zhaih commented on PR #12660:
URL: https://github.com/apache/lucene/pull/12660#issuecomment-1786511544
FYI I pushed a commit
https://github.com/apache/lucene/commit/4576ae09e8885f40cc27424fa8d529aa5c172422
to fix the behavior that vector writer close the executor passed in
--
This is an
rmuir commented on PR #12737:
URL: https://github.com/apache/lucene/pull/12737#issuecomment-1786498331
Also it was just previously confusing to see stuff like vector benchmark
results with 128-bit arm vectors going 8x faster than 32-bit floats which makes
no logical sense. e.g. with this ch
rmuir commented on PR #12737:
URL: https://github.com/apache/lucene/pull/12737#issuecomment-1786433389
I tried naively writing the logic like this with a couple N (8, 16, 32,etc)
with FMA both off and on to see if I can baby this compiler to vectorize, nope,
nothing. I don't think autovect
rmuir commented on code in PR #12737:
URL: https://github.com/apache/lucene/pull/12737#discussion_r1377027128
##
lucene/core/src/java/org/apache/lucene/internal/vectorization/DefaultVectorUtilSupport.java:
##
@@ -17,72 +17,46 @@
package org.apache.lucene.internal.vectorizatio
rmuir commented on PR #12737:
URL: https://github.com/apache/lucene/pull/12737#issuecomment-1786395055
e.g. for dotproduct case, with this patch, despite there being no data
dependencies, compiler literally does 4 `VFMADD*SS` in the loop with different
xmm registers. Instead of just doing 1
rmuir commented on PR #12737:
URL: https://github.com/apache/lucene/pull/12737#issuecomment-1786392160
with all the data dependencies removed, i also gave at least one stab trying
to see if i could trick the compiler into using packed instructions instead of
single floats... would be awesom
romseygeek commented on PR #12736:
URL: https://github.com/apache/lucene/pull/12736#issuecomment-1786088337
This looks great, thank you @daviscook477! Would you be able to add an
entry to CHANGES.txt under the 9.9.0 release?
--
This is an automated message from the Apache Git Service.
T
benwtrent commented on PR #12729:
URL: https://github.com/apache/lucene/pull/12729#issuecomment-1786059581
@jpountz updated. Flat is no longer pluggable, two HNSW formats are exposed.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to
daviscook477 opened a new pull request, #12736:
URL: https://github.com/apache/lucene/pull/12736
### Description
The [javadoc for
Monitor.getQuery](https://github.com/apache/lucene/blob/a0887c7d26df6c9f32afcf8e9f0ff66275115f92/lucene/monitor/src/java/org/apache/lucene/monitor/Monitor.java
nvnmandadhi closed pull request #399: fix explicit type declaration
URL: https://github.com/apache/lucene-solr/pull/399
--
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 unsub
gsmiller closed issue #12319: DaciukMihovAutomatonBuilder#build should probably
take a List instead of a Collection
URL: https://github.com/apache/lucene/issues/12319
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
UR
gsmiller merged PR #12427:
URL: https://github.com/apache/lucene/pull/12427
--
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.ap
gsmiller commented on PR #12427:
URL: https://github.com/apache/lucene/pull/12427#issuecomment-1785924607
+1 looks good to me as well. I like that this small change, 1) makes the API
a little more general, allowing users to provide any Iterable instead of
Collection, and 2) adds an explicit
risdenk merged PR #2681:
URL: https://github.com/apache/lucene-solr/pull/2681
--
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.
jpountz commented on PR #12729:
URL: https://github.com/apache/lucene/pull/12729#issuecomment-1785712360
Thanks, splitting the way you describe would make me happy.
I had not understood that the flat codec was a goal. Now that I think more
about it, I wonder if we should better separa
mikemccand commented on issue #12714:
URL: https://github.com/apache/lucene/issues/12714#issuecomment-1785695227
> Perhaps instead of UnCompiledNode, we could encode it as byte-array (could
take the same format as the FST-encoded binary, but the FST operation works on
absolute address value
asfgit merged PR #12731:
URL: https://github.com/apache/lucene/pull/12731
--
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.apac
benwtrent commented on PR #12729:
URL: https://github.com/apache/lucene/pull/12729#issuecomment-1785563758
OK, @jpountz thinking about it more. To do what you are suggesting, I think
the following would work:
- Force Lucene99HnswVectorsReader & Lucene99HnswVectorsWriter to take a
`F
rmuir commented on PR #12731:
URL: https://github.com/apache/lucene/pull/12731#issuecomment-1785549757
> I think the Panama API should allow the user to figure out how many
parallel units are available to somehow dynamically split work correctly.
I'm not even sure openjdk/hotspot know
uschindler commented on PR #12731:
URL: https://github.com/apache/lucene/pull/12731#issuecomment-1785453474
> Last time i tried to figure out WTF was happening here, I think i
determined that floating point reproducibility was still preventing this from
happening? That there isn't like a "b
jpountz merged PR #12719:
URL: https://github.com/apache/lucene/pull/12719
--
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.apa
benwtrent commented on PR #12729:
URL: https://github.com/apache/lucene/pull/12729#issuecomment-1785376897
@jpountz the goal of this change is not just making code reusable. But:
- Allowing folks who don't want HNSW to take advantage of the per-segment
quantization and logic. Paging
jpountz commented on code in PR #12729:
URL: https://github.com/apache/lucene/pull/12729#discussion_r1376303898
##
lucene/core/src/java/org/apache/lucene/codecs/lucene99/Lucene99HnswVectorsReader.java:
##
@@ -399,41 +281,30 @@ private HnswGraph getGraph(FieldEntry entry) throws
benwtrent closed issue #12700: Should we handle negative scores due to floating
point arithmetic errors?
URL: https://github.com/apache/lucene/issues/12700
--
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
benwtrent merged PR #12727:
URL: https://github.com/apache/lucene/pull/12727
--
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.a
benwtrent merged PR #12726:
URL: https://github.com/apache/lucene/pull/12726
--
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.a
jpountz commented on code in PR #12727:
URL: https://github.com/apache/lucene/pull/12727#discussion_r1376230730
##
lucene/core/src/test/org/apache/lucene/util/TestVectorUtil.java:
##
@@ -115,6 +116,21 @@ public void testNormalizeZeroThrows() {
expectThrows(IllegalArgumentEx
benwtrent commented on PR #12726:
URL: https://github.com/apache/lucene/pull/12726#issuecomment-1785207766
@shubhamvishu I will merge and backport today
--
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
benwtrent merged PR #12728:
URL: https://github.com/apache/lucene/pull/12728
--
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.a
rmuir commented on PR #12731:
URL: https://github.com/apache/lucene/pull/12731#issuecomment-1785178856
Last time i tried to figure out WTF was happening here, I think i determined
that floating point reproducibility was still preventing this from happening?
That there isn't like a "bail out
rmuir commented on PR #12731:
URL: https://github.com/apache/lucene/pull/12731#issuecomment-1785163931
> .. and yes (I've not forgotten), we need something like a
`java.lang.Architecture/Platform`, that is queryable for such low-level support
(rather than resorting to beans - which actually
jpountz commented on PR #12727:
URL: https://github.com/apache/lucene/pull/12727#issuecomment-1785162374
I had a suspicion that the double promotion is not buying us anything in
that case, so I ran a quick test that seems to confirm it:
```java
long equals = 0;
long notEquals =
dungba88 commented on issue #12714:
URL: https://github.com/apache/lucene/issues/12714#issuecomment-1785150257
I ran a small test to see what RAM would be needed for some sample
dictionary using a simple `LinkedHashMap`:
6MB Cache size
62457 items
977KB FST size
The repor
rmuir commented on PR #12731:
URL: https://github.com/apache/lucene/pull/12731#issuecomment-1785145823
> ha! So just removing the overly aggressive unrolling in cosine improves
things.
well, only in combination with switch to FMA. seems then its able to keep
cpu busy multiplying.
benwtrent commented on PR #12727:
URL: https://github.com/apache/lucene/pull/12727#issuecomment-1785060096
@ChrisHegarty added a test for verifying VectorSimilarityFunction returns
scores `>= 0`.
--
This is an automated message from the Apache Git Service.
To respond to the message, pleas
dweiss commented on issue #12714:
URL: https://github.com/apache/lucene/issues/12714#issuecomment-1784860286
I'd check whether there's much gain from the switch first.
Fill-up-then-discard caches often perform quite well and allow for much
easier/faster implementation (both in terms of GC o
dungba88 opened a new pull request, #12735:
URL: https://github.com/apache/lucene/pull/12735
### Description
Clean-up inputCount as it no longer has an active use
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use
37 matches
Mail list logo
