fix custom op tutorial.

tutorials/source_zh_cn/advanced_use/custom_operator.md

@@ -38,7 +38,7 @@
 - 输入输出的名称通过`init_prim_io_names()`函数定义。
 - 输出Tensor的shape推理方法在`infer_shape()`函数中定义，输出Tensor的dtype推理方法在`infer_dtype()`函数中定义。
 
-自定义算子与内置算子的唯一区别是需要通过在`__init__()`函数中导入算子实现函数(`from .square_impl import CusSquareImpl`)来将算子实现注册到后端。本用例在`square_impl.py`中定义了算子实现和算子信息，将在后文中说明。
+自定义算子与内置算子的唯一区别是需要通过在`__init__()`函数中导入算子实现函数(`from square_impl import CusSquareImpl`)来将算子实现注册到后端。本用例在`square_impl.py`中定义了算子实现和算子信息，将在后文中说明。
 
 以Square算子原语`cus_square.py`为例，给出如下示例代码。
 
@@ -53,7 +53,7 @@ class CusSquare(PrimitiveWithInfer):
     @prim_attr_register
     def __init__(self):
         self.init_prim_io_names(inputs=['x'], outputs=['y'])
-        from .square_impl import CusSquareImpl # Import the entry function of the kernel implementation from relative path or PYTHONPATH.
+        from square_impl import CusSquareImpl # Import the entry function of the kernel implementation from relative path or PYTHONPATH.
 
     def infer_shape(self, data_shape):
         return data_shape
@@ -155,7 +155,7 @@ import mindspore.nn as nn
 import mindspore.context as context
 from mindspore import Tensor
 # Import the definition of the CusSquare primtive.
-from .cus_square import CusSquare
+from cus_square import CusSquare
 context.set_context(mode=context.GRAPH_MODE, device_target="Ascend")
 
 class Net(nn.Cell):
@@ -200,7 +200,7 @@ class CusSquare(PrimitiveWithInfer):
     def __init__(self):
         """init CusSquare"""
         self.init_prim_io_names(inputs=['x'], outputs=['y'])
-        from .square_impl import CusSquareImpl
+        from square_impl import CusSquareImpl
 
     def infer_shape(self, data_shape):
         return data_shape