jorgecarleitao commented on pull request #8089:
URL: https://github.com/apache/arrow/pull/8089#issuecomment-684844739


   @alamb , I did not want to open that discussion on this PR because the 
current optimization is already an improvement, regardless of our choice.
   
   However, I do agree with you: I would also prefer keeping the types. The 
main issue for me of `sqrt(f32) -> f64` is that this optimization only works 
for a single operation. E.g. the `exp` in `exp(sqrt(f32))` is evaluated with 
input `f64` because `sqrt` is mapped to `f64`. In other words, `f32` is not 
really used in computations, as it is quickly absorbed into `f64`.
   
   OTOH, as @andygrove pointed out in another thread, spark returns a 
consistent `f64` regardless of the input type.


----------------------------------------------------------------
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