Apache
/
lucene
mirror of https://github.com/apache/lucene.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

optimize BKDWriter's offline comparator a bit

This commit is contained in:
Mike McCandless 2016-03-01 16:08:01 -05:00
parent 0f4f53a8f5
commit 251cdbcee3
2 changed files with 15 additions and 13 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

1
lucene/core/src/java/org/apache/lucene/util/StringHelper.java
View File

@ -377,6 +377,7 @@ public abstract class StringHelper {
* big-endian unsigned values. Returns positive int if a > b,
* negative int if a < b and 0 if a == b */
public static int compare(int count, byte[] a, int aOffset, byte[] b, int bOffset) {
// TODO: dedup this w/ BytesRef.compareTo?
for(int i=0;i<count;i++) {
int cmp = (a[aOffset+i]&0xff) - (b[bOffset+i]&0xff);
if (cmp != 0) {

27
lucene/core/src/java/org/apache/lucene/util/bkd/BKDWriter.java
View File

@ -665,34 +665,35 @@ public class BKDWriter implements Closeable {
// Offline sort:
assert tempInput != null;
final ByteArrayDataInput reader = new ByteArrayDataInput();
Comparator<BytesRef> cmp = new Comparator<BytesRef>() {
private final ByteArrayDataInput readerB = new ByteArrayDataInput();
final ByteArrayDataInput reader = new ByteArrayDataInput();
@Override
public int compare(BytesRef a, BytesRef b) {
reader.reset(a.bytes, a.offset, a.length);
reader.readBytes(scratch1, 0, scratch1.length);
final int docIDA = reader.readVInt();
final long ordA = reader.readVLong();
reader.reset(b.bytes, b.offset, b.length);
reader.readBytes(scratch2, 0, scratch2.length);
final int docIDB = reader.readVInt();
final long ordB = reader.readVLong();
int cmp = StringHelper.compare(bytesPerDim, scratch1, bytesPerDim*dim, scratch2, bytesPerDim*dim);
// First compare the bytes on the dimension we are sorting on:
int cmp = StringHelper.compare(bytesPerDim, a.bytes, a.offset + bytesPerDim*dim, b.bytes, b.offset + bytesPerDim*dim);
if (cmp != 0) {
return cmp;
}
// Tie-break
// Tie-break by docID and then ord:
reader.reset(a.bytes, a.offset + packedBytesLength, a.length);
final int docIDA = reader.readVInt();
final long ordA = reader.readVLong();
reader.reset(b.bytes, b.offset + packedBytesLength, b.length);
final int docIDB = reader.readVInt();
final long ordB = reader.readVLong();
cmp = Integer.compare(docIDA, docIDB);
if (cmp != 0) {
return cmp;
}
// TODO: is this really necessary? If OfflineSorter is stable, we can safely return 0 here, and avoid writing ords?
return Long.compare(ordA, ordB);
}
};