[llvm] [AMDGPU] Promote nestedGEP allocas to vectors (PR #141199)

[Carl Ritson via llvm-commits]

================
@@ -51,17 +51,12 @@ entry:
   ret void
 }
 
-; Check code sequence for amdpal use of scratch for alloca in a compute shader.
-; The scratch descriptor is loaded from offset 0x10 of the GIT, rather than offset
-; 0 in a graphics shader.
-; Prior to GCN3 s_load_dword offsets are dwords, so the offset will be 0x4.
+; After the change that **promotes the alloca to a vector** (GEP‑of‑GEP
+; promotion), no scratch buffer is needed, so the descriptor load should
+; disappear.
----------------
perlfu wrote:

As previously stated.  The test is intended to check scratch descriptor is correctly accessed, thus it must access scratch.  Test should not change.

https://github.com/llvm/llvm-project/pull/141199