[clang] [Clang][AMDGPU] Add ``amdgcn_av("none")`` attribute for atomic expressions (PR #199622)

[Sameer Sahasrabuddhe via cfe-commits]

================
@@ -2532,6 +2532,12 @@ def AMDGPUMaxNumWorkGroups : InheritableAttr {
   let Subjects = SubjectList<[Function], ErrorDiag, "kernel functions">;
 }
 
+def AMDGCNAV : StmtAttr, TargetSpecificAttr<TargetAMDGPU> {
+  let Spellings = [Clang<"amdgcn_av">];
----------------
ssahasra wrote:

That leaves us in a very weird place. What I see is the following:

- Clang attributes have the `amdgpu_` prefix.
- Clang builtins have the `amdgcn_` prefix.
- The LLVM IR metadata for this feature is `amdgcn_av`.

So if I follow other attributes and rename this to `amdgpu_av`, that will look really out of place.

https://github.com/llvm/llvm-project/pull/199622