bruno-roustant commented on a change in pull request #980: LUCENE-8920: Reduce 
the memory used by direct addressing of arcs
URL: https://github.com/apache/lucene-solr/pull/980#discussion_r339983062
 
 

 ##########
 File path: lucene/core/src/java/org/apache/lucene/util/fst/FST.java
 ##########
 @@ -864,6 +1012,68 @@ private long readUnpackedNodeTarget(BytesReader in) 
throws IOException {
     return readNextRealArc(arc, in);
   }
 
+  /**
+   * Reads the presence bits of a direct-addressing node, store them in the 
provided arc {@link Arc#bitTable()} and returns the number of presence bytes.
+   */
+  private int readPresenceBytes(Arc<T> arc, BytesReader in) throws IOException 
{
+    int numPresenceBytes = getNumPresenceBytes(arc.numArcs());
+    long[] presenceBits = new long[numPresenceBytes + 7 >>> 3];
+    int longIndex = 0;
+    int byteIndex = 0;
+    for (int i = 0; i < numPresenceBytes; i++) {
+      presenceBits[longIndex] |= (long) (in.readByte() & 0xFF) << (i << 3);
+      if (++byteIndex == 8) {
+        byteIndex = 0;
+        longIndex++;
+      }
+    }
+    arc.bitTable = presenceBits;
+    assert checkPresenceBytesAreValid(arc);
+    return numPresenceBytes;
+  }
+
+  static boolean checkPresenceBytesAreValid(Arc arc) {
+    assert (arc.bitTable()[0] & 1L) != 0; // First bit must be set.
+    assert (arc.bitTable()[arc.bitTable().length - 1] & (1L << (arc.numArcs() 
- 1))) != 0; // Last bit must be set.
+    assert countBits(arc.bitTable()) <= arc.numArcs(); // Total bit set must 
be <= label range.
+    return true;
+  }
+
+  /**
+   * Counts all bits set in the provided longs.
+   */
+  static int countBits(long[] bits) {
+    int bitCount = 0;
+    for (long l : bits) {
+      bitCount += Long.bitCount(l);
+    }
+    return bitCount;
+  }
+
+  /**
+   * Returns whether the bit at given index is set.
+   */
+  static boolean isBitSet(long[] bits, int bitIndex) {
+    return (bits[bitIndex >> 6] & (1L << (bitIndex & 63))) != 0;
 
 Review comment:
   Thanks for the tip about the removal of the mask.
   
   About the arithmetic shift, my idea was to keep the idea of "division by 
Long.SIZE", but I agree this is not clear. For clarity I finally prefer simply 
"bits[bitIndex / Long.SIZE]".

----------------------------------------------------------------
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.
 
For queries about this service, please contact Infrastructure at:
us...@infra.apache.org


With regards,
Apache Git Services

---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org
For additional commands, e-mail: issues-h...@lucene.apache.org

Reply via email to