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

[Matt Arsenault via llvm-commits]

================
@@ -1511,7 +1511,7 @@ define amdgpu_kernel void @test_preserve_condition_undef_flag(float %arg, i32 %a
 ; GFX1064-NEXT:    s_endpgm
 bb0:
   %tmp = icmp sgt i32 %arg1, 4
-  %undef = call i1 @llvm.amdgcn.class.f32(float poison, i32 undef)
+  %undef = call i1 @llvm.amdgcn.class.f32(float poison, i32 0)
----------------
arsenm wrote:

This case is not great because it's using the class as a loophole to get an undef value. It doesn't matter for the purposes of this test since it should still fold to undef from the first operand 

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