improve BytesRefHash.sort performance by rearranging ids (#15772)

* improve BytesRefHash.sort performance

Adjusting the data order in ids during compaction,

which can improve data access continuity and reduce cache-misses.

finally enhance sort performance by 20% in million-term tests

* Update changes.

Added a new entry to CHANGES.txt to document performance improvement in BytesRefHash.sort.

* add comment in compact

* Fix comment typo in compact method

* update comment

---------

Co-authored-by: Dawid Weiss <dawid.weiss@carrotsearch.com>

Author: tyronecai

lucene/CHANGES.txt

@@ -235,6 +235,8 @@ Optimizations
 
 * GITHUB#15729: Lucene90DocValuesProducer is not prefetching any data for DocValueSkippers anymore (Alexander Reelsen)
 
+* GITHUB#15772: improve BytesRefHash.sort performance by rearranging ids#15772 (tyronecai)
+
 * GITHUB#15742: Optimize int4 dotProduct and squareDistance computations by replacing vector conversions with reinterpret casting + bit manipulation. (Trevor McCulloch, Kaival Parikh)
 
 Bug Fixes

lucene/core/src/java/org/apache/lucene/util/BytesRefHash.java

@@ -178,18 +178,13 @@ public BytesRef get(int bytesID, BytesRef ref) {
    */
   public int[] compact() {
     assert bytesStart != null : "bytesStart is null - not initialized";
-    int upto = 0;
-    for (int i = 0; i < hashSize; i++) {
-      if (ids[i] != -1) {
-        ids[upto] = ids[i] & hashMask;
-        if (upto < i) {
-          ids[i] = -1;
-        }
-        upto++;
-      }
+
+    // id is the sequence number when bytes added to the pool
+    for (int i = 0; i < count; i++) {
+      ids[i] = i;
     }
+    Arrays.fill(ids, count, hashSize, -1);
 
-    assert upto == count;
     lastCount = count;
     return ids;
   }