JonSnow1807 commented on issue #27279: URL: https://github.com/apache/datafusion/issues/27279#issuecomment-6830361890
@Dandandan @comphead I tried finding 2 from this issue. With 32 or fewer groups, each group gets 16 per-batch lanes (row i goes to lane i % 16), folded into the existing state at the end of update_batch. Only integer and decimal sums use it, so results are identical. Floats are unchanged. On TPC-H SF10 Q1 with one partition, accumulator time drops from 335 to 162 ms on a Ryzen 9700X (query -7.2%) and from 490 to 215 ms on an Arm GB10 core (-11.6%). With all cores it is within noise. Other TPC-H and high-cardinality GROUP BYs are unchanged. It touches the same update_batch bodies as #24815. Should I open a PR on top? -- 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] --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
