dungba88 commented on issue #12714:
URL: https://github.com/apache/lucene/issues/12714#issuecomment-1788492690
An updated statistics based on the new implementation.
RAM Limit | Actual cache size | Cached nodes | FST size
---|---|---|---
No limit | 1.4MB | 62457 | 977KB
1MB |
rmuir opened a new pull request, #12743:
URL: https://github.com/apache/lucene/pull/12743
I was looking at this curious little guy which seems close to what we want?
https://github.com/openjdk/jdk/blob/b3fec6b5f32c338ae1a84dd20bdcbd3d9b7186f3/src/hotspot/cpu/aarch64/aarch64_vector_ad.m4#L137
dungba88 commented on PR #12738:
URL: https://github.com/apache/lucene/pull/12738#issuecomment-1788406285
Ok, it's ready for review. I'll add the CHANGES.txt entry once it's approved.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to G
dungba88 commented on code in PR #12738:
URL: https://github.com/apache/lucene/pull/12738#discussion_r1378381108
##
lucene/core/src/java/org/apache/lucene/util/fst/NodeHash.java:
##
@@ -186,119 +194,85 @@ private long hash(FSTCompiler.UnCompiledNode node) {
return h;
}
rmuir commented on PR #12737:
URL: https://github.com/apache/lucene/pull/12737#issuecomment-1788347778
Uwe i realize i may be short with my responses, I ask:
0. get your espresso before continuing.
1. please take the time to look at benchmark results in depth. **make** sure
you see the
rmuir commented on PR #12737:
URL: https://github.com/apache/lucene/pull/12737#issuecomment-1788323948
Given this PR has no regressions for anyone and only causes speedups, I will
pause it for a few weeks. please take the time to read and digest the results.
Uwe, your FMA is slow too
rmuir commented on PR #12737:
URL: https://github.com/apache/lucene/pull/12737#issuecomment-1788320396
also @uschindler i think you read your own results wrong, you can see the
regression with FMA in your case too, same as @dweiss:
```
VectorUtilBenchmark.floatDotProductScalar1
dungba88 commented on code in PR #12738:
URL: https://github.com/apache/lucene/pull/12738#discussion_r1378285803
##
lucene/core/src/java/org/apache/lucene/util/fst/NodeHash.java:
##
@@ -269,36 +283,58 @@ private boolean nodesEqual(FSTCompiler.UnCompiledNode
node, long address)
dungba88 commented on code in PR #12738:
URL: https://github.com/apache/lucene/pull/12738#discussion_r1378285803
##
lucene/core/src/java/org/apache/lucene/util/fst/NodeHash.java:
##
@@ -269,36 +283,58 @@ private boolean nodesEqual(FSTCompiler.UnCompiledNode
node, long address)
dungba88 commented on PR #12738:
URL: https://github.com/apache/lucene/pull/12738#issuecomment-1788188467
I solved most of the nocommits (only 1 left). But I ended up using a
`List` where each item is a node instead of ByteBlockPool due to the
following reasons:
- With ByteBlockPool we h
slow-J commented on code in PR #12741:
URL: https://github.com/apache/lucene/pull/12741#discussion_r1378259917
##
lucene/core/src/java/org/apache/lucene/codecs/lucene99/ForDeltaUtil.java:
##
@@ -0,0 +1,86 @@
+/*
+ * Licensed to the Apache Software Foundation (ASF) under one or m
slow-J commented on code in PR #12741:
URL: https://github.com/apache/lucene/pull/12741#discussion_r1378259917
##
lucene/core/src/java/org/apache/lucene/codecs/lucene99/ForDeltaUtil.java:
##
@@ -0,0 +1,86 @@
+/*
+ * Licensed to the Apache Software Foundation (ASF) under one or m
dweiss commented on issue #12742:
URL: https://github.com/apache/lucene/issues/12742#issuecomment-1788129507
I think the reason for an up-to-date status is because we handle module path
ourselves
```
// Add modular dependencies and their transitive dependencies to
module path.
dweiss commented on issue #12742:
URL: https://github.com/apache/lucene/issues/12742#issuecomment-1788119452
It's not incremental compilation - it's something that is related to task
inputs and (likely) modular dependencies. Seems like we're missing something
that makes the test compilation
uschindler commented on PR #12737:
URL: https://github.com/apache/lucene/pull/12737#issuecomment-1788100282
Let's remove FMA support in this pull request.
I will add a follow-up PR to have a PKG private utility class to query VM
options that also logs a warning consistent (like in ram
robro612 commented on issue #12449:
URL: https://github.com/apache/lucene/issues/12449#issuecomment-1788067055
Hi @asubbu90
My name is Rohan Jha, I'm a Masters student at UT Austin taking a graduate
[Distributed Systems course](https://github.com/vijay03/cs380d-f23). As part of
my c
robro612 commented on issue #12725:
URL: https://github.com/apache/lucene/issues/12725#issuecomment-1788065989
Hi @gf2121 ,
My name is Rohan Jha, I'm a Masters student at UT Austin taking a graduate
[Distributed Systems course](https://github.com/vijay03/cs380d-f23). As part of
my co
dweiss opened a new issue, #12742:
URL: https://github.com/apache/lucene/issues/12742
### Description
Recent builds failed with:
```
org.apache.lucene.search.uhighlight.TestLengthGoalBreakIterator >
testTargetLen FAILED
java.lang.NoSuchMethodError: 'org.apache.lucene.util
rmuir commented on PR #12737:
URL: https://github.com/apache/lucene/pull/12737#issuecomment-1788017728
good point about the C2 detection! If the bigdecimal code gets used it is a
real problem. I hit this case on accident while developing this PR and suddenly
saw JMH outputting scientific no
rmuir commented on PR #12737:
URL: https://github.com/apache/lucene/pull/12737#issuecomment-1787992450
> My AMD, which is a server CPU (in contrast to Threadripper, which is a
workstation CPU), works faster with FMA. How can I enable it with your patch?
You disable it without any chance to
uschindler commented on PR #12737:
URL: https://github.com/apache/lucene/pull/12737#issuecomment-1787985763
Anyways there's something problematic:
FMA is slow with C2 disabled. In Panama Vector mode we detect this earlier
and fallback to scalar variant. But with this patch the scalar
uschindler commented on PR #12737:
URL: https://github.com/apache/lucene/pull/12737#issuecomment-1787973537
Hi,
> So I'd like to avoid use of FMA on AMD, where some cpus it causes
slowdowns compared to mul+add which allows more parallelism due to architecture.
If somebody notic
slow-J commented on PR #12741:
URL: https://github.com/apache/lucene/pull/12741#issuecomment-1787958483
Thanks for the suggestion, I added `Lucene90RWPostingsFormat` in latest
commit and made `Lucene90PostingsFormat` read-only.
--
This is an automated message from the Apache Git Service.
slow-J commented on code in PR #12741:
URL: https://github.com/apache/lucene/pull/12741#discussion_r1378112509
##
lucene/core/src/java/org/apache/lucene/codecs/lucene99/PForUtil.java:
##
@@ -0,0 +1,558 @@
+/*
+ * Licensed to the Apache Software Foundation (ASF) under one or more
slow-J commented on code in PR #12741:
URL: https://github.com/apache/lucene/pull/12741#discussion_r1378112295
##
lucene/core/src/java/org/apache/lucene/codecs/lucene99/ForDeltaUtil.java:
##
@@ -0,0 +1,86 @@
+/*
+ * Licensed to the Apache Software Foundation (ASF) under one or m
slow-J commented on code in PR #12741:
URL: https://github.com/apache/lucene/pull/12741#discussion_r1378112072
##
lucene/CHANGES.txt:
##
@@ -104,6 +104,8 @@ Optimizations
* GITHUB#12552: Make FSTPostingsFormat load FSTs off-heap. (Tony X)
+* GITHUB#12696: Change Postings ba
rmuir commented on PR #12737:
URL: https://github.com/apache/lucene/pull/12737#issuecomment-1787943673
I don't want to commit changes that slow down some people's computers. This
is a way to go about it that is safe and sane. I view this one the same as the
intel downclocking issue (where
uschindler commented on PR #12737:
URL: https://github.com/apache/lucene/pull/12737#issuecomment-1787795291
To prevent any surprises, I backported the branch and also the benchmark
module to 9.x and ran everything with Java 11.0.13 (on my Intel Laptop):
```
branch_9x (Java 11.0.13)
uschindler commented on PR #12737:
URL: https://github.com/apache/lucene/pull/12737#issuecomment-1787740992
Hi,
on my older Ryzen it is faster with FMA enabled (I downgraded your branch
and also verified that the system prints "FMA enabled". Here is my full
benchmark:
```
main,
jpountz merged PR #12713:
URL: https://github.com/apache/lucene/pull/12713
--
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
jpountz commented on issue #12717:
URL: https://github.com/apache/lucene/issues/12717#issuecomment-1787646282
Since we're changing the postings format anyway in
https://github.com/apache/lucene/pull/12741, I wonder if it would be worth
looking into different encodings for these tail posting
jpountz commented on code in PR #12741:
URL: https://github.com/apache/lucene/pull/12741#discussion_r1377911285
##
lucene/core/src/java/org/apache/lucene/codecs/lucene99/ForDeltaUtil.java:
##
@@ -0,0 +1,86 @@
+/*
+ * Licensed to the Apache Software Foundation (ASF) under one or
slow-J commented on issue #12696:
URL: https://github.com/apache/lucene/issues/12696#issuecomment-1787521225
Thanks all for the feedback. Will proceed with removing patching only for
doc blocks (reverting some of https://github.com/apache/lucene/pull/69)
All the changes needed to crea
slow-J opened a new pull request, #12741:
URL: https://github.com/apache/lucene/pull/12741
We are still keeping PFOR for positions only.
This is a partial revert of https://github.com/apache/lucene/pull/69 which
brings back ForDeltaUtil.
Closes #12696
--
This is an autom
rmuir commented on PR #12737:
URL: https://github.com/apache/lucene/pull/12737#issuecomment-1787491191
thanks for benchmarking everyone, really helpful. I'm glad @dweiss caught
the AMD issue.
--
This is an automated message from the Apache Git Service.
To respond to the message, please lo
dungba88 commented on code in PR #12738:
URL: https://github.com/apache/lucene/pull/12738#discussion_r138426
##
lucene/core/src/java/org/apache/lucene/util/fst/NodeHash.java:
##
@@ -214,7 +222,13 @@ private long hash(long node) throws IOException {
* Compares an unfrozen
ChrisHegarty commented on PR #12737:
URL: https://github.com/apache/lucene/pull/12737#issuecomment-1787445189
> > I see the scalar code vectorize, but not optimally for the target CPU -
e.g. `vfmadd231ss %xmm9,%xmm10,%xmm4` on my Rocket Lake. Where as, the vector
API compilation emits instr
romseygeek commented on PR #12736:
URL: https://github.com/apache/lucene/pull/12736#issuecomment-1787418950
Thanks @daviscook477!
--
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 com
zhaih opened a new issue, #12740:
URL: https://github.com/apache/lucene/issues/12740
### Description
Follow up of #12660, currently the HNSW concurrent merge need 2 parameters
to make it work: `numMergeWorker` (num of thread spawn per merge) and an
`ExecutorService` that is used to e
rmuir commented on PR #12737:
URL: https://github.com/apache/lucene/pull/12737#issuecomment-1787409152
> I see the scalar code vectorize, but not optimally for the target CPU -
e.g. `vfmadd231ss %xmm9,%xmm10,%xmm4` on my Rocket Lake. Where as, the vector
API compilation emits instructions t
romseygeek merged PR #12736:
URL: https://github.com/apache/lucene/pull/12736
--
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.
ChrisHegarty commented on PR #12737:
URL: https://github.com/apache/lucene/pull/12737#issuecomment-1787403081
> 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
zhaih commented on PR #12660:
URL: https://github.com/apache/lucene/pull/12660#issuecomment-1787383067
> For my understanding, how does it play out with merge throttling, could we
end up sleeping within tasks that get passed to the configured executor service?
I think we rate limit th
mikemccand commented on code in PR #12738:
URL: https://github.com/apache/lucene/pull/12738#discussion_r1377721562
##
lucene/core/src/java/org/apache/lucene/util/fst/NodeHash.java:
##
@@ -269,36 +283,58 @@ private boolean nodesEqual(FSTCompiler.UnCompiledNode
node, long address
mikemccand commented on PR #12737:
URL: https://github.com/apache/lucene/pull/12737#issuecomment-1787353298
OK I ran again this command: `./gradlew -p lucene/benchmark-jmh assemble &&
java -jar
lucene/benchmark-jmh/build/benchmarks/lucene-benchmark-jmh-10.0.0-SNAPSHOT.jar
float -p size=102
mikemccand commented on PR #12737:
URL: https://github.com/apache/lucene/pull/12737#issuecomment-1787346877
> Why is vector and scalar identical? Cause it is java 17 or am I missing
something.
Sorry, yeah, I ran with Java 17. I'm running again with Java 20 now ...
> Please ad
uschindler commented on PR #12737:
URL: https://github.com/apache/lucene/pull/12737#issuecomment-1787335532
> Thanks @rmuir!
>
> Results from the nightly benchy box (128 core Ryzen Threadripper 3990X):
>
> `main`:
>
> ```
> Benchmark (s
ChrisHegarty commented on PR #12737:
URL: https://github.com/apache/lucene/pull/12737#issuecomment-1787326935
Just some more numbers, for sanity.
Intel Rocket Lake:
main:
```
VectorUtilBenchmark.floatCosineScalar1024 thrpt5 0.844 ± 0.007
ops/us
VectorUti
mikemccand commented on PR #12737:
URL: https://github.com/apache/lucene/pull/12737#issuecomment-1787292187
Thanks @rmuir!
Results from the nightly benchy box (128 core Ryzen Threadripper 3990X):
`main`:
```
Benchmark (size) Mode Cnt
uschindler commented on PR #12737:
URL: https://github.com/apache/lucene/pull/12737#issuecomment-1787287133
> This should solve @dweiss problem:
[f2be84f](https://github.com/apache/lucene/commit/f2be84f822b6c2c7971fa1d434ad82c09174039f)
>
> It should also improve speed of vectorized c
daviscook477 commented on PR #12736:
URL: https://github.com/apache/lucene/pull/12736#issuecomment-1787229329
@romseygeek I have added a CHANGES.txt entry under the 9.9.0 release
describing the bugfix in this PR.
--
This is an automated message from the Apache Git Service.
To respond to t
rmuir commented on PR #12737:
URL: https://github.com/apache/lucene/pull/12737#issuecomment-1787195593
This should solve @dweiss problem:
https://github.com/apache/lucene/pull/12737/commits/f2be84f822b6c2c7971fa1d434ad82c09174039f
It should also improve speed of vectorized case on AMD
uschindler commented on PR #12737:
URL: https://github.com/apache/lucene/pull/12737#issuecomment-1787190466
Please keep FMA for now and allow us benchmarking.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL abo
jpountz commented on issue #12486:
URL: https://github.com/apache/lucene/issues/12486#issuecomment-1787160140
> Lucene is also doing the "accumulate docid deltas into the absolute docid"
too in this loop, but I guess Tantivy does this separately somehow?
I believe Tantivy does the sam
rmuir commented on PR #12737:
URL: https://github.com/apache/lucene/pull/12737#issuecomment-1787121130
we can also detect newer zen (e.g. zen4) by it having 512-bit vectors. so we
just need to detect AMD vs Intel.
anyway, i'm fine with removing FMA from this PR completely actually, an
rmuir commented on PR #12737:
URL: https://github.com/apache/lucene/pull/12737#issuecomment-1787102205
> You can't get zen version or detect Ryzen CPUs easily. Please do not add
cpuinfo parsing as this won't work on alternate platforms like Windows.
We can just retrieve another flag t
stefanvodita commented on code in PR #12547:
URL: https://github.com/apache/lucene/pull/12547#discussion_r1377483050
##
lucene/facet/src/java/org/apache/lucene/facet/taxonomy/FloatTaxonomyFacets.java:
##
@@ -37,33 +37,43 @@ abstract class FloatTaxonomyFacets extends TaxonomyFace
uschindler commented on PR #12737:
URL: https://github.com/apache/lucene/pull/12737#issuecomment-1787077812
I will check on my Zen on policeman Jenkins later.
You can't get zen version or detect Ryzen CPUs easily. Please do not add
cpuinfo parsing as this won't work on alternate platf
mikemccand commented on code in PR #12547:
URL: https://github.com/apache/lucene/pull/12547#discussion_r1377455225
##
lucene/facet/src/java/org/apache/lucene/facet/taxonomy/FloatTaxonomyFacets.java:
##
@@ -37,33 +37,43 @@ abstract class FloatTaxonomyFacets extends TaxonomyFacets
rmuir commented on PR #12737:
URL: https://github.com/apache/lucene/pull/12737#issuecomment-1787047135
hmm i think i read @dweiss results in the wrong order... it seems like a
fairly big difference? we actually regress scalar dotproduct for his cpu here.
And I assume the vector case behaves
mikemccand commented on PR #12716:
URL: https://github.com/apache/lucene/pull/12716#issuecomment-1787033818
Wow, these are interesting results. Thanks @shubhamvishu. It's curious
that, at 3/4 rehash ratio, linear probing is quite a bit slower (103.4 s) than
quadratic (98.1 s).
The
uschindler commented on PR #12737:
URL: https://github.com/apache/lucene/pull/12737#issuecomment-1787029825
Yeah. Please keep it as it gives more correct results.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL
rmuir commented on PR #12737:
URL: https://github.com/apache/lucene/pull/12737#issuecomment-1787025516
> Nice. I checked with AMD Ryzen Threadripper 3970X. Note it's actually
slightly faster when not using FMA...
@dweiss thanks for testing, I expect this on AMD. but "slightly" is very
dweiss commented on code in PR #12738:
URL: https://github.com/apache/lucene/pull/12738#discussion_r1377416721
##
lucene/core/src/java/org/apache/lucene/util/fst/ReverseBytesReader.java:
##
@@ -17,7 +17,7 @@
package org.apache.lucene.util.fst;
/** Reads in reverse from a sin
jpountz commented on issue #12696:
URL: https://github.com/apache/lucene/issues/12696#issuecomment-1786969801
> Normally the IntNRQ (1D points numeric range query) is very noisy, but
maybe this gain is real? p-value seems to think it could be close to real?
I'm not sure how it could n
dungba88 commented on code in PR #12738:
URL: https://github.com/apache/lucene/pull/12738#discussion_r1377383960
##
lucene/core/src/java/org/apache/lucene/util/fst/ReverseBytesReader.java:
##
@@ -17,7 +17,7 @@
package org.apache.lucene.util.fst;
/** Reads in reverse from a s
mikemccand commented on issue #12696:
URL: https://github.com/apache/lucene/issues/12696#issuecomment-1786949852
Thanks for testing @jpountz.
I think at some point we also enabled patching for the freq blocks inside
`.doc` file too?
Normally the `IntNRQ` (1D points numeric rang
jpountz merged PR #12739:
URL: https://github.com/apache/lucene/pull/12739
--
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
dungba88 commented on code in PR #12738:
URL: https://github.com/apache/lucene/pull/12738#discussion_r1377364142
##
lucene/core/src/java/org/apache/lucene/util/fst/ReverseBytesReader.java:
##
@@ -17,7 +17,7 @@
package org.apache.lucene.util.fst;
/** Reads in reverse from a s
dungba88 commented on code in PR #12738:
URL: https://github.com/apache/lucene/pull/12738#discussion_r1377360860
##
lucene/core/src/java/org/apache/lucene/util/fst/NodeHash.java:
##
@@ -269,36 +283,58 @@ private boolean nodesEqual(FSTCompiler.UnCompiledNode
node, long address)
dungba88 commented on code in PR #12738:
URL: https://github.com/apache/lucene/pull/12738#discussion_r1377358613
##
lucene/core/src/java/org/apache/lucene/util/fst/NodeHash.java:
##
@@ -269,36 +283,58 @@ private boolean nodesEqual(FSTCompiler.UnCompiledNode
node, long address)
dungba88 commented on code in PR #12738:
URL: https://github.com/apache/lucene/pull/12738#discussion_r1377355083
##
lucene/core/src/java/org/apache/lucene/util/fst/NodeHash.java:
##
@@ -269,36 +283,58 @@ private boolean nodesEqual(FSTCompiler.UnCompiledNode
node, long address)
dungba88 commented on code in PR #12738:
URL: https://github.com/apache/lucene/pull/12738#discussion_r1377351934
##
lucene/core/src/java/org/apache/lucene/util/fst/FSTCompiler.java:
##
@@ -749,7 +750,6 @@ public void add(IntsRef input, T output) throws IOException
{
// f
dungba88 commented on code in PR #12738:
URL: https://github.com/apache/lucene/pull/12738#discussion_r1377348991
##
lucene/core/src/java/org/apache/lucene/util/fst/FSTCompiler.java:
##
@@ -461,9 +457,14 @@ long addNode(FSTCompiler.UnCompiledNode nodeIn) throws
IOException {
mikemccand commented on code in PR #12738:
URL: https://github.com/apache/lucene/pull/12738#discussion_r1377320240
##
lucene/core/src/java/org/apache/lucene/util/fst/FSTCompiler.java:
##
@@ -461,9 +457,14 @@ long addNode(FSTCompiler.UnCompiledNode nodeIn) throws
IOException {
uschindler commented on PR #12737:
URL: https://github.com/apache/lucene/pull/12737#issuecomment-1786880847
Thanks, great. By the way if you run the incubator impl it prints the FMA
status as log message. Just the scalar one doesn't.
--
This is an automated message from the Apache Git Ser
dweiss commented on PR #12737:
URL: https://github.com/apache/lucene/pull/12737#issuecomment-1786855076
Here's how JMH copies runner VM's input arguments -
```
jvmArgs.addAll(options.getJvmArgs()
.orElseGet(() -> benchmark.getJvmArgs()
.orEl
dweiss commented on PR #12737:
URL: https://github.com/apache/lucene/pull/12737#issuecomment-1786839902
Out of curiosity I checked (by sysouting the status of the constant):
-XX:+-UseFMA is passed to forked JVMs, so these incantations of jmh are
equivalent:
```
java -XX:-UseFMA -jar
jpountz commented on PR #12660:
URL: https://github.com/apache/lucene/pull/12660#issuecomment-1786816049
For my understanding, how does it play out with merge throttling, could we
end up sleeping within tasks that get passed to the configured executor service?
--
This is an automated mess
dweiss commented on PR #12737:
URL: https://github.com/apache/lucene/pull/12737#issuecomment-1786815134
JMH states that:
```
-jvmArgsUse given JVM arguments. Most options are
inherited
from the host VM options, [...]
```
and I've alw
jpountz opened a new pull request, #12739:
URL: https://github.com/apache/lucene/pull/12739
If `mergeQuantizedByteVectorValues` fails with an exception, the temp output
never gets closed. This was found by the test that throws random exceptions.
--
This is an automated message from the Ap
uschindler commented on PR #12737:
URL: https://github.com/apache/lucene/pull/12737#issuecomment-1786790886
@dweiss Are you sure the `-useFMA` was applied to the child jvms of
benchmark. As far as I remember you have to pass the JVM options as separate
parameter to be applied to childs.
-
dweiss commented on PR #12737:
URL: https://github.com/apache/lucene/pull/12737#issuecomment-1786776417
Nice. I checked with AMD Ryzen Threadripper 3970X. Note it's actually
slightly faster when not using FMA...
```
JDK 17, patch
java -jar
lucene\benchmark-jmh\build\benchmarks\
uschindler commented on PR #12737:
URL: https://github.com/apache/lucene/pull/12737#issuecomment-1786711148
Did you aldo test this PR with java 17? Your benchmark does not say which
version it uses.
--
This is an automated message from the Apache Git Service.
To respond to the message, pl
uschindler commented on PR #12737:
URL: https://github.com/apache/lucene/pull/12737#issuecomment-1786709832
One thing: we should test this in lucebe 9.x with java 11, too. I want to
make sure we get same or similar speedup there. Not that there will be a
regression.
--
This is an automat
dungba88 opened a new pull request, #12738:
URL: https://github.com/apache/lucene/pull/12738
### Description
Fix #12714
First attempt to introduce value-based LRU cache in NodeHash. There are some
inefficiencies, but the functionalities work.
--
This is an automated message
dungba88 commented on issue #12714:
URL: https://github.com/apache/lucene/issues/12714#issuecomment-1786608097
I think that makes sense.
I attempted to implement the copy bytes (not optimizing though, and there
are lots of non-optimal bytes read/write).
With the same FST as abo
87 matches
Mail list logo
