[llvm] [AMDGPU] Add verification for amdgcn.init.exec.from.input (PR #128172)

[Matt Arsenault via llvm-commits]

================
@@ -18,10 +17,3 @@ define amdgpu_ps void @init_exec_from_input_fail_not_inreg(i32 inreg %a, i32 %b)
   call void @llvm.amdgcn.init.exec.from.input(i32 %b, i32 0)
   ret void
 }
----------------
arsenm wrote:

Also test a case where it's not an argument (try one with a constant and one with an instruction) 

https://github.com/llvm/llvm-project/pull/128172