LUCENE-1443: Performance improvement for OpenBitSetDISI.inPlaceAnd()

git-svn-id: https://svn.apache.org/repos/asf/lucene/java/trunk@712908 13f79535-47bb-0310-9956-ffa450edef68

CHANGES.txt

@@ -70,6 +70,9 @@ Optimizations
     more efficient (single pass) by not creating & populating an
     intermediate OpenBitSet (Paul Elschot, Mike McCandless)
 
+ 2. LUCENE-1443: Performance improvement for OpenBitSetDISI.inPlaceAnd()
+    (Paul Elschot via yonik)
+
 
 Documentation
 

src/java/org/apache/lucene/util/BitUtil.java

@@ -688,7 +688,7 @@ public class BitUtil {
   public static final byte[] ntzTable = {8,0,1,0,2,0,1,0,3,0,1,0,2,0,1,0,4,0,1,0,2,0,1,0,3,0,1,0,2,0,1,0,5,0,1,0,2,0,1,0,3,0,1,0,2,0,1,0,4,0,1,0,2,0,1,0,3,0,1,0,2,0,1,0,6,0,1,0,2,0,1,0,3,0,1,0,2,0,1,0,4,0,1,0,2,0,1,0,3,0,1,0,2,0,1,0,5,0,1,0,2,0,1,0,3,0,1,0,2,0,1,0,4,0,1,0,2,0,1,0,3,0,1,0,2,0,1,0,7,0,1,0,2,0,1,0,3,0,1,0,2,0,1,0,4,0,1,0,2,0,1,0,3,0,1,0,2,0,1,0,5,0,1,0,2,0,1,0,3,0,1,0,2,0,1,0,4,0,1,0,2,0,1,0,3,0,1,0,2,0,1,0,6,0,1,0,2,0,1,0,3,0,1,0,2,0,1,0,4,0,1,0,2,0,1,0,3,0,1,0,2,0,1,0,5,0,1,0,2,0,1,0,3,0,1,0,2,0,1,0,4,0,1,0,2,0,1,0,3,0,1,0,2,0,1,0};
 
 
-  /** Returns number of trailing zeros in the 64 bit long value. */
+  /** Returns number of trailing zeros in a 64 bit long value. */
   public static int ntz(long val) {
     // A full binary search to determine the low byte was slower than
     // a linear search for nextSetBit().  This is most likely because
@@ -728,6 +728,23 @@ public class BitUtil {
     }
   }
 
+  /** Returns number of trailing zeros in a 32 bit int value. */
+  public static int ntz(int val) {
+    // This implementation does a single binary search at the top level only.
+    // In addition, the case of a non-zero first byte is checked for first
+    // because it is the most common in dense bit arrays.
+
+    int lowByte = val & 0xff;
+    if (lowByte != 0) return ntzTable[lowByte];
+    lowByte = (val>>>8) & 0xff;
+    if (lowByte != 0) return ntzTable[lowByte] + 8;
+    lowByte = (val>>>16) & 0xff;
+    if (lowByte != 0) return ntzTable[lowByte] + 16;
+    // no need to mask off low byte for the last byte.
+    // no need to check for zero on the last byte either.
+    return ntzTable[val>>>24] + 24;
+  }
+
   /** returns 0 based index of first set bit
    * (only works for x!=0)
    * <br/> This is an alternate implementation of ntz()

src/java/org/apache/lucene/util/OpenBitSet.java

@@ -327,6 +327,43 @@ public class OpenBitSet extends DocIdSet implements Cloneable, Serializable {
     bits[wordNum] &= ~bitmask;
   }
 
+  /** Clears a range of bits.  Clearing past the end does not change the size of the set.
+   *
+   * @param startIndex lower index
+   * @param endIndex one-past the last bit to clear
+   */
+  public void clear(int startIndex, int endIndex) {
+    if (endIndex <= startIndex) return;
+
+    int startWord = (startIndex>>6);
+    if (startWord >= wlen) return;
+
+    // since endIndex is one past the end, this is index of the last
+    // word to be changed.
+    int endWord   = ((endIndex-1)>>6);
+
+    long startmask = -1L << startIndex;
+    long endmask = -1L >>> -endIndex;  // 64-(endIndex&0x3f) is the same as -endIndex due to wrap
+
+    // invert masks since we are clearing
+    startmask = ~startmask;
+    endmask = ~endmask;
+
+    if (startWord == endWord) {
+      bits[startWord] &= (startmask | endmask);
+      return;
+    }
+
+    bits[startWord] &= startmask;
+
+    int middle = Math.min(wlen, endWord);
+    Arrays.fill(bits, startWord+1, middle, 0L);
+    if (endWord < wlen) {
+      bits[endWord] &= endmask;
+    }
+  }
+
+
   /** Clears a range of bits.  Clearing past the end does not change the size of the set.
    *
    * @param startIndex lower index

src/java/org/apache/lucene/util/OpenBitSetDISI.java

@@ -59,20 +59,15 @@ public class OpenBitSetDISI extends OpenBitSet {
    * constructor.
    */   
   public void inPlaceAnd(DocIdSetIterator disi) throws IOException {
-    int index = nextSetBit(0);
-    int lastNotCleared = -1;
-    while ((index != -1) && disi.skipTo(index)) {
-      while ((index != -1) && (index < disi.doc())) {
-        fastClear(index);
-        index = nextSetBit(index + 1);
+    int bitSetDoc = nextSetBit(0);
+    while ((bitSetDoc != -1) && disi.skipTo(bitSetDoc)) {
+      int disiDoc = disi.doc();
+      clear(bitSetDoc, disiDoc);
+      bitSetDoc = nextSetBit(disiDoc + 1);
     }
-      if (index == disi.doc()) {
-        lastNotCleared = index;
-        index++;
+    if (bitSetDoc != -1) {
+      clear(bitSetDoc, size());
     }
-      assert (index == -1) || (index > disi.doc());
-    }
-    clear(lastNotCleared+1, size());
   }
 
   /**