uschindler commented on code in PR #12694:
URL: https://github.com/apache/lucene/pull/12694#discussion_r1365474987
##########
lucene/core/src/java20/org/apache/lucene/internal/vectorization/PanamaVectorUtilSupport.java:
##########
@@ -352,6 +382,11 @@ private int dotProductBody512(byte[] a, byte[] b, int
limit) {
// 16-bit multiply: avoid AVX-512 heavy multiply on zmm
Vector<Short> va16 = va8.convertShape(B2S, SHORT_SPECIES, 0);
Vector<Short> vb16 = vb8.convertShape(B2S, SHORT_SPECIES, 0);
+ if (unsigned) {
Review Comment:
the problem is that you won't see this problem in benchmark, because each
benchmark runs in a separate VM which always calls `dotProductBody512` with
always same parameter. Hotspot will for sure optimize this. But if you have
productive code that sometimes uses signed and sometimes unsigned
multiplication, the method will be deoptimized on every change as it runs into
a trap (or like that). That's not what you want.
--
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
