for example list_int32_ree/default 0.83×, primitive_dictionary_1pct_cardinality/default 0.85×, float64_dictionary_1pct_cardinality/default 0.88×. Nothing is consistently slower. Only two string benchmarks reached 1.06×, in the second run only, and this PR doesn’t touch those writers. So the earlier slowdown came from intern being compiled as an out-of-line call, and inlining it fixes that. I’m happy to split the inline change into its own PR if you’d rather land it separately. Otherwise I think this is ready for another look. -- 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]
