The current comparisons are based on float16 implementations which check for NAN.
The current comparisons are based on float16 implementations which check for NAN.