Skip to content

Apache Lucene open-source search software

Skip runs of excluded docs in ReqExclBulkScorer for two-phase clauses (#16307)

Jim Ferenczi446dda966920
diff --git a/lucene/CHANGES.txt b/lucene/CHANGES.txt
index 5d530c76bb0..d51fb3955fa 100644
--- a/lucene/CHANGES.txt
+++ b/lucene/CHANGES.txt
@@ -149,6 +149,10 @@ Improvements
Optimizations
---------------------
+* GITHUB#16307: ReqExclBulkScorer now skips runs of excluded docs via
+ TwoPhaseIterator#docIDRunEnd for two-phase excluded clauses (e.g. a doc-values
+ not-equals filter), instead of advancing one doc at a time. (Jim Ferenczi)
+
* GITHUB#16254: Enable the vectorized findNextGEQ path on aarch64, where it was disabled on cores
with fewer than 8 int lanes (Graviton2 and Graviton4). The vector compare beats the scalar
fallback by +75% on Graviton4 and +35% on Graviton2 (AdvanceBenchmark). (Shitij Bhargava)
diff --git a/lucene/core/src/java/org/apache/lucene/search/ReqExclBulkScorer.java b/lucene/core/src/java/org/apache/lucene/search/ReqExclBulkScorer.java
index da665c0dd95..42f7110cd8e 100644
--- a/lucene/core/src/java/org/apache/lucene/search/ReqExclBulkScorer.java
+++ b/lucene/core/src/java/org/apache/lucene/search/ReqExclBulkScorer.java
@@ -61,8 +61,10 @@ final class ReqExclBulkScorer extends BulkScorer {
// from upTo to docIdRunEnd() are excluded, so we scored up to docIdRunEnd()
upTo = Math.min(exclApproximation.docIDRunEnd(), max);
} else if (exclTwoPhase.matches()) {
- // upTo is excluded so we can consider that we scored up to upTo+1
- upTo += 1;
+ // upTo is excluded; skip the whole run of consecutive excluded docs at once, like the
+ // non-two-phase branch above. Clamp to at least upTo+1 to guarantee progress in case a
+ // TwoPhaseIterator returns a run end that does not move past the current doc.
+ upTo = Math.max(upTo + 1, Math.min(exclTwoPhase.docIDRunEnd(), max));
}
exclDoc = exclApproximation.nextDoc();
} else {
diff --git a/lucene/core/src/java/org/apache/lucene/search/TwoPhaseIterator.java b/lucene/core/src/java/org/apache/lucene/search/TwoPhaseIterator.java
index f4b7410ec9f..8a83e865480 100644
--- a/lucene/core/src/java/org/apache/lucene/search/TwoPhaseIterator.java
+++ b/lucene/core/src/java/org/apache/lucene/search/TwoPhaseIterator.java
@@ -117,10 +117,11 @@ public abstract class TwoPhaseIterator {
* <p><b>Note</b>: It is illegal to call this method when the approximation is exhausted or not
* positioned.
*
- * <p>The default implementation returns the current doc ID of the approximation.
+ * <p>The default implementation assumes runs of a single doc ID and returns the current doc ID of
+ * the approximation plus one, mirroring {@link DocIdSetIterator#docIDRunEnd()}.
*/
public int docIDRunEnd() throws IOException {
- return approximation().docID();
+ return approximation().docID() + 1;
}
/**