jpountz commented on code in PR #14906:
URL: https://github.com/apache/lucene/pull/14906#discussion_r2189422844
##########
lucene/core/src/java/org/apache/lucene/search/ScorerUtil.java:
##########
@@ -157,11 +157,10 @@ static void filterCompetitiveHits(
int newSize = 0;
for (int i = 0; i < buffer.size; ++i) {
- if (buffer.scores[i] >= minRequiredScore) {
- buffer.docs[newSize] = buffer.docs[i];
- buffer.scores[newSize] = buffer.scores[i];
- newSize++;
- }
+ int inc = buffer.scores[i] >= minRequiredScore ? 1 : 0;
+ buffer.docs[newSize] = buffer.docs[i];
+ buffer.scores[newSize] = buffer.scores[i];
+ newSize += inc;
Review Comment:
I checked the assembly, it doesn't use cmov, it looks like it does what the
Java code suggests by adding the result of the comparison to `newSize`.
However, your suggestion made me want to look into making C2 generate cmov
instructions, and the below approach worked:
```java
@Benchmark
public int branchlessCandidateCmov() {
int newSize = 0;
for (int i = 0; i < size; ++i) {
int doc = docs[i];
double score = scores[i];
docs[newSize] = doc;
scores[newSize] = score;
if (score >= minScoreInclusive) {
newSize += 1;
}
}
return newSize;
}
```
Plus it's faster:
```
Benchmark (minScoreInclusive) (size)
Mode Cnt Score Error Units
CompetitiveBenchmark.branchlessCandidate 0 128
thrpt 5 16462.769 ± 1385.801 ops/ms
CompetitiveBenchmark.branchlessCandidate 0.2 128
thrpt 5 8681.387 ± 510.207 ops/ms
CompetitiveBenchmark.branchlessCandidate 0.4 128
thrpt 5 8469.440 ± 279.038 ops/ms
CompetitiveBenchmark.branchlessCandidate 0.5 128
thrpt 5 8403.283 ± 371.047 ops/ms
CompetitiveBenchmark.branchlessCandidate 0.8 128
thrpt 5 8497.105 ± 250.696 ops/ms
CompetitiveBenchmark.branchlessCandidateCmov 0 128
thrpt 5 16974.162 ± 386.891 ops/ms
CompetitiveBenchmark.branchlessCandidateCmov 0.2 128
thrpt 5 10308.811 ± 115.632 ops/ms
CompetitiveBenchmark.branchlessCandidateCmov 0.4 128
thrpt 5 10583.388 ± 434.330 ops/ms
CompetitiveBenchmark.branchlessCandidateCmov 0.5 128
thrpt 5 10368.750 ± 539.356 ops/ms
CompetitiveBenchmark.branchlessCandidateCmov 0.8 128
thrpt 5 10306.593 ± 499.033 ops/ms
```
--
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.
To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org
For queries about this service, please contact Infrastructure at:
us...@infra.apache.org
---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org
For additional commands, e-mail: issues-h...@lucene.apache.org
