Add SIMD-accelerated bulk range evaluation for dense numeric doc values

lucene/CHANGES.txt

@@ -374,6 +374,9 @@ Optimizations
 
 * GITHUB#16001: IndexSearcher.count() was calling query.rewrite twice, a regression since v9.10 (David Smiley)
 
+* GITHUB#16050: Add SIMD-accelerated bulk range evaluation for dense numeric doc values via
+  BatchDocValuesRangeIterator and DocValuesRangeSupport. (Sagar Upadhyaya)
+
 * GITHUB16061#: Improve cost estimation in SortedSetDocValuesRangeQuery when using DocValuesSkipper and the field
   is dense and is the primary sort of the index and reduce the number of doc values visited. (Ignacio Vera)
 

lucene/benchmark-jmh/src/java/org/apache/lucene/benchmark/jmh/MultiFieldDocValuesRangeBenchmark.java

@@ -87,7 +87,7 @@ public static class Params {
     @Param({"1000000", "10000000"})
     public int docCount;
 
-    @Param({"3", "5"})
+    @Param({"1", "3", "5"})
     public int fieldCount;
 
     @Param({CLUSTERED, MIXED, RANDOM, SORTED})
@@ -129,6 +129,17 @@ public void setup(Params params) throws Exception {
           SortedNumericDocValuesField.newSlowRangeQuery("field" + f, range[0], range[1]),
           Occur.FILTER);
     }
+    // For fieldCount=1 on non-sorted patterns, add a MatchAllDocsQuery so
+    // DenseConjunctionBulkScorer is used and intoBitSet() is called on the range iterator
+    // (enabling the SIMD path). Without this, a single-clause BooleanQuery rewrites to the
+    // query itself and goes through DefaultBulkScorer which doesn't call intoBitSet().
+    // For the sorted pattern, field0 is the index sort key so
+    // getDocIdSetIteratorOrNullForPrimarySort
+    // fires and returns DocIdSetIterator.range() — adding MatchAllDocsQuery here would force it
+    // through DenseConjunctionBulkScorer and bypass that fast path, causing a regression.
+    if (params.fieldCount == 1 && !params.dataPattern.equals(SORTED)) {
+      bqBuilder.add(new org.apache.lucene.search.MatchAllDocsQuery(), Occur.FILTER);
+    }
     query = bqBuilder.build();
   }
 

lucene/core/src/java/org/apache/lucene/codecs/lucene90/Lucene90DocValuesProducer.java

@@ -391,6 +391,24 @@ public void close() throws IOException {
   private record DocValuesSkipperEntry(
       long offset, long length, long minValue, long maxValue, int docCount, int maxDocId) {}
 
+  // Cached VectorizationProvider instance to avoid repeated stack walks in ensureCaller()
+  private static final org.apache.lucene.internal.vectorization.DocValuesRangeSupport
+      DOC_VALUES_RANGE_SUPPORT =
+          org.apache.lucene.internal.vectorization.VectorizationProvider.getInstance()
+              .getDocValuesRangeSupport();
+
+  static void rangeIntoBitSet(
+      org.apache.lucene.util.LongValues values,
+      int fromDoc,
+      int toDoc,
+      long minValue,
+      long maxValue,
+      org.apache.lucene.util.FixedBitSet bitSet,
+      int offset) {
+    DOC_VALUES_RANGE_SUPPORT.rangeIntoBitSet(
+        values, fromDoc, toDoc, minValue, maxValue, bitSet, offset);
+  }
+
   private static class NumericEntry {
     long[] table;
     int blockShift;
@@ -610,6 +628,19 @@ public long longValue() throws IOException {
               public long longValue() throws IOException {
                 return values.get(doc);
               }
+
+              @Override
+              public void rangeIntoBitSet(
+                  int fromDoc,
+                  int toDoc,
+                  long minValue,
+                  long maxValue,
+                  org.apache.lucene.util.FixedBitSet bitSet,
+                  int offset) {
+                // Bulk range evaluation via DocValuesRangeSupport
+                Lucene90DocValuesProducer.rangeIntoBitSet(
+                    values, fromDoc, toDoc, minValue, maxValue, bitSet, offset);
+              }
             };
           } else {
             final long mul = entry.gcd;
@@ -619,6 +650,23 @@ public long longValue() throws IOException {
               public long longValue() throws IOException {
                 return mul * values.get(doc) + delta;
               }
+
+              @Override
+              public void rangeIntoBitSet(
+                  int fromDoc,
+                  int toDoc,
+                  long minValue,
+                  long maxValue,
+                  org.apache.lucene.util.FixedBitSet bitSet,
+                  int offset) {
+                // Per-doc evaluation for gcd/delta encoded fields
+                for (int d = fromDoc; d < toDoc; d++) {
+                  long v = mul * values.get(d) + delta;
+                  if (v >= minValue && v <= maxValue) {
+                    bitSet.set(d - offset);
+                  }
+                }
+              }
             };
           }
         }

lucene/core/src/java/org/apache/lucene/document/SortedNumericDocValuesRangeQuery.java

@@ -25,6 +25,7 @@
 import org.apache.lucene.index.LeafReaderContext;
 import org.apache.lucene.index.NumericDocValues;
 import org.apache.lucene.index.SortedNumericDocValues;
+import org.apache.lucene.search.BatchDocValuesRangeIterator;
 import org.apache.lucene.search.ConstantScoreScorerSupplier;
 import org.apache.lucene.search.ConstantScoreWeight;
 import org.apache.lucene.search.DocIdSetIterator;
@@ -145,6 +146,12 @@ public ScorerSupplier scorerSupplier(LeafReaderContext context) throws IOExcepti
               return ConstantScoreScorerSupplier.fromIterator(
                   psIterator, score(), scoreMode, maxDoc);
             }
+            // Use batch iterator for bulk block evaluation via intoBitSet()
+            return ConstantScoreScorerSupplier.fromIterator(
+                new BatchDocValuesRangeIterator(singleton, skipper, lowerValue, upperValue),
+                score(),
+                scoreMode,
+                maxDoc);
           }
           return ConstantScoreScorerSupplier.fromIterator(
               TwoPhaseIterator.asDocIdSetIterator(

lucene/core/src/java/org/apache/lucene/index/NumericDocValues.java

@@ -20,6 +20,7 @@
 import java.io.IOException;
 import org.apache.lucene.search.BooleanClause.Occur;
 import org.apache.lucene.search.FieldExistsQuery;
+import org.apache.lucene.util.FixedBitSet;
 
 /** A per-document numeric value. */
 public abstract class NumericDocValues extends DocValuesIterator {
@@ -91,4 +92,33 @@ public void longValues(int size, int[] docs, long[] values, long defaultValue)
       values[i] = value;
     }
   }
+
+  /**
+   * Fills a {@link org.apache.lucene.util.FixedBitSet} with the doc IDs in {@code [fromDoc, toDoc)}
+   * whose values are in {@code [minValue, maxValue]}. This is a bulk operation that avoids per-doc
+   * virtual dispatch overhead.
+   *
+   * <p>The default implementation falls back to per-doc evaluation via {@link #advanceExact} and
+   * {@link #longValue}. Subclasses with random-access storage (e.g., dense fixed-bitsPerValue
+   * fields) can override this for significantly better performance.
+   *
+   * @param fromDoc first doc ID to evaluate (inclusive)
+   * @param toDoc last doc ID to evaluate (exclusive)
+   * @param minValue lower bound of the range (inclusive)
+   * @param maxValue upper bound of the range (inclusive)
+   * @param bitSet the bitset to fill
+   * @param offset subtracted from each doc ID before setting the bit
+   */
+  public void rangeIntoBitSet(
+      int fromDoc, int toDoc, long minValue, long maxValue, FixedBitSet bitSet, int offset)
+      throws IOException {
+    for (int d = fromDoc; d < toDoc; d++) {
+      if (advanceExact(d)) {
+        long v = longValue();
+        if (v >= minValue && v <= maxValue) {
+          bitSet.set(d - offset);
+        }
+      }
+    }
+  }
 }

lucene/core/src/java/org/apache/lucene/internal/vectorization/DefaultDocValuesRangeSupport.java

@@ -0,0 +1,46 @@
+/*
+ * Licensed to the Apache Software Foundation (ASF) under one or more
+ * contributor license agreements.  See the NOTICE file distributed with
+ * this work for additional information regarding copyright ownership.
+ * The ASF licenses this file to You under the Apache License, Version 2.0
+ * (the "License"); you may not use this file except in compliance with
+ * the License.  You may obtain a copy of the License at
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+package org.apache.lucene.internal.vectorization;
+
+import org.apache.lucene.util.FixedBitSet;
+import org.apache.lucene.util.LongValues;
+
+/** Scalar (non-SIMD) implementation of {@link DocValuesRangeSupport}. */
+final class DefaultDocValuesRangeSupport implements DocValuesRangeSupport {
+
+  static final DefaultDocValuesRangeSupport INSTANCE = new DefaultDocValuesRangeSupport();
+
+  private DefaultDocValuesRangeSupport() {}
+
+  @Override
+  public void rangeIntoBitSet(
+      LongValues values,
+      int fromDoc,
+      int toDoc,
+      long minValue,
+      long maxValue,
+      FixedBitSet bitSet,
+      int offset) {
+    // Scalar fallback implementation
+    for (int d = fromDoc; d < toDoc; d++) {
+      long v = values.get(d);
+      if (v >= minValue && v <= maxValue) {
+        bitSet.set(d - offset);
+      }
+    }
+  }
+}

lucene/core/src/java/org/apache/lucene/internal/vectorization/DefaultVectorizationProvider.java

@@ -50,4 +50,9 @@ public FlatVectorsScorer getLucene99ScalarQuantizedVectorsScorer() {
   public PostingDecodingUtil newPostingDecodingUtil(IndexInput input) {
     return new PostingDecodingUtil(input);
   }
+
+  @Override
+  public DocValuesRangeSupport getDocValuesRangeSupport() {
+    return DefaultDocValuesRangeSupport.INSTANCE;
+  }
 }

lucene/core/src/java/org/apache/lucene/internal/vectorization/DocValuesRangeSupport.java

@@ -0,0 +1,53 @@
+/*
+ * Licensed to the Apache Software Foundation (ASF) under one or more
+ * contributor license agreements.  See the NOTICE file distributed with
+ * this work for additional information regarding copyright ownership.
+ * The ASF licenses this file to You under the Apache License, Version 2.0
+ * (the "License"); you may not use this file except in compliance with
+ * the License.  You may obtain a copy of the License at
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+package org.apache.lucene.internal.vectorization;
+
+import org.apache.lucene.util.FixedBitSet;
+import org.apache.lucene.util.LongValues;
+
+/**
+ * Interface for SIMD-accelerated doc values range operations.
+ *
+ * <p>Implementations fill a {@link FixedBitSet} with the doc IDs in a range whose values satisfy a
+ * numeric range predicate. The default scalar implementation is used when the Panama Vector API is
+ * unavailable; a SIMD-accelerated implementation is used otherwise.
+ *
+ * @lucene.internal
+ */
+public interface DocValuesRangeSupport {
+
+  /**
+   * Fills {@code bitSet} with the doc IDs in {@code [fromDoc, toDoc)} whose values (read via {@code
+   * values}) are in {@code [minValue, maxValue]}.
+   *
+   * @param values random-access reader for the doc values
+   * @param fromDoc first doc ID to evaluate (inclusive)
+   * @param toDoc last doc ID to evaluate (exclusive)
+   * @param minValue lower bound of the range (inclusive)
+   * @param maxValue upper bound of the range (inclusive)
+   * @param bitSet the bitset to fill
+   * @param offset subtracted from each doc ID before setting the bit
+   */
+  void rangeIntoBitSet(
+      LongValues values,
+      int fromDoc,
+      int toDoc,
+      long minValue,
+      long maxValue,
+      FixedBitSet bitSet,
+      int offset);
+}

lucene/core/src/java/org/apache/lucene/internal/vectorization/VectorizationProvider.java

@@ -115,6 +115,13 @@ public static VectorizationProvider getInstance() {
   /** Create a new {@link PostingDecodingUtil} for the given {@link IndexInput}. */
   public abstract PostingDecodingUtil newPostingDecodingUtil(IndexInput input) throws IOException;
 
+  /**
+   * Returns a {@link DocValuesRangeSupport} instance for bulk numeric range evaluation. The
+   * returned instance uses SIMD when available (Panama Vector API), falling back to a scalar loop
+   * otherwise.
+   */
+  public abstract DocValuesRangeSupport getDocValuesRangeSupport();
+
   // *** Lookup mechanism: ***
 
   private static final Logger LOG = Logger.getLogger(VectorizationProvider.class.getName());
@@ -213,6 +220,7 @@ private static Optional<Module> lookupVectorModule() {
           "org.apache.lucene.util.VectorUtil",
           "org.apache.lucene.codecs.lucene104.Lucene104PostingsReader",
           "org.apache.lucene.codecs.lucene104.PostingIndexInput",
+          "org.apache.lucene.codecs.lucene90.Lucene90DocValuesProducer",
           "org.apache.lucene.tests.util.TestSysoutsLimits");
 
   private static final StackWalker STACKWALKER =