[llvm] [AMDGPU] Remove the AnnotateKernelFeatures pass (PR #130198)

[Jun Wang via llvm-commits]

================
@@ -20,11 +20,14 @@ define amdgpu_kernel void @lds_atomic_dec_ret_i32(ptr addrspace(1) %out, ptr add
 ; CI-NEXT:    s_load_dwordx2 s[0:1], s[8:9], 0x0
 ; CI-NEXT:    v_mov_b32_e32 v0, 42
 ; CI-NEXT:    s_mov_b32 m0, -1
+; CI-NEXT:    s_add_i32 s12, s12, s17
 ; CI-NEXT:    s_waitcnt lgkmcnt(0)
 ; CI-NEXT:    v_mov_b32_e32 v1, s2
 ; CI-NEXT:    ds_dec_rtn_u32 v2, v1, v0
 ; CI-NEXT:    s_waitcnt lgkmcnt(0)
 ; CI-NEXT:    v_mov_b32_e32 v0, s0
+; CI-NEXT:    s_mov_b32 flat_scratch_lo, s13
----------------
jwanggit86 wrote:

> Maybe add the attribute to those test cases to avoid updating these check lines

Do you mean something like this:
```diff
 attributes #0 = { nounwind speculatable willreturn memory(none) }
-attributes #1 = { nounwind }
+attributes #1 = { nounwind "amdgpu-no-flat-scratch-init"}
 attributes #2 = { nounwind memory(none) }

 !0 = !{i32 5, i32 6}
```

https://github.com/llvm/llvm-project/pull/130198