mccullocht commented on issue #15697:
URL: https://github.com/apache/lucene/issues/15697#issuecomment-3930769434
Seeing the same kind of warmup behavior on a Mac M2. I'll check x86 when I'm
at home tonight.
```
# Warmup Iteration 1: Feb 19, 2026 2:10:55 PM
org.apache.lucene.internal.vectorization.PanamaVectorizationProvider
logIncubatorSetup
INFO: Java vector incubator API enabled; uses preferredBitSize=128
13.816 ops/us
# Warmup Iteration 2: 18.402 ops/us
# Warmup Iteration 3: 2.481 ops/us
# Warmup Iteration 4: 2.656 ops/us
Iteration 1: 2.555 ops/us
Iteration 2: 2.510 ops/us
Iteration 3: 2.553 ops/us
Iteration 4: 2.693 ops/us
Iteration 5: 2.668 ops/us
```
I remember there being an issue with call site pollution in the
ByteVectorLoader implementations but I don't think that would trigger in the
benchmark since both sides are using `ArrayLoader`. I ran async-profiler in the
benchmark and it seemed to indicate that the following lines were very
expensive:
```
Vector<Integer> intAcc0 = acc0.convert(S2I, 0);
Vector<Integer> intAcc1 = acc0.convert(S2I, 1);
Vector<Integer> intAcc2 = acc1.convert(S2I, 0);
Vector<Integer> intAcc3 = acc1.convert(S2I, 1);
```
--
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: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
