YUZHEthefool commented on PR #10600:
URL: https://github.com/apache/arrow-rs/pull/10600#issuecomment-5232170191

   > are we aiming to do some optimizations on the rank function for view types?
   
   I wanted to explore, although I don't have a winning optimization yet.
   I prototyped an indexed comparator that compares the inline prefix first and 
only resolves the backing buffer on collisions. It regressed by 20–43%: 
byte_view_rank already materializes the slice descriptors once, while my 
version added a view lookup and branches to the hot comparison loop.
   The benchmarks alone are probably not compelling enough. A more promising 
approach is to reuse the strategy already proven by sort_byte_view: cache the 
raw view, or a wider local sort key, and only access the backing buffer for key 
collisions. I also plan to add a long-value/same-prefix case related to #7350, 
since random [0-400] values rarely exercise the expensive prefix-collision path.
   I'll mark this PR as draft and turn it into a perf PR only if I can 
demonstrate a clear improvement; otherwise I'll close the benchmark-only PR.


-- 
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]

Reply via email to