emkornfield commented on pull request #8320:
URL: https://github.com/apache/arrow/pull/8320#issuecomment-702615585


   > I also notice that we call internal::GreaterThanBitmap for each 64 levels, 
which always goes through the dynamic dispatch indirection (meaning two 
function calls, I think). We could call GreaterThanBitmapImpl but that requires 
compiling a specialized version of level_conversion_inc.h for AVX2, otherwise 
we lose performance.
   
   yeah it isn't ideal, it is possible there is a better factoring in there but 
it seemed hard to do and isolate BMI2 special instructions, I guess if this 
isn't too much slower then BMI2 on intel we could potentially collapse 
everything, but I would not expect that to be the case.


----------------------------------------------------------------
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.

For queries about this service, please contact Infrastructure at:
[email protected]


Reply via email to