jpountz commented on PR #14275:
URL: https://github.com/apache/lucene/pull/14275#issuecomment-2748960716

   I've been thinking a bit more about this. The two potential use-cases that 
come to mind are the following:
    - Diverging from the way Lucene does memory management (off-heap / on-heap 
and related stuff).
    - Using more optimized logic for some routines (prefix sums, vector 
comparisons, etc.), e.g. using architecture-specific native code that the JVM 
could hardly generate itself.
   
   As discussed above, I want codecs to own the way memory is managed, so I 
don't like giving options to customize it.
   
   Regarding native optimizations, Lucene core is unlikely to ever include 
native extensions, so I can see an argument for it given the number of 
limitations of the vector API that we already found. But then I don't think 
that codecs are the right extension point for this sort of things, they are too 
high-level and would force re-implementing (copy-pasting in practice) the whole 
read-side logic, just optimizing the relevant bits. I would rather like it to 
look like `DefaultVectorizationProvider`/`PanamaVectorizationProvider` where 
optimized implementations of some performance-critical routines get plugged in 
it runtime in a way that is orthogonal to codecs?


-- 
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

Reply via email to