[llvm] [AMDGPU] Add IR-level pass to rewrite away address space 7 (PR #77952)

[Krzysztof Drewniak via llvm-commits]

================
@@ -352,3 +352,107 @@ loop.exit:
 exit:
   ret float %sum
 }
+
+;; Test that the optimization works correctly for phi nodes that repeat the same
----------------
krzysz00 wrote:

I'd characterize switching

```llvm
%phi.rsrc = phi [ %v.rsrc, %l1 ], [ %v.rsrc, %l2 ], ...
%phi.off = phi [ %v1.off, %l1 ], [ %v2.off, %l2 ], ...
```
(the natural lowering of `phi`)

to 
```llvm
%phi.rsrc = %v.rsrc
%phi.off = [ %v1.off, %l1 ], [ %v2.off, %l2 ]
```
as an optimization.

https://github.com/llvm/llvm-project/pull/77952