How to prevent ops to be decomposed in custom compile backend

compiler
1

Take Batchnorm2d as an example.
My bachend is this:

def xtorch_compile_backend(
    gm: torch.fx.GraphModule, example_inputs: Sequence[Any], **kwargs
) -> torch.nn.Module:

    try:
        fake_mode = detect_fake_mode(example_inputs)

        with unittest.mock.patch.object(
            fake_mode, "allow_non_fake_inputs", True
        ), fake_mode:
            gm = apply_pre_aot_passes(gm)
            logger.debug(f"After pre-AOT passes: \n{gm.graph}")

            gm = aot_export_joint_simple(
                gm,
                example_inputs,
                trace_joint=False,
                decompositions=get_decompositions(),
            )
            logger.debug(f"After aot export: \n{gm.graph}")

            gm = apply_post_aot_passes(gm)
            logger.debug(f"After post-AOT and remove_sym_nodes passes: \n{gm.graph}")
            converted_module = convert_module(gm, example_inputs, options)

            return converted_module

I need convert placeholders to get_attr, thus i use aot_export_joint_simple .
torch.nn.functional.batch_norm will be conver to torch.ops.aten._native_batch_norm_legit.no_stats, which is same for many other ops. How to prevent this.