[llvm] [AMDGPU] Add AMDGPU specific variadic operation MCExprs (PR #82022)

[Pierre van Houtryve via llvm-commits]

================
@@ -1512,6 +1512,24 @@ The AMDGPU backend supports the following calling conventions:
 
      =============================== ==========================================================
 
+AMDGPU MCExpr
+-------------
+
+As part of the AMDGPU MC layer, AMDGPU provides the following target specific
+``MCExpr``\s.
+
+  .. table:: AMDGPU MCExpr types:
+     :name: amdgpu-mcexpr-table
+
+     =================== ======================================================================
----------------
Pierre-vh wrote:

I'd just make 4 columns, these are essentially functions after all :
- Expr 
- Number of operands
- Return value
- Eventual notes (can be omitted at this time as none of these are tricky)

So for this case you'd have "1 or more" in the operands column, and the return value just says "the maximum value between all arguments" and "all arguments bitwise-or'd together" (not sure about the wording for the latter)

https://github.com/llvm/llvm-project/pull/82022