[PATCH] D53670: Revert "Teach __libcpp_is_floating_point that __fp16 and _Float16 are"

[Louis Dionne via Phabricator]

ldionne added a comment.

In my opinion, the proper way forward is to standardize a notion of "extended floating point types" and establish a set of things that those must support. Then, when vendors want to provide such types, they can make sure that we (the library implementation) can implement the required functionality for their custom types to be proper floating point types. This would mean `abs` , `isnan`, `numeric_limits`, etc...

We don't have such a notion right now. If we did, we could arguably make `_Float16` and `__float128` such extended floating point types, but in all cases `__fp16` would be dead in the water. I hear that someone is going to propose such floating point types -- I would be inclined to wait and see how this pans out before we spend more cycles trying to figure the issue out ourselves.


Repository:
  rL LLVM

https://reviews.llvm.org/D53670