FelixYBW commented on PR #5326: URL: https://github.com/apache/incubator-gluten/pull/5326#issuecomment-2044269942
> OK! Just out of curiosity, how much performance improvement can using uint16_t bring compared to uint32_t? It's up to queries and CPU. You may test in your case. Memory latency can be >50x of L2 cache latency. Split operation is top 5 hotspot in our TPCH test. If batch size is unlimited, it may more hurt Velox pipeline's performance. Velox is vectorized engine, basic policy is that one operator's output is used by following operator's input. So if batch size is too large it will leads to big cache miss and hurt performance. It's why we use 4K rows as default now. -- 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]
