Rich-T-kid commented on code in PR #11715:
URL: https://github.com/apache/arrow-rs/pull/11715#discussion_r4560638257
##########
arrow-select/src/interleave.rs:
##########
@@ -551,6 +625,10 @@ fn interleave_list<O: OffsetSizeTrait>(
// ranges, avoiding both the intermediate child_indices Vec allocation
and
// MutableArrayData's function pointer indirection.
field.data_type() => (list_primitive_helper),
+ // Short lists offer no bulk-copy advantage over the scalar gather.
+ DataType::Struct(fields) if capacity > indices.len() &&
can_interleave_struct_ranges(fields) => {
Review Comment:
This path starts being used at an average child length of 2, but the
checked-in benchmark only covers 15. I reran base against this head in both
orders: at length 2 the PR was 19–20% slower without nulls and about 52% slower
with nulls; at length 3 it was about 6% and 36–39% slower. Length 4 was roughly
the crossover. Could we raise the cutoff and keep benchmark cases around the
chosen boundary?
--
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]