original-brownbear commented on PR #13894:
URL: https://github.com/apache/lucene/pull/13894#issuecomment-2426662403
I think we simply underestimate the variance luceneutil which results in
p-values that are too low. See
https://github.com/mikemccand/luceneutil/pull/308 for a suggested fix.
uschindler commented on PR #13894:
URL: https://github.com/apache/lucene/pull/13894#issuecomment-2410803086
I was a bit afraid that this might have been caused by missing "prefetch()"
implementation, but for MemorySegmentIndexInput also the random access slice
uses the correct implementatio
uschindler commented on PR #13894:
URL: https://github.com/apache/lucene/pull/13894#issuecomment-2410749511
I dont think this is a real slowdown caused by the commit here. It is more
caused by the Hotspot optimizer misinterpreting something. We should get the
assembly code from the benchmar
original-brownbear commented on PR #13894:
URL: https://github.com/apache/lucene/pull/13894#issuecomment-2410525472
Looks to me like a few inlining decisions changed (judging by the profiling
and the relative weight of inlined versions vs non-inlined versions of some
method). I'm not even e
jpountz commented on PR #13894:
URL: https://github.com/apache/lucene/pull/13894#issuecomment-2410306410
Looks like there was a slowdown on some taxo facets tasks (which use binary
doc values to store the taxonomy). `OrHighMedDayTaxoFacets`,
`AndHighHighDayTaxoFacets` and `MedTermDayTaxoFac
original-brownbear merged PR #13894:
URL: https://github.com/apache/lucene/pull/13894
--
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...
original-brownbear commented on PR #13894:
URL: https://github.com/apache/lucene/pull/13894#issuecomment-2408457832
Thanks Adrien!
--
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
