Cannot mix cross thread reduction with Tensorize

Questions
1

Hi, I want to use Tensorize to store reduce’s result, but i got error(Cannot mix cross thread reduction with Tensorize) when building the schedule. If I delete the sch[real_output].tensorize() , the error will go away. I was wondering that tensorize and kCrossThreadReduction cannot be used at the same time? and if so, how can i use tensorize in reduce? thanks a lot!

    data_in   = op.input_tensors[0]
    data_out  = op.output(0)
    warp_size = 64
    IL        = sch.cache_read(data_in, "local", [op])
    if len(sch[data_out].op.axis) > 0:
        all_reduce = False
        num_thread = 64
        target     = tvm.target.Target.current()
        if target and (target.kind.name == "opencl" or target.kind.name == "metal"):
            # without it, CL_INVALID_WORK_GROUP_SIZE occurred when running test_topi_reduce.py
            # don't know why
            num_thread = 16
        block_x  = te.thread_axis("blockIdx.x")
        thread_x = te.thread_axis((0, num_thread), "threadIdx.x")
        thread_y = te.thread_axis((0, num_thread), "threadIdx.y")
    else:
        all_reduce = True
        num_thread = tvm.target.Target.current(allow_none=False).max_num_threads
        thread_x   = te.thread_axis((0, warp_size), "threadIdx.x")

    # Fuse and refactor the reduce axis
    fused_reduce = sch[data_out].fuse(
        *[sch[data_out].op.reduce_axis[i] for i in range(len(sch[data_out].op.reduce_axis))]
    )

    ko, ki      = sch[data_out].split(fused_reduce, factor=warp_size) #ko:outer(61),ki:inner(1024)
   
    data_out_rf = sch.rfactor(data_out, ki)       #把ki单独拎出来，并对其做归并，包含两个操作


    sch[IL].compute_at(sch[data_out_rf], sch[data_out_rf].op.reduce_axis[0])

    sch[IL].tensorize(
        IL.op.axis[-1],
        intrin_xxxx_load_tensor_image(data_in.storage_scope, "local",
            data_in.dtype, (1,), (0,), (warp_size,))
    ) 


    tx          = sch[data_out].op.reduce_axis[0] #此时的sch[data_out].op只有一个reduce_axis，负责计算输出

    sch[data_out].bind(tx, thread_x)   #把对1024个数据进行求和的任务绑定到每个线程上，实现并行化计算
 
    sch[data_out_rf].compute_at(sch[data_out], tx)    #将sch[data_out_rf]附着到目标sch[data_out]指定的tx方向，减少了一个for循环

    real_output = data_out
    result = sch[real_output].set_store_predicate(thread_x.equal(0))
    sch[real_output].tensorize(
        tx,
        intrin_xxxx_store_tensor_image(
            "local",
            real_output.storage_scope,
            real_output.dtype,
            (1,)
        )
    )