viirya opened a new issue, #5900:
URL: https://github.com/apache/datafusion-comet/issues/5900

   Normalizing multi-entry `map<string,int>` batches currently slices each 
row's keys and invokes Arrow sorting, allocating temporary indices and prefix 
tuples for every map. The singleton optimization in #5887 avoids this work for 
one-entry maps, but multi-entry maps still pay these costs.
   
   Reuse an Arrow-compatible prefix-tuple scratch buffer across rows for 
multi-entry string-key/int-value maps, while preserving the singleton fast 
path, exact duplicate-key permutations, sliced offsets, null buffers, schema 
metadata, and unsupported-key errors. Also fill rebased offsets in bulk for 
all-empty visible batches.
   
   Matched release microbenchmarks show about 3x faster normalization for 
2–10-entry forward maps and 33–39% faster normalization for 2–50-entry maps in 
independent paired measurements. Validation covers normalization, hashing, 
combined execution, nulls, mixed cardinalities, Unicode keys, and fallback 
types.
   
   Related to #5818 and the matched benchmark work in #5822.
   


-- 
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]

Reply via email to