[libc] [llvm] [libc][math][c23] Add fsqrtf16 C23 math functions. (PR #131153)

[Harrison Hao via llvm-commits]

harrisonGPU wrote:

> > > The function with `float16` input and `float16` output should be named `sqrtf16`, not `fsqrtf16`.
> > 
> > 
> > Hi Tue. Do you mean that `fsqrtf16` takes a float as input and returns a float16? I also noticed that `fmodf16` uses float16 for both its input and output. So I am not sure.
> 
> No, `fmod` is a family of functions. You have `fmod, fmodf, fmodl, fmodf16, fmodf128`. And those have output types the same as input types. Similarly, `sqrt` is a family of functions. You have `sqrt, sqrtf, sqrtl, sqrtf16, sqrtf128`, which have output types the same as input types.
> 
> You also have newer C23 functions that have output types narrower than input types. Like `fsqrt` is related to `sqrt`, taking `double` input, but returning a `float` (hence `f` prefix), or `f16sqrtf` is related to `sqrtf`, taking `float` input (hence `f` suffix) and returning a `float16`(hence `f16` prefix). Since they are supposed to have narrower output types, you don't have `fsqrtf16` which would be interpreted as taking `float16` input and returning a `float`.

Thanks, Tue. I really understand what you mean, so we don't need to implement fsqrtf16.

https://github.com/llvm/llvm-project/pull/131153