Skip to content

Speed up tail computation in MemorySegmentES91OSQVectorsScorer #132001

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 3 commits into from
Jul 29, 2025
Merged

Speed up tail computation in MemorySegmentES91OSQVectorsScorer #132001

Show file tree
Hide file tree
Changes from 1 commit
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
b4998b4
Speed up tail computation in MemorySegmentES91OSQVectorsScorer
iverase Jul 28, 2025
6299dca
update comment
iverase Jul 28, 2025
7f0ff42
Merge branch 'main' into speedTail
iverase Jul 28, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
2 changes: 1 addition & 1 deletion benchmarks/src/main/java/org/elasticsearch/benchmark/vector/OSQScorerBenchmark.java
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -52,7 +52,7 @@ public class OSQScorerBenchmark {
LogConfigurator.configureESLogging(); // native access requires logging to be initialized
}

@Param({ "1024" })
@Param({ "384", "782", "1024" })
int dims;

int length;
Expand Down
73 changes: 65 additions & 8 deletions ...a/org/elasticsearch/simdvec/internal/vectorization/MemorySegmentES91OSQVectorsScorer.java
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -18,6 +18,7 @@

import org.apache.lucene.index.VectorSimilarityFunction;
import org.apache.lucene.store.IndexInput;
import org.apache.lucene.util.BitUtil;
import org.apache.lucene.util.VectorUtil;
import org.elasticsearch.simdvec.ES91OSQVectorsScorer;

Expand Down Expand Up @@ -120,6 +121,20 @@ private long quantizeScore256(byte[] q) throws IOException {
}
// tail as bytes
in.seek(offset);
for (final int upperBound = length & -Long.BYTES; i < upperBound; i += Long.BYTES) {
final long value = in.readLong();
subRet0 += Long.bitCount((long) BitUtil.VH_LE_LONG.get(q, i) & value);
subRet1 += Long.bitCount((long) BitUtil.VH_LE_LONG.get(q, i + length) & value);
subRet2 += Long.bitCount((long) BitUtil.VH_LE_LONG.get(q, i + 2 * length) & value);
subRet3 += Long.bitCount((long) BitUtil.VH_LE_LONG.get(q, i + 3 * length) & value);
}
for (final int upperBound = length & -Integer.BYTES; i < upperBound; i += Integer.BYTES) {
final int value = in.readInt();
subRet0 += Integer.bitCount((int) BitUtil.VH_LE_INT.get(q, i) & value);
subRet1 += Integer.bitCount((int) BitUtil.VH_LE_INT.get(q, i + length) & value);
subRet2 += Integer.bitCount((int) BitUtil.VH_LE_INT.get(q, i + 2 * length) & value);
subRet3 += Integer.bitCount((int) BitUtil.VH_LE_INT.get(q, i + 3 * length) & value);
}
for (; i < length; i++) {
int dValue = in.readByte() & 0xFF;
subRet0 += Integer.bitCount((q[i] & dValue) & 0xFF);
Expand Down Expand Up @@ -160,12 +175,26 @@ private long quantizeScore128(byte[] q) throws IOException {
subRet3 += sum3.reduceLanes(VectorOperators.ADD);
// tail as bytes
in.seek(offset);
for (final int upperBound = length & -Long.BYTES; i < upperBound; i += Long.BYTES) {
final long value = in.readLong();
subRet0 += Long.bitCount((long) BitUtil.VH_LE_LONG.get(q, i) & value);
subRet1 += Long.bitCount((long) BitUtil.VH_LE_LONG.get(q, i + length) & value);
subRet2 += Long.bitCount((long) BitUtil.VH_LE_LONG.get(q, i + 2 * length) & value);
subRet3 += Long.bitCount((long) BitUtil.VH_LE_LONG.get(q, i + 3 * length) & value);
}
for (final int upperBound = length & -Integer.BYTES; i < upperBound; i += Integer.BYTES) {
final int value = in.readInt();
subRet0 += Integer.bitCount((int) BitUtil.VH_LE_INT.get(q, i) & value);
subRet1 += Integer.bitCount((int) BitUtil.VH_LE_INT.get(q, i + length) & value);
subRet2 += Integer.bitCount((int) BitUtil.VH_LE_INT.get(q, i + 2 * length) & value);
subRet3 += Integer.bitCount((int) BitUtil.VH_LE_INT.get(q, i + 3 * length) & value);
}
for (; i < length; i++) {
int dValue = in.readByte() & 0xFF;
subRet0 += Integer.bitCount((dValue & q[i]) & 0xFF);
subRet1 += Integer.bitCount((dValue & q[i + length]) & 0xFF);
subRet2 += Integer.bitCount((dValue & q[i + 2 * length]) & 0xFF);
subRet3 += Integer.bitCount((dValue & q[i + 3 * length]) & 0xFF);
subRet0 += Integer.bitCount((q[i] & dValue) & 0xFF);
subRet1 += Integer.bitCount((q[i + length] & dValue) & 0xFF);
subRet2 += Integer.bitCount((q[i + 2 * length] & dValue) & 0xFF);
subRet3 += Integer.bitCount((q[i + 3 * length] & dValue) & 0xFF);
}
return subRet0 + (subRet1 << 1) + (subRet2 << 2) + (subRet3 << 3);
}
Expand Down Expand Up @@ -217,12 +246,26 @@ private void quantizeScore128Bulk(byte[] q, int count, float[] scores) throws IO
subRet3 += sum3.reduceLanes(VectorOperators.ADD);
// tail as bytes
in.seek(offset);
for (final int upperBound = length & -Long.BYTES; i < upperBound; i += Long.BYTES) {
final long value = in.readLong();
subRet0 += Long.bitCount((long) BitUtil.VH_LE_LONG.get(q, i) & value);
subRet1 += Long.bitCount((long) BitUtil.VH_LE_LONG.get(q, i + length) & value);
subRet2 += Long.bitCount((long) BitUtil.VH_LE_LONG.get(q, i + 2 * length) & value);
subRet3 += Long.bitCount((long) BitUtil.VH_LE_LONG.get(q, i + 3 * length) & value);
}
for (final int upperBound = length & -Integer.BYTES; i < upperBound; i += Integer.BYTES) {
final int value = in.readInt();
subRet0 += Integer.bitCount((int) BitUtil.VH_LE_INT.get(q, i) & value);
subRet1 += Integer.bitCount((int) BitUtil.VH_LE_INT.get(q, i + length) & value);
subRet2 += Integer.bitCount((int) BitUtil.VH_LE_INT.get(q, i + 2 * length) & value);
subRet3 += Integer.bitCount((int) BitUtil.VH_LE_INT.get(q, i + 3 * length) & value);
}
for (; i < length; i++) {
int dValue = in.readByte() & 0xFF;
subRet0 += Integer.bitCount((dValue & q[i]) & 0xFF);
subRet1 += Integer.bitCount((dValue & q[i + length]) & 0xFF);
subRet2 += Integer.bitCount((dValue & q[i + 2 * length]) & 0xFF);
subRet3 += Integer.bitCount((dValue & q[i + 3 * length]) & 0xFF);
subRet0 += Integer.bitCount((q[i] & dValue) & 0xFF);
subRet1 += Integer.bitCount((q[i + length] & dValue) & 0xFF);
subRet2 += Integer.bitCount((q[i + 2 * length] & dValue) & 0xFF);
subRet3 += Integer.bitCount((q[i + 3 * length] & dValue) & 0xFF);
}
scores[iter] = subRet0 + (subRet1 << 1) + (subRet2 << 2) + (subRet3 << 3);
}
Expand Down Expand Up @@ -283,6 +326,20 @@ private void quantizeScore256Bulk(byte[] q, int count, float[] scores) throws IO
}
// tail as bytes
in.seek(offset);
for (final int upperBound = length & -Long.BYTES; i < upperBound; i += Long.BYTES) {
final long value = in.readLong();
subRet0 += Long.bitCount((long) BitUtil.VH_LE_LONG.get(q, i) & value);
subRet1 += Long.bitCount((long) BitUtil.VH_LE_LONG.get(q, i + length) & value);
subRet2 += Long.bitCount((long) BitUtil.VH_LE_LONG.get(q, i + 2 * length) & value);
subRet3 += Long.bitCount((long) BitUtil.VH_LE_LONG.get(q, i + 3 * length) & value);
}
for (final int upperBound = length & -Integer.BYTES; i < upperBound; i += Integer.BYTES) {
final int value = in.readInt();
subRet0 += Integer.bitCount((int) BitUtil.VH_LE_INT.get(q, i) & value);
subRet1 += Integer.bitCount((int) BitUtil.VH_LE_INT.get(q, i + length) & value);
subRet2 += Integer.bitCount((int) BitUtil.VH_LE_INT.get(q, i + 2 * length) & value);
subRet3 += Integer.bitCount((int) BitUtil.VH_LE_INT.get(q, i + 3 * length) & value);
}
for (; i < length; i++) {
int dValue = in.readByte() & 0xFF;
subRet0 += Integer.bitCount((q[i] & dValue) & 0xFF);
Expand Down