[libclc] [libclc] Use unchecked division for f64 AMDGPU reciprocal (PR #203809)

[Joseph Huber via cfe-commits]

jhuber6 wrote:

@wenju-he General question, there  are a lot of math functions that are used by CUDA/HIP but aren't defined in a standard like `libc` or `opencl`. I'm curious where you think something like this would live, I believe the functions that AMD provides for HIP that aren't included in `opencl` are:
```
  j0 j1 jn  y0 y1 yn  i0/i1
  erfinv  erfcinv  erfcx
  ncdf  ncdfinv
  rcbrt  rhypot
  nearbyint  scalbn/scalbln
```
So, basically, if we wanted to upstream all of this, it'd need to live somewhere and I'd assume `libclc` would absorb it.

https://github.com/llvm/llvm-project/pull/203809