[llvm] AMDGPU: Migrate more tests away from undef (PR #131314)

[Matt Arsenault via llvm-commits]

================
@@ -1189,7 +1197,9 @@ bb1:                                              ; preds = %bb3, %bb0
   %tmp0 = phi i32 [ 8, %bb0 ], [ %tmp4, %bb3 ]
   %tmp1 = add nsw i32 %tmp0, -1
   %tmp2 = getelementptr inbounds i32, ptr addrspace(3) %out, i32 %tmp1
-  br i1 undef, label %bb2, label %bb3
+  %cond = call i32 asm "; def $0","=s"()
----------------
arsenm wrote:

I tried that first but it perturbed codegen a lot more 

https://github.com/llvm/llvm-project/pull/131314