Skip to content
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

Speed up IndexedDISI Sparse #AdvanceExactWithinBlock for tiny step advance #12324

Merged
merged 4 commits into from
Jun 13, 2023
Merged

Speed up IndexedDISI Sparse #AdvanceExactWithinBlock for tiny step advance #12324

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
6a791d7
cache sparse next doc
gf2121 May 22, 2023
7ddde17
add CHANGES
gf2121 May 22, 2023
31c7c12
feedback fix: remove integer.max_value
gf2121 Jun 1, 2023
3823a25
feedback fix: 65536->BLOCK_SIZE
gf2121 Jun 2, 2023
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: 2 additions & 0 deletions lucene/CHANGES.txt
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -131,6 +131,8 @@ Improvements
Optimizations
---------------------

* GITHUB#12324: Speed up sparse block advanceExact with tiny step in IndexedDISI. (Guo Feng)

* GITHUB#12270 Don't generate stacktrace in CollectionTerminatedException. (Armin Braun)

* GITHUB#12286 Toposort use iterator to avoid stackoverflow. (Tang Donghai)
Expand Down
16 changes: 12 additions & 4 deletions lucene/core/src/java/org/apache/lucene/codecs/lucene90/IndexedDISI.java
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -110,12 +110,12 @@ public final class IndexedDISI extends DocIdSetIterator {
private static void flush(
int block, FixedBitSet buffer, int cardinality, byte denseRankPower, IndexOutput out)
throws IOException {
assert block >= 0 && block < 65536;
assert block >= 0 && block < BLOCK_SIZE;
out.writeShort((short) block);
assert cardinality > 0 && cardinality <= 65536;
assert cardinality > 0 && cardinality <= BLOCK_SIZE;
out.writeShort((short) (cardinality - 1));
if (cardinality > MAX_ARRAY_LENGTH) {
if (cardinality != 65536) { // all docs are set
if (cardinality != BLOCK_SIZE) { // all docs are set
if (denseRankPower != -1) {
final byte[] rank = createRank(buffer, denseRankPower);
out.writeBytes(rank, rank.length);
Expand Down Expand Up @@ -418,6 +418,7 @@ public static RandomAccessInput createJumpTable(

// SPARSE variables
boolean exists;
int nextExistDocInBlock = -1;
zhaih marked this conversation as resolved.
Show resolved Hide resolved

// DENSE variables
long word;
Expand Down Expand Up @@ -495,7 +496,8 @@ private void readBlockHeader() throws IOException {
if (numValues <= MAX_ARRAY_LENGTH) {
method = Method.SPARSE;
blockEnd = slice.getFilePointer() + (numValues << 1);
} else if (numValues == 65536) {
nextExistDocInBlock = -1;
} else if (numValues == BLOCK_SIZE) {
method = Method.ALL;
blockEnd = slice.getFilePointer();
gap = block - index - 1;
Expand Down Expand Up @@ -550,6 +552,7 @@ boolean advanceWithinBlock(IndexedDISI disi, int target) throws IOException {
if (doc >= targetInBlock) {
disi.doc = disi.block | doc;
disi.exists = true;
disi.nextExistDocInBlock = doc;
return true;
}
}
Expand All @@ -560,13 +563,18 @@ boolean advanceWithinBlock(IndexedDISI disi, int target) throws IOException {
boolean advanceExactWithinBlock(IndexedDISI disi, int target) throws IOException {
final int targetInBlock = target & 0xFFFF;
// TODO: binary search
if (disi.nextExistDocInBlock > targetInBlock) {
assert !disi.exists;
return false;
}
if (target == disi.doc) {
return disi.exists;
}
for (; disi.index < disi.nextBlockIndex; ) {
int doc = Short.toUnsignedInt(disi.slice.readShort());
disi.index++;
if (doc >= targetInBlock) {
disi.nextExistDocInBlock = doc;
if (doc != targetInBlock) {
disi.index--;
disi.slice.seek(disi.slice.getFilePointer() - Short.BYTES);
Expand Down