mightsleep commented on PR #11554: URL: https://github.com/apache/arrow-rs/pull/11554#issuecomment-6198711306
measured it: read_window on BitChunks is as fast as the local reader if the check is one compare (len >= 16 and b <= len - 16). get(b..)?.first_chunk()? is two compares per word, about 11 % slower, so I'll fix that here. I'd like to do the merge kernels from #11437, and they need the same read at popcount-advanced positions, with two cursors. That needs a padded read valid at any position, and without a tail copy made once it gets 2.8x slower when one side is under 16 bytes. So I'd move the reader with read and read_padded into arrow-buffer in a follow-up and use it from and_then and merge. is it ok? -- 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]
