perf: short-circuit and collect_bool for IN list with column references#20694
Merged
adriangb merged 3 commits intoapache:mainfrom Mar 5, 2026
Merged
perf: short-circuit and collect_bool for IN list with column references#20694adriangb merged 3 commits intoapache:mainfrom
adriangb merged 3 commits intoapache:mainfrom
Conversation
Contributor
Author
|
run benchmark in_list |
|
🤖 Hi @zhangxffff, thanks for the request (#20694 (comment)). |
Contributor
Author
|
@Dandandan @adriangb This PR adds a short-circuit optimization that breaks early when all rows already match, and incorporates the suggestions from #20428 (BooleanBuffer::collect_bool and first-expr initialization). Would appreciate your review when you have a chance. |
Contributor
|
run benchmark in_list |
|
🤖 |
|
🤖: Benchmark completed Details
|
Contributor
🚀 |
neilconway
reviewed
Mar 4, 2026
Co-authored-by: Adrian Garcia Badaracco <1755071+adriangb@users.noreply.github.com>
Contributor
|
I'm curious if the |
Contributor
Author
Benchmark result (before vs after vs after_null_count): For nulls=20% cases: after version showed ~3-5% regressions due to calling true_count() on every iteration. after_null_count eliminates this, matching before (e.g. list=28/match=100%/nulls=20%: 100.9µs vs 104.5µs). For the in_list_cols/Utf8 cases: the benchmark implicitly contains ~20% nulls, so the |
adriangb
approved these changes
Mar 5, 2026
Contributor
|
run benchmark in_list |
|
🤖 |
|
🤖: Benchmark completed Details
|
github-merge-queue Bot
pushed a commit
that referenced
this pull request
Mar 16, 2026
## Which issue does this PR close? N/A ## Rationale for this change `BooleanBuffer::collect_bool()` is faster than `map().collect()`. Per discussion on #20694; originally suggested by @Dandandan. ## What changes are included in this PR? - Implement optimization - Add benchmark for nested struct comparison ## Are these changes tested? Yes, covered by existing tests. ## Are there any user-facing changes? No. ## AI usage Multiple AI tools were used to iterate on this PR. I have reviewed and understand the resulting code. --------- Co-authored-by: Andrew Lamb <andrew@nerdnetworks.org>
de-bgunter
pushed a commit
to de-bgunter/datafusion
that referenced
this pull request
Mar 24, 2026
…es (apache#20694) ## Which issue does this PR close? <!-- We generally require a GitHub issue to be filed for all bug fixes and enhancements and this helps us generate change logs for our releases. You can link an issue to this PR using the GitHub syntax. For example `Closes apache#123` indicates that this PR will close issue apache#123. --> - Closes apache#20428 . ## Rationale for this change Third PR in the IN list optimization series (split from apache#20428): - PR1: benchmarks (apache#20444, merged) - PR2: Arrow vectorized eq kernel (apache#20528, merged) - **PR3 (this): short-circuit, collect_bool, and first-expr initialization** <!-- Why are you proposing this change? If this is already explained clearly in the issue then this section is not needed. Explaining clearly why changes are proposed helps reviewers understand your changes and offer better suggestions for fixes. --> ## What changes are included in this PR? - **Short-circuit break**: convert `try_fold` to `for` loop; when all non-null rows are already `true`, skip remaining list items (up to 27x faster for match=100%/nulls=0%) - **`BooleanBuffer::collect_bool`**: use in `make_comparator` fallback path for nested types instead `(0..n).map().collect()` (suggested by @Dandandan in apache#20428 ) - **First-expr initialization**: evaluate the first list expression directly as the accumulator, avoiding a redundant `or_kleene(all_false, rhs)` (suggested by @Dandandan in apache#20428 ) - **Tests**: added 3 new tests covering short-circuit, short-circuit with nulls, and struct column references (make_comparator fallback path) <!-- There is no need to duplicate the description in the issue here but it is sometimes worth providing a summary of the individual changes in this PR. --> ## Are these changes tested? Yes, and add test to cover short-circuit, short-circuit with nulls, and struct column references (make_comparator fallback path) Benchmark result: ``` (zhangxffff) zhangxffff/datafusion@95d3d60664da ~/W/datafusion ((bcc52cd))> critcmp after before group after before ----- ----- ------ in_list_cols/Int32/list=28/match=0%/nulls=0% 1.02 93.8±1.80µs ? ?/sec 1.00 91.8±1.52µs ? ?/sec in_list_cols/Int32/list=28/match=0%/nulls=20% 1.03 105.3±1.95µs ? ?/sec 1.00 102.2±1.59µs ? ?/sec in_list_cols/Int32/list=28/match=100%/nulls=0% 1.00 3.4±0.07µs ? ?/sec 27.14 91.7±1.52µs ? ?/sec in_list_cols/Int32/list=28/match=100%/nulls=20% 1.07 107.7±1.91µs ? ?/sec 1.00 100.4±1.33µs ? ?/sec in_list_cols/Int32/list=28/match=50%/nulls=0% 1.00 50.1±1.15µs ? ?/sec 1.84 92.4±1.36µs ? ?/sec in_list_cols/Int32/list=28/match=50%/nulls=20% 1.05 105.1±1.49µs ? ?/sec 1.00 100.0±0.84µs ? ?/sec in_list_cols/Int32/list=3/match=0%/nulls=0% 1.00 9.9±0.17µs ? ?/sec 1.01 10.1±0.19µs ? ?/sec in_list_cols/Int32/list=3/match=0%/nulls=20% 1.02 11.0±0.18µs ? ?/sec 1.00 10.8±0.16µs ? ?/sec in_list_cols/Int32/list=3/match=100%/nulls=0% 1.00 3.3±0.06µs ? ?/sec 2.95 9.9±0.16µs ? ?/sec in_list_cols/Int32/list=3/match=100%/nulls=20% 1.01 10.9±0.19µs ? ?/sec 1.00 10.8±0.09µs ? ?/sec in_list_cols/Int32/list=3/match=50%/nulls=0% 1.00 10.0±0.17µs ? ?/sec 1.00 9.9±0.18µs ? ?/sec in_list_cols/Int32/list=3/match=50%/nulls=20% 1.05 11.3±0.24µs ? ?/sec 1.00 10.8±0.11µs ? ?/sec in_list_cols/Int32/list=8/match=0%/nulls=0% 1.02 26.7±0.58µs ? ?/sec 1.00 26.2±0.50µs ? ?/sec in_list_cols/Int32/list=8/match=0%/nulls=20% 1.04 29.6±0.57µs ? ?/sec 1.00 28.5±0.45µs ? ?/sec in_list_cols/Int32/list=8/match=100%/nulls=0% 1.00 3.4±0.05µs ? ?/sec 7.78 26.2±0.36µs ? ?/sec in_list_cols/Int32/list=8/match=100%/nulls=20% 1.05 30.0±0.65µs ? ?/sec 1.00 28.7±0.55µs ? ?/sec in_list_cols/Int32/list=8/match=50%/nulls=0% 1.03 26.7±0.59µs ? ?/sec 1.00 26.0±0.37µs ? ?/sec in_list_cols/Int32/list=8/match=50%/nulls=20% 1.04 29.9±0.57µs ? ?/sec 1.00 28.7±0.46µs ? ?/sec in_list_cols/Utf8/list=28/match=0% 1.17 155.0±2.44µs ? ?/sec 1.00 132.8±2.97µs ? ?/sec in_list_cols/Utf8/list=28/match=100% 1.02 726.6±14.54µs ? ?/sec 1.00 712.4±9.09µs ? ?/sec in_list_cols/Utf8/list=28/match=50% 1.02 1070.1±13.06µs ? ?/sec 1.00 1051.8±8.17µs ? ?/sec in_list_cols/Utf8/list=3/match=0% 1.14 16.4±0.37µs ? ?/sec 1.00 14.4±0.22µs ? ?/sec in_list_cols/Utf8/list=3/match=100% 1.02 68.0±1.29µs ? ?/sec 1.00 66.5±0.99µs ? ?/sec in_list_cols/Utf8/list=3/match=50% 1.15 107.6±2.05µs ? ?/sec 1.00 93.6±1.88µs ? ?/sec in_list_cols/Utf8/list=8/match=0% 1.16 44.0±0.61µs ? ?/sec 1.00 37.9±0.95µs ? ?/sec in_list_cols/Utf8/list=8/match=100% 1.00 190.4±2.71µs ? ?/sec 1.03 195.7±2.01µs ? ?/sec in_list_cols/Utf8/list=8/match=50% 1.03 295.9±4.45µs ? ?/sec 1.00 287.3±3.26µs ? ?/sec ``` <!-- We typically require tests for all PRs in order to: 1. Prevent the code from being accidentally broken by subsequent changes 2. Serve as another way to document the expected behavior of the code If tests are not included in your PR, please explain why (for example, are they covered by existing tests)? --> ## Are there any user-facing changes? <!-- If there are user-facing changes then we may require documentation to be updated before approving the PR. --> <!-- If there are any breaking changes to public APIs, please add the `api change` label. --> --------- Co-authored-by: Adrian Garcia Badaracco <1755071+adriangb@users.noreply.github.com>
de-bgunter
pushed a commit
to de-bgunter/datafusion
that referenced
this pull request
Mar 24, 2026
## Which issue does this PR close? N/A ## Rationale for this change `BooleanBuffer::collect_bool()` is faster than `map().collect()`. Per discussion on apache#20694; originally suggested by @Dandandan. ## What changes are included in this PR? - Implement optimization - Add benchmark for nested struct comparison ## Are these changes tested? Yes, covered by existing tests. ## Are there any user-facing changes? No. ## AI usage Multiple AI tools were used to iterate on this PR. I have reviewed and understand the resulting code. --------- Co-authored-by: Andrew Lamb <andrew@nerdnetworks.org>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
4 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Which issue does this PR close?
Rationale for this change
Third PR in the IN list optimization series (split from #20428):
What changes are included in this PR?
try_foldtoforloop; when all non-null rows are alreadytrue, skip remaining list items (up to 27x faster for match=100%/nulls=0%)BooleanBuffer::collect_bool: use inmake_comparatorfallback path for nested types instead(0..n).map().collect()(suggested by @Dandandan in perf: Optimize IN list with column references evaluation #20428 )or_kleene(all_false, rhs)(suggested by @Dandandan in perf: Optimize IN list with column references evaluation #20428 )Are these changes tested?
Yes, and add test to cover short-circuit, short-circuit with nulls, and struct column references (make_comparator fallback path)
Benchmark result:
Are there any user-facing changes?