[libclc] libclc: Use frexp builtins to implement frexp for amdgpu (PR #185637)

[Matt Arsenault via cfe-commits]

================
@@ -0,0 +1,47 @@
+//===----------------------------------------------------------------------===//
----------------
arsenm wrote:

>  if a target decides to implement its own version of .inc file with same name

They should just not do that then? I can't defend against someone in the future being confused by the confusing thing they introduced. I think this is where it should be. The generic names have a library of possible implementations, the target should be naming their implementation with a target prefix 

https://github.com/llvm/llvm-project/pull/185637