[AMDGPU] Add back the missed barrier_arrive builtin definition

Author: changpeng

clang/include/clang/Basic/BuiltinsAMDGPU.def

@@ -659,6 +659,8 @@ TARGET_BUILTIN(__builtin_amdgcn_global_store_async_from_lds_b8, "vc*1c*3IiIi", "
 TARGET_BUILTIN(__builtin_amdgcn_global_store_async_from_lds_b32, "vi*1i*3IiIi", "nc", "gfx1250-insts")
 TARGET_BUILTIN(__builtin_amdgcn_global_store_async_from_lds_b64, "vV2i*1V2i*3IiIi", "nc", "gfx1250-insts")
 TARGET_BUILTIN(__builtin_amdgcn_global_store_async_from_lds_b128, "vV4i*1V4i*3IiIi", "nc", "gfx1250-insts")
+TARGET_BUILTIN(__builtin_amdgcn_ds_atomic_async_barrier_arrive_b64, "vLi*3", "nc", "gfx1250-insts")
+TARGET_BUILTIN(__builtin_amdgcn_ds_atomic_barrier_arrive_rtn_b64, "LiLi*3Li", "nc", "gfx1250-insts")
 
 TARGET_BUILTIN(__builtin_amdgcn_tensor_load_to_lds, "vV4iV8iV4iV4iIi", "nc", "gfx1250-insts")
 TARGET_BUILTIN(__builtin_amdgcn_tensor_load_to_lds_d2, "vV4iV8iIi", "nc", "gfx1250-insts")