andishgar commented on issue #48123:
URL: https://github.com/apache/arrow/issues/48123#issuecomment-3540605049

   >I'm not sure I agree with this interpretation. I don't think floats with 
different signs should be considered equal, even approximately.
   
   Regarding this, I used `boost::math::float_distance` to measure ULPs. In my 
test between positive and negative `denorm_min`, it shows that the distance is 
`2`. I’m not sure whether it is meaningful to compare floats with different 
signs, but if we want to compare based on ULPs, the distance between these two 
numbers is 2.
   
   >Did you measure it? On which benchmark? If there's a faster version that 
gives similar results then why not. But this seems a bit unrelated to the task 
for making Float16 comparison "more meaningful".
   
   Actually, my purpose is to change the implementation and also port it to 
Float16. However, you mentioned that I should write a benchmark to verify this 
claim. (Note that I haven’t written any benchmark yet; I am just relying on the 
findings I discovered during my search.)
   
   If I need to write a benchmark, that’s fine—I can start it. I think I should 
open a separate issue first.
   
   


-- 
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: [email protected]

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

Reply via email to