From 04f23b7105eddb838fa223a7d00572e200473aa7 Mon Sep 17 00:00:00 2001
From: wuzhanfei <wuzhanfei@baidu.com>
Date: Mon, 5 Dec 2022 10:47:05 +0000
Subject: [PATCH 01/16] assert increment

---
 .../fleet/utils/hybrid_parallel_inference.py  |   4 +-
 python/paddle/fluid/layers/control_flow.py    | 119 +-----------------
 python/paddle/fluid/layers/rnn.py             |   2 +-
 python/paddle/fluid/optimizer.py              |   8 +-
 .../auto_parallel/test_while_op_partition.py  |   4 +-
 .../fleet/hybrid_parallel_inference_helper.py |   4 +-
 .../fluid/tests/unittests/dist_transformer.py |   5 +-
 .../unittests/dygraph_to_static/test_loop.py  |   9 +-
 .../unittests/npu/test_increment_op_npu.py    |   2 +-
 .../tests/unittests/npu/test_while_op_npu.py  |   9 +-
 .../unittests/test_array_read_write_op.py     |   9 +-
 .../fluid/tests/unittests/test_assert_op.py   |  13 +-
 .../test_async_ssa_graph_executor_mnist.py    |   2 +-
 .../test_dynamic_rnn_stop_gradient.py         |   4 +-
 .../unittests/test_eager_deletion_while_op.py |   9 +-
 .../tests/unittests/test_executor_and_mul.py  |   7 +-
 .../unittests/test_fetch_lod_tensor_array.py  |   5 +-
 .../fluid/tests/unittests/test_profiler.py    |   4 +-
 .../tests/unittests/test_while_loop_op.py     |  31 ++---
 .../fluid/tests/unittests/test_while_op.py    |  11 +-
 .../tests/unittests/xpu/test_while_op_xpu.py  |   9 +-
 .../paddle/jit/dy2static/convert_operators.py |   4 +-
 22 files changed, 91 insertions(+), 183 deletions(-)

diff --git a/python/paddle/distributed/fleet/utils/hybrid_parallel_inference.py b/python/paddle/distributed/fleet/utils/hybrid_parallel_inference.py
index 49aed0862f697..13688589383e1 100644
--- a/python/paddle/distributed/fleet/utils/hybrid_parallel_inference.py
+++ b/python/paddle/distributed/fleet/utils/hybrid_parallel_inference.py
@@ -62,7 +62,7 @@ class HybridParallelInferenceHelper:
                 element_in_arr = layers.array_read(array=arr, i=step_idx)
                 # write placehold data to global lod_tensor_array,
                 # it need for send_v2 of lod_tensor_array
-                layers.increment(x=step_idx, value=1.0, in_place=True)
+                paddle.static.nn.control_flow.increment(x=step_idx, value=1.0, in_place=True)
                 layers.array_write(element_in_arr, i=step_idx, array=arr)
 
             with paddle.fluid.device_guard(f'{device}:0'):
@@ -135,7 +135,7 @@ class HybridParallelInferenceHelper:
             with while_op.block():
                 with paddle.fluid.device_guard(f'{device}:all'):
                     input = layers.array_read(array=data, i=step_idx)
-                    layers.increment(x=step_idx, value=1.0, in_place=True)
+                    paddle.static.nn.control_flow.increment(x=step_idx, value=1.0, in_place=True)
                     layers.array_write(input, i=step_idx, array=data)
 
                 with paddle.fluid.device_guard(f'{device}:0'):
diff --git a/python/paddle/fluid/layers/control_flow.py b/python/paddle/fluid/layers/control_flow.py
index cd49f94e035b8..c2cc745de52c9 100755
--- a/python/paddle/fluid/layers/control_flow.py
+++ b/python/paddle/fluid/layers/control_flow.py
@@ -54,7 +54,6 @@
 __all__ = [
     'While',
     'Switch',
-    'increment',
     'array_write',
     'array_read',
     'cond',
@@ -62,7 +61,6 @@
     'StaticRNN',
     'reorder_lod_tensor_by_rank',
     'Print',
-    'Assert',
     'is_empty',
     'case',
     'switch_case',
@@ -455,78 +453,6 @@ def Print(
     return output
 
 
-def Assert(cond, data=None, summarize=20, name=None):
-    '''
-    This API creates an op that asserts the given condition is true. If the
-    condition is false, prints the tensors in data. ``summarize`` specifies the
-    number of the elements in the tensors to print.
-
-    Args:
-        cond (Variable): The boolean condition tensor whose numel should be 1.
-        data (list|tuple, optional): list or tuple of tensors to print when
-            condition is not true. If it's ``None``, no tensor will be printed.
-            The default value is ``None``.
-        summarize (int, optional): Number of elements in the tensor to be
-            printed. If its value is -1, then all elements in the tensor will
-            be printed. The default value is 20.
-        name (str, optional): The default value is ``None`` . Normally users
-            don't have to set this parameter. For more information, please
-            refer to :ref:`api_guide_Name` .
-
-    Returns:
-        Operator: the created operation.
-
-    Raises:
-        TypeError: If ``cond`` is not boolean Variable.
-        TypeError: If ``data`` is not a list or tuple or ``None``.
-        TypeError: If ``summarize`` is not int.
-        TypeError: If ``name`` is not a string or ``None`` .
-        fluid.core.EnforceNotMet: If the condition is False in running time.
-
-    Examples:
-        .. code-block:: python
-
-            import paddle.fluid as fluid
-            import paddle.fluid.layers as layers
-
-            x = layers.fill_constant(shape=[2, 3], dtype='float32', value=2.0)
-            condition = layers.reduce_max(x) < 1.0 # False
-            layers.Assert(condition, [x], 10, "example_assert_layer")
-
-            exe = fluid.Executor()
-            try:
-                exe.run(fluid.default_main_program())
-                # Print x and throws paddle.fluid.core.EnforceNotMet exception
-                # Example printed message for x:
-                #
-                # Variable: fill_constant_0.tmp_0
-                #   - lod: {}
-                #   - place: CPUPlace()
-                #   - shape: [2, 3]
-                #   - layout: NCHW
-                #   - dtype: float
-                #   - data: [2 2 2 2 2 2]
-            except fluid.core.EnforceNotMet as e:
-                print("Assert Exception Example")
-
-    '''
-    check_variable_and_dtype(cond, "cond", ["bool"], "fluid.layers.Assert")
-    check_type(data, "data", (list, tuple, type(None)), "fluid.layers.Assert")
-    check_type(summarize, "summarize", int, "fluid.layers.Assert")
-    check_type(name, "name", (str, type(None)), "fluid.layers.Assert")
-
-    layer_name = name if name else ('assert_' + cond.name)
-    helper = LayerHelper(layer_name, **locals())
-
-    op = helper.append_op(
-        type="assert",
-        inputs={"Cond": cond, "Data": [] if data is None else list(data)},
-        attrs={"summarize": summarize},
-    )
-
-    return op
-
-
 class BlockGuard:
     """
     BlockGuard class.
@@ -1216,7 +1142,7 @@ class While:
             cond = paddle.less_than(x=i, y=loop_len)
             while_op = fluid.layers.While(cond=cond)
             with while_op.block():
-                i = fluid.layers.increment(x=i, value=1, in_place=True)
+                i = paddle.static.nn.control_flow.increment(x=i, value=1, in_place=True)
                 paddle.assign(paddle.less_than(x=i, y=loop_len), cond)
 
             exe = fluid.Executor(fluid.CPUPlace())
@@ -1244,7 +1170,7 @@ class While:
             with while_op.block():
                 sums_tensor = fluid.layers.elementwise_add(x=data, y=data)
                 fluid.layers.assign(sums_tensor, sums)  # Update the value of sums_tensor defined in While to the sums which defined outside of While through layers.assign
-                i = fluid.layers.increment(x=i, value=1, in_place=True)
+                i = paddle.static.nn.control_flow.increment(x=i, value=1, in_place=True)
                 data = fluid.layers.elementwise_add(x=data, y=one)
                 paddle.assign(paddle.less_than(x=i, y=loop_len), cond)
 
@@ -1607,47 +1533,6 @@ def max_sequence_len(rank_table):
     return res
 
 
-def increment(x, value=1.0, in_place=True):
-    """
-    The OP is usually used for control flow to increment the data of :attr:`x` by an amount :attr:`value`.
-    Notice that the number of elements in :attr:`x` must be equal to 1.
-
-    Parameters:
-        x (Variable): A tensor that must always contain only one element, its data type supports
-            float32, float64, int32 and int64.
-        value (float, optional): The amount to increment the data of :attr:`x`. Default: 1.0.
-        in_place (bool, optional): Whether the OP should be performed in-place. Default: True.
-
-    Returns:
-        Variable: The elementwise-incremented tensor with the same shape and data type as :attr:`x`.
-
-    Examples:
-        .. code-block:: python
-
-          import paddle.fluid as fluid
-          counter = fluid.layers.zeros(shape=[1], dtype='float32') # [0.]
-          fluid.layers.increment(counter) # [1.]
-    """
-    if in_dygraph_mode():
-        return _C_ops.increment_(x, value)
-
-    check_variable_and_dtype(
-        x, 'x', ['float32', 'float64', 'int32', 'int64'], 'increment'
-    )
-    helper = LayerHelper("increment", **locals())
-    if not in_place:
-        out = helper.create_variable_for_type_inference(dtype=x.dtype)
-    else:
-        out = x
-    helper.append_op(
-        type='increment',
-        inputs={'X': [x]},
-        outputs={'Out': [out]},
-        attrs={'step': float(value)},
-    )
-    return out
-
-
 def array_write(x, i, array=None):
     """
     This OP writes the input ``x`` into the i-th position of the ``array``
diff --git a/python/paddle/fluid/layers/rnn.py b/python/paddle/fluid/layers/rnn.py
index 8b5721438d2e5..a1ea375a9be45 100644
--- a/python/paddle/fluid/layers/rnn.py
+++ b/python/paddle/fluid/layers/rnn.py
@@ -21,7 +21,7 @@
 from paddle.utils import deprecated
 from . import nn
 from . import tensor
-from . import control_flow
+from paddle.static.nn import control_flow
 from . import utils
 from . import sequence_lod
 from .utils import *
diff --git a/python/paddle/fluid/optimizer.py b/python/paddle/fluid/optimizer.py
index ea10b49e9cc6f..7397db5bf8c23 100755
--- a/python/paddle/fluid/optimizer.py
+++ b/python/paddle/fluid/optimizer.py
@@ -7283,7 +7283,9 @@ def minimize(self, loss, startup_program=None):
                 dtype='int32',
                 persistable=True,
             )
-            layers.increment(x=step, value=1.0, in_place=True)
+            paddle.static.nn.control_flow.increment(
+                x=step, value=1.0, in_place=True
+            )
 
             # lookahead
             zero_var = layers.fill_constant(
@@ -7517,7 +7519,9 @@ def _get_gm_cond_var(self, main_block):
 
         with device_guard("cpu"):
             # step_var = (step_var + 1) % k_step
-            layers.increment(x=step_var, value=1.0, in_place=True)
+            paddle.static.nn.control_flow.increment(
+                x=step_var, value=1.0, in_place=True
+            )
             main_block.append_op(
                 type='elementwise_mod',
                 inputs={'X': step_var, 'Y': k_step_var},
diff --git a/python/paddle/fluid/tests/unittests/auto_parallel/test_while_op_partition.py b/python/paddle/fluid/tests/unittests/auto_parallel/test_while_op_partition.py
index fcfd783f71f6d..92204cc7dd39a 100644
--- a/python/paddle/fluid/tests/unittests/auto_parallel/test_while_op_partition.py
+++ b/python/paddle/fluid/tests/unittests/auto_parallel/test_while_op_partition.py
@@ -189,7 +189,9 @@ def get_program():
             cur_pred = mlp_while(pre_input)
 
             # 更新循环条件
-            i = fluid.layers.increment(x=i, value=1, in_place=True)
+            i = paddle.static.nn.control_flow.increment(
+                x=i, value=1, in_place=True
+            )
             fluid.layers.array_write(cur_pred, array=input_array, i=i)
             paddle.assign(paddle.less_than(x=i, y=loop_len), cond)
 
diff --git a/python/paddle/fluid/tests/unittests/collective/fleet/hybrid_parallel_inference_helper.py b/python/paddle/fluid/tests/unittests/collective/fleet/hybrid_parallel_inference_helper.py
index 542b1ba637936..c903359f68239 100644
--- a/python/paddle/fluid/tests/unittests/collective/fleet/hybrid_parallel_inference_helper.py
+++ b/python/paddle/fluid/tests/unittests/collective/fleet/hybrid_parallel_inference_helper.py
@@ -89,7 +89,9 @@ def test_hybrid_parallel_inference_helper_mp1pp2(self):
             with while_op.block():
                 with paddle.fluid.device_guard(f'{device}:all'):
                     input = layers.array_read(array=data, i=step_idx)
-                    layers.increment(x=step_idx, value=1.0, in_place=True)
+                    paddle.static.nn.control_flow.increment(
+                        x=step_idx, value=1.0, in_place=True
+                    )
                     layers.array_write(input, i=step_idx, array=data)
 
                 with paddle.fluid.device_guard(f'{device}:0'):
diff --git a/python/paddle/fluid/tests/unittests/dist_transformer.py b/python/paddle/fluid/tests/unittests/dist_transformer.py
index c6165dd753537..fad6c62cc3313 100644
--- a/python/paddle/fluid/tests/unittests/dist_transformer.py
+++ b/python/paddle/fluid/tests/unittests/dist_transformer.py
@@ -28,6 +28,7 @@
 import paddle.fluid as fluid
 import paddle.fluid.layers as layers
 import paddle.nn.functional as F
+import paddle.static.nn.control_flow as control_flow
 
 const_para_attr = fluid.ParamAttr(initializer=fluid.initializer.Constant(0.001))
 const_bias_attr = const_para_attr
@@ -1813,7 +1814,7 @@ def beam_search():
                     shape=[-1, 1, 1],
                     dtype=pre_ids.dtype,
                 ),
-                y=layers.increment(x=step_idx, value=1.0, in_place=False),
+                y=control_flow.increment(x=step_idx, value=1.0, in_place=False),
                 axis=0,
             )
             logits = wrap_decoder(
@@ -1852,7 +1853,7 @@ def beam_search():
                 end_id=eos_idx,
             )
 
-            layers.increment(x=step_idx, value=1.0, in_place=True)
+            control_flow.increment(x=step_idx, value=1.0, in_place=True)
             # update states
             layers.array_write(selected_ids, i=step_idx, array=ids)
             layers.array_write(selected_scores, i=step_idx, array=scores)
diff --git a/python/paddle/fluid/tests/unittests/dygraph_to_static/test_loop.py b/python/paddle/fluid/tests/unittests/dygraph_to_static/test_loop.py
index 06b17978d6687..9da4598425d07 100644
--- a/python/paddle/fluid/tests/unittests/dygraph_to_static/test_loop.py
+++ b/python/paddle/fluid/tests/unittests/dygraph_to_static/test_loop.py
@@ -19,6 +19,7 @@
 
 import paddle
 import paddle.fluid as fluid
+import paddle.static.nn.control_flow as control_flow
 from paddle.jit.api import declarative
 from paddle.jit.dy2static.loop_transformer import NameVisitor
 from paddle.utils import gast
@@ -83,7 +84,7 @@ def while_loop_dyfunc_with_none(x):
 def for_loop_dyfunc(max_len):
     for i in range(max_len):
         ret = fluid.layers.zeros(shape=[1], dtype='float32')
-        fluid.layers.increment(ret, value=2.0, in_place=True)
+        control_flow.increment(ret, value=2.0, in_place=True)
     return ret
 
 
@@ -104,14 +105,14 @@ def for_loop_dyfunc2(max_len):
 def for_loop_dyfunc3(max_len):
     ret = fluid.layers.zeros(shape=[1], dtype='float32')
     for i in range(1, 10, 2):
-        fluid.layers.increment(ret, value=2.0, in_place=True)
+        control_flow.increment(ret, value=2.0, in_place=True)
     return ret
 
 
 def for_loop_dyfunc4(max_len):
     ret = fluid.layers.zeros(shape=[1], dtype='float32')
     for i in range(10, 1, -2):
-        fluid.layers.increment(ret, value=2.0, in_place=True)
+        control_flow.increment(ret, value=2.0, in_place=True)
     return ret
 
 
@@ -119,7 +120,7 @@ def for_loop_dyfunc_not_support(max_len):
     ret = fluid.layers.zeros(shape=[1], dtype='float32')
     a = -2
     for i in range(10, 1, a):
-        fluid.layers.increment(ret, value=2.0, in_place=True)
+        control_flow.increment(ret, value=2.0, in_place=True)
     return ret
 
 
diff --git a/python/paddle/fluid/tests/unittests/npu/test_increment_op_npu.py b/python/paddle/fluid/tests/unittests/npu/test_increment_op_npu.py
index 39f11737bce6c..568f4d7a5a768 100644
--- a/python/paddle/fluid/tests/unittests/npu/test_increment_op_npu.py
+++ b/python/paddle/fluid/tests/unittests/npu/test_increment_op_npu.py
@@ -121,7 +121,7 @@ def test_npu(self):
 
         with paddle.static.program_guard(main_prog, startup_prog):
             a = paddle.static.data(name="a", shape=[1], dtype='float32')
-            b = fluid.layers.increment(a)
+            b = paddle.static.nn.control_flow.increment(a)
 
         place = paddle.NPUPlace(NPUPlace)
 
diff --git a/python/paddle/fluid/tests/unittests/npu/test_while_op_npu.py b/python/paddle/fluid/tests/unittests/npu/test_while_op_npu.py
index c63f11b85910c..043903b6e5d6e 100644
--- a/python/paddle/fluid/tests/unittests/npu/test_while_op_npu.py
+++ b/python/paddle/fluid/tests/unittests/npu/test_while_op_npu.py
@@ -21,6 +21,7 @@
 from paddle.fluid.backward import append_backward
 import numpy
 from paddle.fluid import compiler, Program, program_guard
+import paddle.static.nn.control_flow as control_flow
 
 paddle.enable_static()
 
@@ -43,9 +44,9 @@ def simple_net(self):
         init = layers.zeros(shape=[10], dtype='float32')
         mem_array = layers.array_write(x=init, i=i)
         data_array = layers.array_write(x=d0, i=i)
-        i = layers.increment(i)
+        i = control_flow.increment(i)
         layers.array_write(d1, i, array=data_array)
-        i = layers.increment(i)
+        i = control_flow.increment(i)
         layers.array_write(d2, i, array=data_array)
         i = layers.zeros(shape=[1], dtype='int32')
         i = layers.cast(i, 'int64')
@@ -71,7 +72,7 @@ def simple_net(self):
             prev = layers.array_read(array=mem_array, i=i)
             result = layers.sums(input=[d, prev])
 
-            i = layers.increment(x=i, in_place=True)
+            i = control_flow.increment(x=i, in_place=True)
             layers.array_write(result, i=i, array=mem_array)
             paddle.assign(paddle.less_than(x=i, y=array_len), cond)
 
@@ -80,7 +81,7 @@ def simple_net(self):
                 prev2 = layers.array_read(array=mem_array, i=j)
                 result2 = layers.sums(input=[d2, prev2])
 
-                j = layers.increment(x=j, in_place=True)
+                j = control_flow.increment(x=j, in_place=True)
                 layers.array_write(result2, i=j, array=mem_array)
                 paddle.assign(paddle.less_than(x=j, y=array_len2), cond2)
         sum_result = layers.array_read(array=mem_array, i=j)
diff --git a/python/paddle/fluid/tests/unittests/test_array_read_write_op.py b/python/paddle/fluid/tests/unittests/test_array_read_write_op.py
index 907bb65cfce25..3cb9e7b33dc1e 100644
--- a/python/paddle/fluid/tests/unittests/test_array_read_write_op.py
+++ b/python/paddle/fluid/tests/unittests/test_array_read_write_op.py
@@ -20,6 +20,7 @@
 import paddle.fluid as fluid
 import paddle.fluid.core as core
 import paddle.fluid.layers as layers
+import paddle.static.nn.control_flow as control_flow
 from paddle.fluid import Program, program_guard
 from paddle.fluid.backward import append_backward
 from paddle.fluid.executor import Executor
@@ -30,17 +31,17 @@ def _test_read_write(x):
     i = layers.zeros(shape=[1], dtype='int64')
     i.stop_gradient = False
     arr = layers.array_write(x=x[0], i=i)
-    i = layers.increment(x=i)
+    i = control_flow.increment(x=i)
     arr = layers.array_write(x=x[1], i=i, array=arr)
-    i = layers.increment(x=i)
+    i = control_flow.increment(x=i)
     arr = layers.array_write(x=x[2], i=i, array=arr)
 
     i = layers.zeros(shape=[1], dtype='int64')
     i.stop_gradient = False
     a0 = layers.array_read(array=arr, i=i)
-    i = layers.increment(x=i)
+    i = control_flow.increment(x=i)
     a1 = layers.array_read(array=arr, i=i)
-    i = layers.increment(x=i)
+    i = control_flow.increment(x=i)
     a2 = layers.array_read(array=arr, i=i)
 
     mean_a0 = paddle.mean(a0)
diff --git a/python/paddle/fluid/tests/unittests/test_assert_op.py b/python/paddle/fluid/tests/unittests/test_assert_op.py
index d59194aef56cb..f62ad38d459d1 100644
--- a/python/paddle/fluid/tests/unittests/test_assert_op.py
+++ b/python/paddle/fluid/tests/unittests/test_assert_op.py
@@ -17,6 +17,7 @@
 import paddle
 import paddle.fluid as fluid
 import paddle.fluid.layers as layers
+from paddle.static.nn.control_flow import Assert
 
 
 class TestAssertOp(unittest.TestCase):
@@ -33,7 +34,7 @@ def net_func():
             condition = layers.fill_constant(
                 shape=[1], dtype='bool', value=True
             )
-            layers.Assert(condition, [])
+            Assert(condition, [])
 
         self.run_network(net_func)
 
@@ -42,7 +43,7 @@ def net_func():
             condition = layers.fill_constant(
                 shape=[1], dtype='bool', value=False
             )
-            layers.Assert(condition)
+            Assert(condition)
 
         with self.assertRaises(ValueError):
             self.run_network(net_func)
@@ -52,7 +53,7 @@ def net_func():
             condition = layers.fill_constant(
                 shape=[1, 2], dtype='bool', value=True
             )
-            layers.Assert(condition, [])
+            Assert(condition, [])
 
         with self.assertRaises(ValueError):
             self.run_network(net_func)
@@ -62,7 +63,7 @@ def net_func():
             zero = layers.fill_constant(shape=[1], dtype='int64', value=0)
             one = layers.fill_constant(shape=[1], dtype='int64', value=1)
             condition = paddle.less_than(one, zero)  # False
-            layers.Assert(condition, [zero, one])
+            Assert(condition, [zero, one])
 
         print("test_assert_print_data")
         with self.assertRaises(ValueError):
@@ -72,7 +73,7 @@ def test_assert_summary(self):
         def net_func():
             x = layers.fill_constant(shape=[10], dtype='float32', value=2.0)
             condition = paddle.max(x) < 1.0
-            layers.Assert(condition, (x,), 5)
+            Assert(condition, (x,), 5)
 
         print("test_assert_summary")
         with self.assertRaises(ValueError):
@@ -82,7 +83,7 @@ def test_assert_summary_greater_than_size(self):
         def net_func():
             x = layers.fill_constant(shape=[2, 3], dtype='float32', value=2.0)
             condition = paddle.max(x) < 1.0
-            layers.Assert(condition, [x], 10, name="test")
+            Assert(condition, [x], 10, name="test")
 
         print("test_assert_summary_greater_than_size")
         with self.assertRaises(ValueError):
diff --git a/python/paddle/fluid/tests/unittests/test_async_ssa_graph_executor_mnist.py b/python/paddle/fluid/tests/unittests/test_async_ssa_graph_executor_mnist.py
index 26eb0a628ab9a..27fb3e36bb007 100644
--- a/python/paddle/fluid/tests/unittests/test_async_ssa_graph_executor_mnist.py
+++ b/python/paddle/fluid/tests/unittests/test_async_ssa_graph_executor_mnist.py
@@ -62,7 +62,7 @@ def convolutional_neural_network(use_py_reader):
         acc = paddle.static.accuracy(input=prediction, label=label)
         i = fluid.layers.zeros(shape=[1], dtype='int64')
         array = fluid.layers.array_write(x=prediction, i=i)
-        fluid.layers.increment(i)
+        paddle.static.nn.control_flow.increment(i)
         fluid.layers.array_write(x=acc, i=i, array=array)
         return array, img, label, prediction, avg_loss, acc, py_reader
 
diff --git a/python/paddle/fluid/tests/unittests/test_dynamic_rnn_stop_gradient.py b/python/paddle/fluid/tests/unittests/test_dynamic_rnn_stop_gradient.py
index 3e3eefd5d278d..f839563344a86 100644
--- a/python/paddle/fluid/tests/unittests/test_dynamic_rnn_stop_gradient.py
+++ b/python/paddle/fluid/tests/unittests/test_dynamic_rnn_stop_gradient.py
@@ -51,7 +51,9 @@ def build_and_run_program(place, batch_size, beam_size, stop_gradient=False):
         topk_coordinates = paddle.stack([batch_pos, indices], axis=2)
         topk_coordinates.stop_gradient = stop_gradient
         score = paddle.gather_nd(x, topk_coordinates)
-        layers.increment(x=step_idx, value=1.0, in_place=True)
+        paddle.static.nn.control_flow.increment(
+            x=step_idx, value=1.0, in_place=True
+        )
         layers.array_write(score, i=step_idx, array=scores)
         length_cond = paddle.less_than(x=step_idx, y=max_len)
         layers.assign(length_cond, cond)
diff --git a/python/paddle/fluid/tests/unittests/test_eager_deletion_while_op.py b/python/paddle/fluid/tests/unittests/test_eager_deletion_while_op.py
index 943642b857cce..bebb534319cb3 100644
--- a/python/paddle/fluid/tests/unittests/test_eager_deletion_while_op.py
+++ b/python/paddle/fluid/tests/unittests/test_eager_deletion_while_op.py
@@ -26,6 +26,7 @@
 import paddle.fluid.compiler as compiler
 import paddle.fluid.core as core
 import paddle.fluid.layers as layers
+import paddle.static.nn.control_flow as control_flow
 from paddle.fluid.executor import Executor
 
 paddle.enable_static()
@@ -83,10 +84,10 @@ def run_main(self, place, with_data_parallel):
         mem_array = layers.array_write(x=init, i=i)
         data_array = layers.array_write(x=d0, i=i)
 
-        i = layers.increment(i)
+        i = control_flow.increment(i)
         layers.array_write(d1, i, array=data_array)
 
-        i = layers.increment(i)
+        i = control_flow.increment(i)
         layers.array_write(d2, i, array=data_array)
 
         i = layers.zeros(shape=[1], dtype='int64')
@@ -112,7 +113,7 @@ def run_main(self, place, with_data_parallel):
             prev = paddle.reshape(prev, shape=[10])
             result = layers.sums(input=[d, prev])
 
-            i = layers.increment(x=i, in_place=True)
+            i = control_flow.increment(x=i, in_place=True)
             layers.array_write(result, i=i, array=mem_array)
             paddle.assign(paddle.less_than(x=i, y=array_len), cond)
             with while_op2.block():
@@ -122,7 +123,7 @@ def run_main(self, place, with_data_parallel):
                 prev2 = paddle.reshape(prev2, shape=[10])
                 result2 = layers.sums(input=[d2, prev2])
 
-                j = layers.increment(x=j, in_place=True)
+                j = control_flow.increment(x=j, in_place=True)
                 layers.array_write(result2, i=j, array=mem_array)
                 paddle.assign(paddle.less_than(x=j, y=array_len2), cond2)
 
diff --git a/python/paddle/fluid/tests/unittests/test_executor_and_mul.py b/python/paddle/fluid/tests/unittests/test_executor_and_mul.py
index 639f84295b24b..912d5a8490bfe 100644
--- a/python/paddle/fluid/tests/unittests/test_executor_and_mul.py
+++ b/python/paddle/fluid/tests/unittests/test_executor_and_mul.py
@@ -16,8 +16,9 @@
 
 import numpy as np
 
+import paddle.static.nn.control_flow as control_flow
 from paddle.fluid.executor import Executor
-from paddle.fluid.layers import array_write, data, increment, mul, zeros
+from paddle.fluid.layers import array_write, data, mul, zeros
 
 
 class TestExecutor(unittest.TestCase):
@@ -26,13 +27,13 @@ def test_mul(self):
         a = data(name='a', shape=[784], dtype='float32')
         array = array_write(x=a, i=i)
 
-        i = increment(i)
+        i = control_flow.increment(i)
         b = data(
             name='b', shape=[784, 100], dtype='float32', append_batch_size=False
         )
         array_write(x=b, i=i, array=array)
 
-        i = increment(i)
+        i = control_flow.increment(i)
         out = mul(x=a, y=b)
         array_write(x=out, i=i, array=array)
 
diff --git a/python/paddle/fluid/tests/unittests/test_fetch_lod_tensor_array.py b/python/paddle/fluid/tests/unittests/test_fetch_lod_tensor_array.py
index b2869d8888237..9a127edce18f1 100644
--- a/python/paddle/fluid/tests/unittests/test_fetch_lod_tensor_array.py
+++ b/python/paddle/fluid/tests/unittests/test_fetch_lod_tensor_array.py
@@ -20,6 +20,7 @@
 
 import paddle.fluid as fluid
 import paddle.fluid.layers as layers
+import paddle.static.nn.control_flow as control_flow
 
 
 class TestFetchLoDTensorArray(unittest.TestCase):
@@ -35,9 +36,9 @@ def build_program(self, main_program, startup_program):
                 opt.minimize(loss)
 
                 array = layers.array_write(x=img, i=i)
-                i = layers.increment(i)
+                i = control_flow.increment(i)
                 layers.array_write(x=label, i=i, array=array)
-                i = layers.increment(i)
+                i = control_flow.increment(i)
                 layers.array_write(x=loss, i=i, array=array)
 
                 return loss, array
diff --git a/python/paddle/fluid/tests/unittests/test_profiler.py b/python/paddle/fluid/tests/unittests/test_profiler.py
index 62d46d4cadc48..0f7459dd96893 100644
--- a/python/paddle/fluid/tests/unittests/test_profiler.py
+++ b/python/paddle/fluid/tests/unittests/test_profiler.py
@@ -50,7 +50,9 @@ def build_program(self, compile_program=True):
             with while_op.block():
                 hidden_n = fluid.layers.fc(input=hidden1, size=64, act='relu')
                 layers.array_write(hidden_n, i, data_arr)
-                fluid.layers.increment(x=counter, value=1, in_place=True)
+                paddle.static.nn.control_flow.increment(
+                    x=counter, value=1, in_place=True
+                )
                 paddle.assign(paddle.less_than(x=counter, y=until), cond)
 
             hidden_n = layers.array_read(data_arr, i)
diff --git a/python/paddle/fluid/tests/unittests/test_while_loop_op.py b/python/paddle/fluid/tests/unittests/test_while_loop_op.py
index 8c94834c9a28b..19193fb348399 100644
--- a/python/paddle/fluid/tests/unittests/test_while_loop_op.py
+++ b/python/paddle/fluid/tests/unittests/test_while_loop_op.py
@@ -20,6 +20,7 @@
 import paddle.fluid as fluid
 import paddle.fluid.core as core
 import paddle.fluid.layers as layers
+import paddle.static.nn.control_flow as control_flow
 from paddle.fluid.backward import append_backward
 from paddle.fluid.framework import Program, program_guard
 
@@ -59,7 +60,7 @@ def cond(i, mem):
 
         def body(i, mem):
             mem = paddle.add(x=mem, y=one)
-            i = layers.increment(i)
+            i = control_flow.increment(i)
             return [i, mem]
 
         main_program = Program()
@@ -100,7 +101,7 @@ def body(i, ten, test_dict, test_list, test_list_dict):
                 test_list_dict[0]["test_key"]
             )
 
-            i = layers.increment(i)
+            i = control_flow.increment(i)
             return [i, ten, test_dict, test_list, test_list_dict]
 
         main_program = Program()
@@ -168,7 +169,7 @@ def internal_cond(j, init, sums):
             def internal_body(j, init, sums):
                 init = paddle.add(x=init, y=ones)
                 sums = paddle.add(x=init, y=sums)
-                j = layers.increment(j)
+                j = control_flow.increment(j)
                 return [j, init, sums]
 
             result = layers.while_loop(
@@ -178,7 +179,7 @@ def internal_body(j, init, sums):
             init = result[1]
             sums = result[2]
             sums = paddle.add(x=init, y=sums)
-            i = layers.increment(i)
+            i = control_flow.increment(i)
             return [i, j, init, sums]
 
         main_program = Program()
@@ -223,7 +224,7 @@ def cond(i, x):
 
         def body(i, x):
             x = paddle.multiply(x=i, y=i)
-            i = layers.increment(i)
+            i = control_flow.increment(i)
             return [i, x]
 
         main_program = Program()
@@ -318,7 +319,7 @@ def internal_body(j, x, mem_array):
                 inner_prev = layers.array_read(array=mem_array, i=j)
                 inner_sum_0 = paddle.add(x=inner_data, y=inner_prev)
                 inner_sum_1 = paddle.add(x=x, y=inner_sum_0)
-                j = layers.increment(x=j, in_place=True)
+                j = control_flow.increment(x=j, in_place=True)
                 layers.array_write(inner_sum_1, i=j, array=mem_array)
                 return [j, x, mem_array]
 
@@ -326,7 +327,7 @@ def internal_body(j, x, mem_array):
             outer_prev = layers.array_read(array=mem_array, i=i)
             outer_sum_0 = paddle.add(x=outer_data, y=outer_prev)
             outer_sum_1 = paddle.add(x=x, y=outer_sum_0)
-            i = layers.increment(x=i, in_place=True)
+            i = control_flow.increment(x=i, in_place=True)
             layers.array_write(outer_sum_1, i=i, array=mem_array)
             j, x, mem_array = layers.while_loop(
                 internal_cond, internal_body, [j, x, mem_array]
@@ -346,9 +347,9 @@ def internal_body(j, x, mem_array):
             init = layers.zeros(shape=[10], dtype='float32')
             mem_array = layers.array_write(x=init, i=i)
             data_array = layers.array_write(x=d0, i=i)
-            i = layers.increment(i)
+            i = control_flow.increment(i)
             layers.array_write(d1, i, array=data_array)
-            i = layers.increment(i)
+            i = control_flow.increment(i)
             layers.array_write(d2, i, array=data_array)
             i = layers.zeros(shape=[1], dtype='int64')
             i.stop_gradient = True
@@ -438,7 +439,7 @@ def cond_returns_constant(i):
             return 1
 
         def cond_returns_not_bool_tensor(i):
-            return layers.increment(i)
+            return control_flow.increment(i)
 
         def cond_returns_bool_tensor(i):
             return paddle.less_than(i, ten)
@@ -450,14 +451,14 @@ def cond_receives_two_args(i, ten):
             return paddle.less_than(i, ten)
 
         def body(i):
-            return layers.increment(i)
+            return control_flow.increment(i)
 
         def body_returns_error_length(i):
-            i = layers.increment(i)
+            i = control_flow.increment(i)
             return [i, i]
 
         def body_returns_error_type(i, ten):
-            return layers.increment(i)
+            return control_flow.increment(i)
 
         def cond_returns_with_mutable_dict(i, test_dict):
             return i > 0
@@ -466,7 +467,7 @@ def body_returns_with_mutable_dict(i, test_dict):
             test_dict['new_key'] = layers.fill_constant(
                 shape=[1], dtype='int64', value=1
             )
-            return layers.increment(i), test_dict
+            return control_flow.increment(i), test_dict
 
         def cond_returns_with_mutable_list(i, test_list):
             return i > 0
@@ -475,7 +476,7 @@ def body_returns_with_mutable_list(i, test_list):
             test_list.append(
                 layers.fill_constant(shape=[1], dtype='int64', value=1)
             )
-            return layers.increment(i), test_list
+            return control_flow.increment(i), test_list
 
         main_program = Program()
         startup_program = Program()
diff --git a/python/paddle/fluid/tests/unittests/test_while_op.py b/python/paddle/fluid/tests/unittests/test_while_op.py
index f77d9767f3c8b..b277cb334480b 100644
--- a/python/paddle/fluid/tests/unittests/test_while_op.py
+++ b/python/paddle/fluid/tests/unittests/test_while_op.py
@@ -20,6 +20,7 @@
 import paddle.fluid as fluid
 import paddle.fluid.core as core
 import paddle.fluid.layers as layers
+import paddle.static.nn.control_flow as control_flow
 from paddle.fluid.backward import append_backward
 from paddle.fluid.executor import Executor
 
@@ -42,9 +43,9 @@ def simple_net(self):
         init = layers.zeros(shape=[10], dtype='float32')
         mem_array = layers.array_write(x=init, i=i)
         data_array = layers.array_write(x=d0, i=i)
-        i = layers.increment(i)
+        i = control_flow.increment(i)
         layers.array_write(d1, i, array=data_array)
-        i = layers.increment(i)
+        i = control_flow.increment(i)
         layers.array_write(d2, i, array=data_array)
         i = layers.zeros(shape=[1], dtype='int64')
         i.stop_gradient = True
@@ -63,7 +64,7 @@ def simple_net(self):
             prev = layers.array_read(array=mem_array, i=i)
             result = layers.sums(input=[d, prev])
 
-            i = layers.increment(x=i, in_place=True)
+            i = control_flow.increment(x=i, in_place=True)
             layers.array_write(result, i=i, array=mem_array)
             paddle.assign(paddle.less_than(x=i, y=array_len), cond)
 
@@ -72,7 +73,7 @@ def simple_net(self):
                 prev2 = layers.array_read(array=mem_array, i=j)
                 result2 = layers.sums(input=[d2, prev2])
 
-                j = layers.increment(x=j, in_place=True)
+                j = control_flow.increment(x=j, in_place=True)
                 layers.array_write(result2, i=j, array=mem_array)
                 paddle.assign(paddle.less_than(x=j, y=array_len2), cond2)
         sum_result = layers.array_read(array=mem_array, i=j)
@@ -134,7 +135,7 @@ def test_error(self):
 
             def test_bad_x():
                 x = [1, 2, 3]
-                fluid.layers.increment(x)
+                control_flow.increment(x)
 
             self.assertRaises(TypeError, test_bad_x)
 
diff --git a/python/paddle/fluid/tests/unittests/xpu/test_while_op_xpu.py b/python/paddle/fluid/tests/unittests/xpu/test_while_op_xpu.py
index e52e8fdceb7ed..33056f8c9b832 100644
--- a/python/paddle/fluid/tests/unittests/xpu/test_while_op_xpu.py
+++ b/python/paddle/fluid/tests/unittests/xpu/test_while_op_xpu.py
@@ -19,6 +19,7 @@
 import paddle
 import paddle.fluid as fluid
 import paddle.fluid.layers as layers
+import paddle.static.nn.control_flow as control_flow
 from paddle.fluid.backward import append_backward
 from paddle.fluid.executor import Executor
 
@@ -41,9 +42,9 @@ def simple_net(self):
         init = layers.zeros(shape=[10], dtype='float32')
         mem_array = layers.array_write(x=init, i=i)
         data_array = layers.array_write(x=d0, i=i)
-        i = layers.increment(i)
+        i = control_flow.increment(i)
         layers.array_write(d1, i, array=data_array)
-        i = layers.increment(i)
+        i = control_flow.increment(i)
         layers.array_write(d2, i, array=data_array)
         i = layers.zeros(shape=[1], dtype='int64')
         i.stop_gradient = True
@@ -62,7 +63,7 @@ def simple_net(self):
             prev = layers.array_read(array=mem_array, i=i)
             result = layers.sums(input=[d, prev])
 
-            i = layers.increment(x=i, in_place=True)
+            i = control_flow.increment(x=i, in_place=True)
             layers.array_write(result, i=i, array=mem_array)
             paddle.assign(paddle.less_than(x=i, y=array_len), cond)
 
@@ -71,7 +72,7 @@ def simple_net(self):
                 prev2 = layers.array_read(array=mem_array, i=j)
                 result2 = layers.sums(input=[d2, prev2])
 
-                j = layers.increment(x=j, in_place=True)
+                j = control_flow.increment(x=j, in_place=True)
                 layers.array_write(result2, i=j, array=mem_array)
                 paddle.assign(paddle.less_than(x=j, y=array_len2), cond2)
         sum_result = layers.array_read(array=mem_array, i=j)
diff --git a/python/paddle/jit/dy2static/convert_operators.py b/python/paddle/jit/dy2static/convert_operators.py
index 1d3e23a4b96b7..886aefa94e30a 100644
--- a/python/paddle/jit/dy2static/convert_operators.py
+++ b/python/paddle/jit/dy2static/convert_operators.py
@@ -14,12 +14,13 @@
 
 import re
 import paddle
+from paddle.static.nn.control_flow import Assert, increment
 from paddle.fluid.data_feeder import convert_dtype
 from paddle.jit.dy2static.variable_trans_func import (
     to_static_variable,
 )
 from paddle.fluid.framework import core, Variable
-from paddle.fluid.layers import Assert, Print
+from paddle.fluid.layers import Print
 from paddle.fluid.layers import (
     array_read,
     array_write,
@@ -35,7 +36,6 @@
 from paddle.fluid.layers.control_flow import (
     cond,
     while_loop,
-    increment,
 )
 from .return_transformer import (
     RETURN_NO_VALUE_VAR_NAME,

From cf743858d6c026bc5bc8667f47708348d6fab5d6 Mon Sep 17 00:00:00 2001
From: wuzhanfei <wuzhanfei@baidu.com>
Date: Tue, 6 Dec 2022 03:32:54 +0000
Subject: [PATCH 02/16] add control_flow.py

---
 python/paddle/static/nn/control_flow.py | 143 ++++++++++++++++++++++++
 1 file changed, 143 insertions(+)
 create mode 100644 python/paddle/static/nn/control_flow.py

diff --git a/python/paddle/static/nn/control_flow.py b/python/paddle/static/nn/control_flow.py
new file mode 100644
index 0000000000000..c2f69c69ec882
--- /dev/null
+++ b/python/paddle/static/nn/control_flow.py
@@ -0,0 +1,143 @@
+#   Copyright (c) 2022 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+
+from paddle import _C_ops
+from paddle.fluid.data_feeder import check_type, check_variable_and_dtype
+from paddle.fluid.layer_helper import LayerHelper
+from paddle.framework import in_dygraph_mode
+
+__all__ = [
+    'Assert',
+    'increment',
+]
+
+
+def Assert(cond, data=None, summarize=20, name=None):
+    '''
+    This API creates an op that asserts the given condition is true. If the
+    condition is false, prints the tensors in data. ``summarize`` specifies the
+    number of the elements in the tensors to print.
+
+    Args:
+        cond (Variable): The boolean condition tensor whose numel should be 1.
+        data (list|tuple, optional): list or tuple of tensors to print when
+            condition is not true. If it's ``None``, no tensor will be printed.
+            The default value is ``None``.
+        summarize (int, optional): Number of elements in the tensor to be
+            printed. If its value is -1, then all elements in the tensor will
+            be printed. The default value is 20.
+        name (str, optional): The default value is ``None`` . Normally users
+            don't have to set this parameter. For more information, please
+            refer to :ref:`api_guide_Name` .
+
+    Returns:
+        Operator: the created operation.
+
+    Raises:
+        TypeError: If ``cond`` is not boolean Variable.
+        TypeError: If ``data`` is not a list or tuple or ``None``.
+        TypeError: If ``summarize`` is not int.
+        TypeError: If ``name`` is not a string or ``None`` .
+        framework.core.EnforceNotMet: If the condition is False in running time.
+
+    Examples:
+        .. code-block:: python
+
+            import paddle
+            from paddle.static.nn.control_flow import Assert
+
+            x = paddle.full(shape=[2, 3], dtype='float32', value=2.0)
+            condition = paddle.max(x) < 1.0 # False
+            Assert(condition, [x], 10, "example_assert_layer")
+
+            exe = paddle.static.Executor()
+            try:
+                exe.run(paddle.static.default_main_program())
+                # Print x and throws paddle.framework.core.EnforceNotMet exception
+                # Example printed message for x:
+                #
+                # Variable: fill_constant_0.tmp_0
+                #   - lod: {}
+                #   - place: CPUPlace()
+                #   - shape: [2, 3]
+                #   - layout: NCHW
+                #   - dtype: float
+                #   - data: [2 2 2 2 2 2]
+            except paddle.framework.core.EnforceNotMet as e:
+                print("Assert Exception Example")
+
+    '''
+    check_variable_and_dtype(
+        cond, "cond", ["bool"], "static.nn.control_flow.Assert"
+    )
+    check_type(
+        data, "data", (list, tuple, type(None)), "static.nn.control_flow.Assert"
+    )
+    check_type(summarize, "summarize", int, "static.nn.control_flow.Assert")
+    check_type(name, "name", (str, type(None)), "static.nn.control_flow.Assert")
+
+    layer_name = name if name else ('assert_' + cond.name)
+    helper = LayerHelper(layer_name, **locals())
+
+    op = helper.append_op(
+        type="assert",
+        inputs={"Cond": cond, "Data": [] if data is None else list(data)},
+        attrs={"summarize": summarize},
+    )
+
+    return op
+
+
+def increment(x, value=1.0, in_place=True):
+    """
+    The OP is usually used for control flow to increment the data of :attr:`x` by an amount :attr:`value`.
+    Notice that the number of elements in :attr:`x` must be equal to 1.
+
+    Parameters:
+        x (Variable): A tensor that must always contain only one element, its data type supports
+            float32, float64, int32 and int64.
+        value (float, optional): The amount to increment the data of :attr:`x`. Default: 1.0.
+        in_place (bool, optional): Whether the OP should be performed in-place. Default: True.
+
+    Returns:
+        Variable: The elementwise-incremented tensor with the same shape and data type as :attr:`x`.
+
+    Examples:
+        .. code-block:: python
+
+          import paddle
+          from paddle.static.nn.control_flow import increment
+
+          counter = paddle.zeros(shape=[1], dtype='float32') # [0.]
+          increment(counter) # [1.]
+    """
+    if in_dygraph_mode():
+        return _C_ops.increment_(x, value)
+
+    check_variable_and_dtype(
+        x, 'x', ['float32', 'float64', 'int32', 'int64'], 'increment'
+    )
+    helper = LayerHelper("increment", **locals())
+    if not in_place:
+        out = helper.create_variable_for_type_inference(dtype=x.dtype)
+    else:
+        out = x
+    helper.append_op(
+        type='increment',
+        inputs={'X': [x]},
+        outputs={'Out': [out]},
+        attrs={'step': float(value)},
+    )
+    return out

From 3e917420c6bd995c56925532ca113d8de2246f92 Mon Sep 17 00:00:00 2001
From: wuzhanfei <wuzhanfei@baidu.com>
Date: Tue, 6 Dec 2022 07:47:11 +0000
Subject: [PATCH 03/16] fix circular dependency

---
 python/paddle/fluid/layers/rnn.py                | 9 ++++++---
 python/paddle/jit/dy2static/convert_operators.py | 5 ++++-
 python/paddle/static/nn/control_flow.py          | 2 +-
 3 files changed, 11 insertions(+), 5 deletions(-)

diff --git a/python/paddle/fluid/layers/rnn.py b/python/paddle/fluid/layers/rnn.py
index a1ea375a9be45..9df2820f02556 100644
--- a/python/paddle/fluid/layers/rnn.py
+++ b/python/paddle/fluid/layers/rnn.py
@@ -21,7 +21,7 @@
 from paddle.utils import deprecated
 from . import nn
 from . import tensor
-from paddle.static.nn import control_flow
+from . import control_flow
 from . import utils
 from . import sequence_lod
 from .utils import *
@@ -1535,8 +1535,9 @@ def _maybe_copy(state, new_state, step_mask):
             next_finished,
             next_sequence_lengths,
         )
+        from paddle.static.nn.control_flow import increment
 
-        control_flow.increment(x=step_idx_tensor, value=1.0, in_place=True)
+        increment(x=step_idx_tensor, value=1.0, in_place=True)
         step_idx += 1
 
         cond = paddle.logical_not(paddle.all(finished))
@@ -1695,7 +1696,9 @@ def _create_array_out_of_while(dtype):
             outputs,
             outputs_arrays,
         )
-        control_flow.increment(x=step_idx, value=1.0, in_place=True)
+        from paddle.static.nn.control_flow import increment
+
+        increment(x=step_idx, value=1.0, in_place=True)
         # update the global_finished first, since it might be also in states of
         # decoder, which otherwise would write a stale finished status to array
         tensor.assign(next_finished, global_finished)
diff --git a/python/paddle/jit/dy2static/convert_operators.py b/python/paddle/jit/dy2static/convert_operators.py
index 886aefa94e30a..a5d2ca14cd9b6 100644
--- a/python/paddle/jit/dy2static/convert_operators.py
+++ b/python/paddle/jit/dy2static/convert_operators.py
@@ -14,7 +14,6 @@
 
 import re
 import paddle
-from paddle.static.nn.control_flow import Assert, increment
 from paddle.fluid.data_feeder import convert_dtype
 from paddle.jit.dy2static.variable_trans_func import (
     to_static_variable,
@@ -731,6 +730,8 @@ def convert_assert(cond, message=""):
     if isinstance(cond, Variable):
         cond = cast(cond, "bool")
         # NOTE: message is not used because Paddle Assert has no corresponding parameter to use.
+        from paddle.static.nn.control_flow import Assert
+
         return Assert(cond)
     else:
         assert cond, message
@@ -783,6 +784,8 @@ def cond(i, new_array):
     def body(i, new_array):
         item = array_read(array=array, i=i)
         array_write(item, paddle.tensor.array_length(new_array), new_array)
+        from paddle.static.nn.control_flow import increment
+
         i = increment(i)
         return i, new_array
 
diff --git a/python/paddle/static/nn/control_flow.py b/python/paddle/static/nn/control_flow.py
index c2f69c69ec882..1eecde1cb6c86 100644
--- a/python/paddle/static/nn/control_flow.py
+++ b/python/paddle/static/nn/control_flow.py
@@ -14,8 +14,8 @@
 
 
 from paddle import _C_ops
+from paddle.common_ops_import import LayerHelper
 from paddle.fluid.data_feeder import check_type, check_variable_and_dtype
-from paddle.fluid.layer_helper import LayerHelper
 from paddle.framework import in_dygraph_mode
 
 __all__ = [

From 2080f9db3ab2bbc17372a62e3852e232c5ccc8f2 Mon Sep 17 00:00:00 2001
From: wuzhanfei <wuzhanfei@baidu.com>
Date: Tue, 6 Dec 2022 12:24:35 +0000
Subject: [PATCH 04/16] add interface to paddle.static.nn

---
 .../fleet/utils/hybrid_parallel_inference.py  |  4 +--
 python/paddle/fluid/layers/control_flow.py    |  4 +--
 python/paddle/fluid/layers/rnn.py             |  6 ++--
 python/paddle/fluid/optimizer.py              |  8 ++---
 .../auto_parallel/test_while_op_partition.py  |  4 +--
 .../fluid/tests/unittests/dist_transformer.py |  7 ++--
 .../unittests/dygraph_to_static/test_loop.py  | 10 +++---
 .../unittests/npu/test_increment_op_npu.py    |  2 +-
 .../tests/unittests/npu/test_while_op_npu.py  | 10 +++---
 .../unittests/test_array_read_write_op.py     | 10 +++---
 .../test_async_ssa_graph_executor_mnist.py    |  2 +-
 .../test_dynamic_rnn_stop_gradient.py         |  4 +--
 .../unittests/test_eager_deletion_while_op.py | 10 +++---
 .../tests/unittests/test_executor_and_mul.py  |  6 ++--
 .../unittests/test_fetch_lod_tensor_array.py  |  6 ++--
 .../fluid/tests/unittests/test_profiler.py    |  4 +--
 .../tests/unittests/test_while_loop_op.py     | 32 +++++++++----------
 .../fluid/tests/unittests/test_while_op.py    | 12 +++----
 .../tests/unittests/xpu/test_while_op_xpu.py  | 10 +++---
 .../paddle/jit/dy2static/convert_operators.py |  3 +-
 python/paddle/static/nn/__init__.py           |  3 ++
 21 files changed, 75 insertions(+), 82 deletions(-)

diff --git a/python/paddle/distributed/fleet/utils/hybrid_parallel_inference.py b/python/paddle/distributed/fleet/utils/hybrid_parallel_inference.py
index 13688589383e1..3281884783cf0 100644
--- a/python/paddle/distributed/fleet/utils/hybrid_parallel_inference.py
+++ b/python/paddle/distributed/fleet/utils/hybrid_parallel_inference.py
@@ -62,7 +62,7 @@ class HybridParallelInferenceHelper:
                 element_in_arr = layers.array_read(array=arr, i=step_idx)
                 # write placehold data to global lod_tensor_array,
                 # it need for send_v2 of lod_tensor_array
-                paddle.static.nn.control_flow.increment(x=step_idx, value=1.0, in_place=True)
+                paddle.static.nn.increment(x=step_idx, value=1.0, in_place=True)
                 layers.array_write(element_in_arr, i=step_idx, array=arr)
 
             with paddle.fluid.device_guard(f'{device}:0'):
@@ -135,7 +135,7 @@ class HybridParallelInferenceHelper:
             with while_op.block():
                 with paddle.fluid.device_guard(f'{device}:all'):
                     input = layers.array_read(array=data, i=step_idx)
-                    paddle.static.nn.control_flow.increment(x=step_idx, value=1.0, in_place=True)
+                    paddle.static.nn.increment(x=step_idx, value=1.0, in_place=True)
                     layers.array_write(input, i=step_idx, array=data)
 
                 with paddle.fluid.device_guard(f'{device}:0'):
diff --git a/python/paddle/fluid/layers/control_flow.py b/python/paddle/fluid/layers/control_flow.py
index c2cc745de52c9..ef448f357b87a 100755
--- a/python/paddle/fluid/layers/control_flow.py
+++ b/python/paddle/fluid/layers/control_flow.py
@@ -1142,7 +1142,7 @@ class While:
             cond = paddle.less_than(x=i, y=loop_len)
             while_op = fluid.layers.While(cond=cond)
             with while_op.block():
-                i = paddle.static.nn.control_flow.increment(x=i, value=1, in_place=True)
+                i = paddle.static.nn.increment(x=i, value=1, in_place=True)
                 paddle.assign(paddle.less_than(x=i, y=loop_len), cond)
 
             exe = fluid.Executor(fluid.CPUPlace())
@@ -1170,7 +1170,7 @@ class While:
             with while_op.block():
                 sums_tensor = fluid.layers.elementwise_add(x=data, y=data)
                 fluid.layers.assign(sums_tensor, sums)  # Update the value of sums_tensor defined in While to the sums which defined outside of While through layers.assign
-                i = paddle.static.nn.control_flow.increment(x=i, value=1, in_place=True)
+                i = paddle.static.nn.increment(x=i, value=1, in_place=True)
                 data = fluid.layers.elementwise_add(x=data, y=one)
                 paddle.assign(paddle.less_than(x=i, y=loop_len), cond)
 
diff --git a/python/paddle/fluid/layers/rnn.py b/python/paddle/fluid/layers/rnn.py
index 9df2820f02556..0ec0ba68e997b 100644
--- a/python/paddle/fluid/layers/rnn.py
+++ b/python/paddle/fluid/layers/rnn.py
@@ -1535,9 +1535,8 @@ def _maybe_copy(state, new_state, step_mask):
             next_finished,
             next_sequence_lengths,
         )
-        from paddle.static.nn.control_flow import increment
 
-        increment(x=step_idx_tensor, value=1.0, in_place=True)
+        paddle.static.nn.increment(x=step_idx_tensor, value=1.0, in_place=True)
         step_idx += 1
 
         cond = paddle.logical_not(paddle.all(finished))
@@ -1696,9 +1695,8 @@ def _create_array_out_of_while(dtype):
             outputs,
             outputs_arrays,
         )
-        from paddle.static.nn.control_flow import increment
 
-        increment(x=step_idx, value=1.0, in_place=True)
+        paddle.static.nn.increment(x=step_idx, value=1.0, in_place=True)
         # update the global_finished first, since it might be also in states of
         # decoder, which otherwise would write a stale finished status to array
         tensor.assign(next_finished, global_finished)
diff --git a/python/paddle/fluid/optimizer.py b/python/paddle/fluid/optimizer.py
index 7397db5bf8c23..d49fe0c70159d 100755
--- a/python/paddle/fluid/optimizer.py
+++ b/python/paddle/fluid/optimizer.py
@@ -7283,9 +7283,7 @@ def minimize(self, loss, startup_program=None):
                 dtype='int32',
                 persistable=True,
             )
-            paddle.static.nn.control_flow.increment(
-                x=step, value=1.0, in_place=True
-            )
+            paddle.static.nn.increment(x=step, value=1.0, in_place=True)
 
             # lookahead
             zero_var = layers.fill_constant(
@@ -7519,9 +7517,7 @@ def _get_gm_cond_var(self, main_block):
 
         with device_guard("cpu"):
             # step_var = (step_var + 1) % k_step
-            paddle.static.nn.control_flow.increment(
-                x=step_var, value=1.0, in_place=True
-            )
+            paddle.static.nn.increment(x=step_var, value=1.0, in_place=True)
             main_block.append_op(
                 type='elementwise_mod',
                 inputs={'X': step_var, 'Y': k_step_var},
diff --git a/python/paddle/fluid/tests/unittests/auto_parallel/test_while_op_partition.py b/python/paddle/fluid/tests/unittests/auto_parallel/test_while_op_partition.py
index 92204cc7dd39a..b8d495d553068 100644
--- a/python/paddle/fluid/tests/unittests/auto_parallel/test_while_op_partition.py
+++ b/python/paddle/fluid/tests/unittests/auto_parallel/test_while_op_partition.py
@@ -189,9 +189,7 @@ def get_program():
             cur_pred = mlp_while(pre_input)
 
             # 更新循环条件
-            i = paddle.static.nn.control_flow.increment(
-                x=i, value=1, in_place=True
-            )
+            i = paddle.static.nn.increment(x=i, value=1, in_place=True)
             fluid.layers.array_write(cur_pred, array=input_array, i=i)
             paddle.assign(paddle.less_than(x=i, y=loop_len), cond)
 
diff --git a/python/paddle/fluid/tests/unittests/dist_transformer.py b/python/paddle/fluid/tests/unittests/dist_transformer.py
index fad6c62cc3313..af402adbcfa6f 100644
--- a/python/paddle/fluid/tests/unittests/dist_transformer.py
+++ b/python/paddle/fluid/tests/unittests/dist_transformer.py
@@ -28,7 +28,6 @@
 import paddle.fluid as fluid
 import paddle.fluid.layers as layers
 import paddle.nn.functional as F
-import paddle.static.nn.control_flow as control_flow
 
 const_para_attr = fluid.ParamAttr(initializer=fluid.initializer.Constant(0.001))
 const_bias_attr = const_para_attr
@@ -1814,7 +1813,9 @@ def beam_search():
                     shape=[-1, 1, 1],
                     dtype=pre_ids.dtype,
                 ),
-                y=control_flow.increment(x=step_idx, value=1.0, in_place=False),
+                y=paddle.static.nn.increment(
+                    x=step_idx, value=1.0, in_place=False
+                ),
                 axis=0,
             )
             logits = wrap_decoder(
@@ -1853,7 +1854,7 @@ def beam_search():
                 end_id=eos_idx,
             )
 
-            control_flow.increment(x=step_idx, value=1.0, in_place=True)
+            paddle.static.nn.increment(x=step_idx, value=1.0, in_place=True)
             # update states
             layers.array_write(selected_ids, i=step_idx, array=ids)
             layers.array_write(selected_scores, i=step_idx, array=scores)
diff --git a/python/paddle/fluid/tests/unittests/dygraph_to_static/test_loop.py b/python/paddle/fluid/tests/unittests/dygraph_to_static/test_loop.py
index 9da4598425d07..1efb3669d26e9 100644
--- a/python/paddle/fluid/tests/unittests/dygraph_to_static/test_loop.py
+++ b/python/paddle/fluid/tests/unittests/dygraph_to_static/test_loop.py
@@ -19,9 +19,9 @@
 
 import paddle
 import paddle.fluid as fluid
-import paddle.static.nn.control_flow as control_flow
 from paddle.jit.api import declarative
 from paddle.jit.dy2static.loop_transformer import NameVisitor
+from paddle.static.nn import increment
 from paddle.utils import gast
 
 SEED = 2020
@@ -84,7 +84,7 @@ def while_loop_dyfunc_with_none(x):
 def for_loop_dyfunc(max_len):
     for i in range(max_len):
         ret = fluid.layers.zeros(shape=[1], dtype='float32')
-        control_flow.increment(ret, value=2.0, in_place=True)
+        increment(ret, value=2.0, in_place=True)
     return ret
 
 
@@ -105,14 +105,14 @@ def for_loop_dyfunc2(max_len):
 def for_loop_dyfunc3(max_len):
     ret = fluid.layers.zeros(shape=[1], dtype='float32')
     for i in range(1, 10, 2):
-        control_flow.increment(ret, value=2.0, in_place=True)
+        increment(ret, value=2.0, in_place=True)
     return ret
 
 
 def for_loop_dyfunc4(max_len):
     ret = fluid.layers.zeros(shape=[1], dtype='float32')
     for i in range(10, 1, -2):
-        control_flow.increment(ret, value=2.0, in_place=True)
+        increment(ret, value=2.0, in_place=True)
     return ret
 
 
@@ -120,7 +120,7 @@ def for_loop_dyfunc_not_support(max_len):
     ret = fluid.layers.zeros(shape=[1], dtype='float32')
     a = -2
     for i in range(10, 1, a):
-        control_flow.increment(ret, value=2.0, in_place=True)
+        increment(ret, value=2.0, in_place=True)
     return ret
 
 
diff --git a/python/paddle/fluid/tests/unittests/npu/test_increment_op_npu.py b/python/paddle/fluid/tests/unittests/npu/test_increment_op_npu.py
index 568f4d7a5a768..542ced34256ae 100644
--- a/python/paddle/fluid/tests/unittests/npu/test_increment_op_npu.py
+++ b/python/paddle/fluid/tests/unittests/npu/test_increment_op_npu.py
@@ -121,7 +121,7 @@ def test_npu(self):
 
         with paddle.static.program_guard(main_prog, startup_prog):
             a = paddle.static.data(name="a", shape=[1], dtype='float32')
-            b = paddle.static.nn.control_flow.increment(a)
+            b = paddle.static.nn.increment(a)
 
         place = paddle.NPUPlace(NPUPlace)
 
diff --git a/python/paddle/fluid/tests/unittests/npu/test_while_op_npu.py b/python/paddle/fluid/tests/unittests/npu/test_while_op_npu.py
index 043903b6e5d6e..74c5c64800caa 100644
--- a/python/paddle/fluid/tests/unittests/npu/test_while_op_npu.py
+++ b/python/paddle/fluid/tests/unittests/npu/test_while_op_npu.py
@@ -21,7 +21,7 @@
 from paddle.fluid.backward import append_backward
 import numpy
 from paddle.fluid import compiler, Program, program_guard
-import paddle.static.nn.control_flow as control_flow
+from paddle.static.nn import increment
 
 paddle.enable_static()
 
@@ -44,9 +44,9 @@ def simple_net(self):
         init = layers.zeros(shape=[10], dtype='float32')
         mem_array = layers.array_write(x=init, i=i)
         data_array = layers.array_write(x=d0, i=i)
-        i = control_flow.increment(i)
+        i = increment(i)
         layers.array_write(d1, i, array=data_array)
-        i = control_flow.increment(i)
+        i = increment(i)
         layers.array_write(d2, i, array=data_array)
         i = layers.zeros(shape=[1], dtype='int32')
         i = layers.cast(i, 'int64')
@@ -72,7 +72,7 @@ def simple_net(self):
             prev = layers.array_read(array=mem_array, i=i)
             result = layers.sums(input=[d, prev])
 
-            i = control_flow.increment(x=i, in_place=True)
+            i = increment(x=i, in_place=True)
             layers.array_write(result, i=i, array=mem_array)
             paddle.assign(paddle.less_than(x=i, y=array_len), cond)
 
@@ -81,7 +81,7 @@ def simple_net(self):
                 prev2 = layers.array_read(array=mem_array, i=j)
                 result2 = layers.sums(input=[d2, prev2])
 
-                j = control_flow.increment(x=j, in_place=True)
+                j = increment(x=j, in_place=True)
                 layers.array_write(result2, i=j, array=mem_array)
                 paddle.assign(paddle.less_than(x=j, y=array_len2), cond2)
         sum_result = layers.array_read(array=mem_array, i=j)
diff --git a/python/paddle/fluid/tests/unittests/test_array_read_write_op.py b/python/paddle/fluid/tests/unittests/test_array_read_write_op.py
index 3cb9e7b33dc1e..8f6f3bd58dd49 100644
--- a/python/paddle/fluid/tests/unittests/test_array_read_write_op.py
+++ b/python/paddle/fluid/tests/unittests/test_array_read_write_op.py
@@ -20,28 +20,28 @@
 import paddle.fluid as fluid
 import paddle.fluid.core as core
 import paddle.fluid.layers as layers
-import paddle.static.nn.control_flow as control_flow
 from paddle.fluid import Program, program_guard
 from paddle.fluid.backward import append_backward
 from paddle.fluid.executor import Executor
 from paddle.fluid.framework import default_main_program
+from paddle.static.nn import increment
 
 
 def _test_read_write(x):
     i = layers.zeros(shape=[1], dtype='int64')
     i.stop_gradient = False
     arr = layers.array_write(x=x[0], i=i)
-    i = control_flow.increment(x=i)
+    i = increment(x=i)
     arr = layers.array_write(x=x[1], i=i, array=arr)
-    i = control_flow.increment(x=i)
+    i = increment(x=i)
     arr = layers.array_write(x=x[2], i=i, array=arr)
 
     i = layers.zeros(shape=[1], dtype='int64')
     i.stop_gradient = False
     a0 = layers.array_read(array=arr, i=i)
-    i = control_flow.increment(x=i)
+    i = increment(x=i)
     a1 = layers.array_read(array=arr, i=i)
-    i = control_flow.increment(x=i)
+    i = increment(x=i)
     a2 = layers.array_read(array=arr, i=i)
 
     mean_a0 = paddle.mean(a0)
diff --git a/python/paddle/fluid/tests/unittests/test_async_ssa_graph_executor_mnist.py b/python/paddle/fluid/tests/unittests/test_async_ssa_graph_executor_mnist.py
index 27fb3e36bb007..b3b666aa27af6 100644
--- a/python/paddle/fluid/tests/unittests/test_async_ssa_graph_executor_mnist.py
+++ b/python/paddle/fluid/tests/unittests/test_async_ssa_graph_executor_mnist.py
@@ -62,7 +62,7 @@ def convolutional_neural_network(use_py_reader):
         acc = paddle.static.accuracy(input=prediction, label=label)
         i = fluid.layers.zeros(shape=[1], dtype='int64')
         array = fluid.layers.array_write(x=prediction, i=i)
-        paddle.static.nn.control_flow.increment(i)
+        paddle.static.nn.increment(i)
         fluid.layers.array_write(x=acc, i=i, array=array)
         return array, img, label, prediction, avg_loss, acc, py_reader
 
diff --git a/python/paddle/fluid/tests/unittests/test_dynamic_rnn_stop_gradient.py b/python/paddle/fluid/tests/unittests/test_dynamic_rnn_stop_gradient.py
index f839563344a86..ded90600ed804 100644
--- a/python/paddle/fluid/tests/unittests/test_dynamic_rnn_stop_gradient.py
+++ b/python/paddle/fluid/tests/unittests/test_dynamic_rnn_stop_gradient.py
@@ -51,9 +51,7 @@ def build_and_run_program(place, batch_size, beam_size, stop_gradient=False):
         topk_coordinates = paddle.stack([batch_pos, indices], axis=2)
         topk_coordinates.stop_gradient = stop_gradient
         score = paddle.gather_nd(x, topk_coordinates)
-        paddle.static.nn.control_flow.increment(
-            x=step_idx, value=1.0, in_place=True
-        )
+        paddle.static.nn.increment(x=step_idx, value=1.0, in_place=True)
         layers.array_write(score, i=step_idx, array=scores)
         length_cond = paddle.less_than(x=step_idx, y=max_len)
         layers.assign(length_cond, cond)
diff --git a/python/paddle/fluid/tests/unittests/test_eager_deletion_while_op.py b/python/paddle/fluid/tests/unittests/test_eager_deletion_while_op.py
index bebb534319cb3..2b38eb8f7c68a 100644
--- a/python/paddle/fluid/tests/unittests/test_eager_deletion_while_op.py
+++ b/python/paddle/fluid/tests/unittests/test_eager_deletion_while_op.py
@@ -26,8 +26,8 @@
 import paddle.fluid.compiler as compiler
 import paddle.fluid.core as core
 import paddle.fluid.layers as layers
-import paddle.static.nn.control_flow as control_flow
 from paddle.fluid.executor import Executor
+from paddle.static.nn import increment
 
 paddle.enable_static()
 fluid.core._set_eager_deletion_mode(0.0, 1.0, True)
@@ -84,10 +84,10 @@ def run_main(self, place, with_data_parallel):
         mem_array = layers.array_write(x=init, i=i)
         data_array = layers.array_write(x=d0, i=i)
 
-        i = control_flow.increment(i)
+        i = increment(i)
         layers.array_write(d1, i, array=data_array)
 
-        i = control_flow.increment(i)
+        i = increment(i)
         layers.array_write(d2, i, array=data_array)
 
         i = layers.zeros(shape=[1], dtype='int64')
@@ -113,7 +113,7 @@ def run_main(self, place, with_data_parallel):
             prev = paddle.reshape(prev, shape=[10])
             result = layers.sums(input=[d, prev])
 
-            i = control_flow.increment(x=i, in_place=True)
+            i = increment(x=i, in_place=True)
             layers.array_write(result, i=i, array=mem_array)
             paddle.assign(paddle.less_than(x=i, y=array_len), cond)
             with while_op2.block():
@@ -123,7 +123,7 @@ def run_main(self, place, with_data_parallel):
                 prev2 = paddle.reshape(prev2, shape=[10])
                 result2 = layers.sums(input=[d2, prev2])
 
-                j = control_flow.increment(x=j, in_place=True)
+                j = increment(x=j, in_place=True)
                 layers.array_write(result2, i=j, array=mem_array)
                 paddle.assign(paddle.less_than(x=j, y=array_len2), cond2)
 
diff --git a/python/paddle/fluid/tests/unittests/test_executor_and_mul.py b/python/paddle/fluid/tests/unittests/test_executor_and_mul.py
index 912d5a8490bfe..265e612a13a98 100644
--- a/python/paddle/fluid/tests/unittests/test_executor_and_mul.py
+++ b/python/paddle/fluid/tests/unittests/test_executor_and_mul.py
@@ -16,9 +16,9 @@
 
 import numpy as np
 
-import paddle.static.nn.control_flow as control_flow
 from paddle.fluid.executor import Executor
 from paddle.fluid.layers import array_write, data, mul, zeros
+from paddle.static.nn import increment
 
 
 class TestExecutor(unittest.TestCase):
@@ -27,13 +27,13 @@ def test_mul(self):
         a = data(name='a', shape=[784], dtype='float32')
         array = array_write(x=a, i=i)
 
-        i = control_flow.increment(i)
+        i = increment(i)
         b = data(
             name='b', shape=[784, 100], dtype='float32', append_batch_size=False
         )
         array_write(x=b, i=i, array=array)
 
-        i = control_flow.increment(i)
+        i = increment(i)
         out = mul(x=a, y=b)
         array_write(x=out, i=i, array=array)
 
diff --git a/python/paddle/fluid/tests/unittests/test_fetch_lod_tensor_array.py b/python/paddle/fluid/tests/unittests/test_fetch_lod_tensor_array.py
index 9a127edce18f1..8ca1a4d118fda 100644
--- a/python/paddle/fluid/tests/unittests/test_fetch_lod_tensor_array.py
+++ b/python/paddle/fluid/tests/unittests/test_fetch_lod_tensor_array.py
@@ -20,7 +20,7 @@
 
 import paddle.fluid as fluid
 import paddle.fluid.layers as layers
-import paddle.static.nn.control_flow as control_flow
+from paddle.static.nn import increment
 
 
 class TestFetchLoDTensorArray(unittest.TestCase):
@@ -36,9 +36,9 @@ def build_program(self, main_program, startup_program):
                 opt.minimize(loss)
 
                 array = layers.array_write(x=img, i=i)
-                i = control_flow.increment(i)
+                i = increment(i)
                 layers.array_write(x=label, i=i, array=array)
-                i = control_flow.increment(i)
+                i = increment(i)
                 layers.array_write(x=loss, i=i, array=array)
 
                 return loss, array
diff --git a/python/paddle/fluid/tests/unittests/test_profiler.py b/python/paddle/fluid/tests/unittests/test_profiler.py
index 0f7459dd96893..629dc72e3ae2b 100644
--- a/python/paddle/fluid/tests/unittests/test_profiler.py
+++ b/python/paddle/fluid/tests/unittests/test_profiler.py
@@ -50,9 +50,7 @@ def build_program(self, compile_program=True):
             with while_op.block():
                 hidden_n = fluid.layers.fc(input=hidden1, size=64, act='relu')
                 layers.array_write(hidden_n, i, data_arr)
-                paddle.static.nn.control_flow.increment(
-                    x=counter, value=1, in_place=True
-                )
+                paddle.static.nn.increment(x=counter, value=1, in_place=True)
                 paddle.assign(paddle.less_than(x=counter, y=until), cond)
 
             hidden_n = layers.array_read(data_arr, i)
diff --git a/python/paddle/fluid/tests/unittests/test_while_loop_op.py b/python/paddle/fluid/tests/unittests/test_while_loop_op.py
index 19193fb348399..fbcc6b568f325 100644
--- a/python/paddle/fluid/tests/unittests/test_while_loop_op.py
+++ b/python/paddle/fluid/tests/unittests/test_while_loop_op.py
@@ -20,9 +20,9 @@
 import paddle.fluid as fluid
 import paddle.fluid.core as core
 import paddle.fluid.layers as layers
-import paddle.static.nn.control_flow as control_flow
 from paddle.fluid.backward import append_backward
 from paddle.fluid.framework import Program, program_guard
+from paddle.static.nn import increment
 
 paddle.enable_static()
 
@@ -60,7 +60,7 @@ def cond(i, mem):
 
         def body(i, mem):
             mem = paddle.add(x=mem, y=one)
-            i = control_flow.increment(i)
+            i = increment(i)
             return [i, mem]
 
         main_program = Program()
@@ -101,7 +101,7 @@ def body(i, ten, test_dict, test_list, test_list_dict):
                 test_list_dict[0]["test_key"]
             )
 
-            i = control_flow.increment(i)
+            i = increment(i)
             return [i, ten, test_dict, test_list, test_list_dict]
 
         main_program = Program()
@@ -169,7 +169,7 @@ def internal_cond(j, init, sums):
             def internal_body(j, init, sums):
                 init = paddle.add(x=init, y=ones)
                 sums = paddle.add(x=init, y=sums)
-                j = control_flow.increment(j)
+                j = increment(j)
                 return [j, init, sums]
 
             result = layers.while_loop(
@@ -179,7 +179,7 @@ def internal_body(j, init, sums):
             init = result[1]
             sums = result[2]
             sums = paddle.add(x=init, y=sums)
-            i = control_flow.increment(i)
+            i = increment(i)
             return [i, j, init, sums]
 
         main_program = Program()
@@ -224,7 +224,7 @@ def cond(i, x):
 
         def body(i, x):
             x = paddle.multiply(x=i, y=i)
-            i = control_flow.increment(i)
+            i = increment(i)
             return [i, x]
 
         main_program = Program()
@@ -319,7 +319,7 @@ def internal_body(j, x, mem_array):
                 inner_prev = layers.array_read(array=mem_array, i=j)
                 inner_sum_0 = paddle.add(x=inner_data, y=inner_prev)
                 inner_sum_1 = paddle.add(x=x, y=inner_sum_0)
-                j = control_flow.increment(x=j, in_place=True)
+                j = increment(x=j, in_place=True)
                 layers.array_write(inner_sum_1, i=j, array=mem_array)
                 return [j, x, mem_array]
 
@@ -327,7 +327,7 @@ def internal_body(j, x, mem_array):
             outer_prev = layers.array_read(array=mem_array, i=i)
             outer_sum_0 = paddle.add(x=outer_data, y=outer_prev)
             outer_sum_1 = paddle.add(x=x, y=outer_sum_0)
-            i = control_flow.increment(x=i, in_place=True)
+            i = increment(x=i, in_place=True)
             layers.array_write(outer_sum_1, i=i, array=mem_array)
             j, x, mem_array = layers.while_loop(
                 internal_cond, internal_body, [j, x, mem_array]
@@ -347,9 +347,9 @@ def internal_body(j, x, mem_array):
             init = layers.zeros(shape=[10], dtype='float32')
             mem_array = layers.array_write(x=init, i=i)
             data_array = layers.array_write(x=d0, i=i)
-            i = control_flow.increment(i)
+            i = increment(i)
             layers.array_write(d1, i, array=data_array)
-            i = control_flow.increment(i)
+            i = increment(i)
             layers.array_write(d2, i, array=data_array)
             i = layers.zeros(shape=[1], dtype='int64')
             i.stop_gradient = True
@@ -439,7 +439,7 @@ def cond_returns_constant(i):
             return 1
 
         def cond_returns_not_bool_tensor(i):
-            return control_flow.increment(i)
+            return increment(i)
 
         def cond_returns_bool_tensor(i):
             return paddle.less_than(i, ten)
@@ -451,14 +451,14 @@ def cond_receives_two_args(i, ten):
             return paddle.less_than(i, ten)
 
         def body(i):
-            return control_flow.increment(i)
+            return increment(i)
 
         def body_returns_error_length(i):
-            i = control_flow.increment(i)
+            i = increment(i)
             return [i, i]
 
         def body_returns_error_type(i, ten):
-            return control_flow.increment(i)
+            return increment(i)
 
         def cond_returns_with_mutable_dict(i, test_dict):
             return i > 0
@@ -467,7 +467,7 @@ def body_returns_with_mutable_dict(i, test_dict):
             test_dict['new_key'] = layers.fill_constant(
                 shape=[1], dtype='int64', value=1
             )
-            return control_flow.increment(i), test_dict
+            return increment(i), test_dict
 
         def cond_returns_with_mutable_list(i, test_list):
             return i > 0
@@ -476,7 +476,7 @@ def body_returns_with_mutable_list(i, test_list):
             test_list.append(
                 layers.fill_constant(shape=[1], dtype='int64', value=1)
             )
-            return control_flow.increment(i), test_list
+            return increment(i), test_list
 
         main_program = Program()
         startup_program = Program()
diff --git a/python/paddle/fluid/tests/unittests/test_while_op.py b/python/paddle/fluid/tests/unittests/test_while_op.py
index b277cb334480b..3ecdb84821ac9 100644
--- a/python/paddle/fluid/tests/unittests/test_while_op.py
+++ b/python/paddle/fluid/tests/unittests/test_while_op.py
@@ -20,9 +20,9 @@
 import paddle.fluid as fluid
 import paddle.fluid.core as core
 import paddle.fluid.layers as layers
-import paddle.static.nn.control_flow as control_flow
 from paddle.fluid.backward import append_backward
 from paddle.fluid.executor import Executor
+from paddle.static.nn import increment
 
 paddle.enable_static()
 
@@ -43,9 +43,9 @@ def simple_net(self):
         init = layers.zeros(shape=[10], dtype='float32')
         mem_array = layers.array_write(x=init, i=i)
         data_array = layers.array_write(x=d0, i=i)
-        i = control_flow.increment(i)
+        i = increment(i)
         layers.array_write(d1, i, array=data_array)
-        i = control_flow.increment(i)
+        i = increment(i)
         layers.array_write(d2, i, array=data_array)
         i = layers.zeros(shape=[1], dtype='int64')
         i.stop_gradient = True
@@ -64,7 +64,7 @@ def simple_net(self):
             prev = layers.array_read(array=mem_array, i=i)
             result = layers.sums(input=[d, prev])
 
-            i = control_flow.increment(x=i, in_place=True)
+            i = increment(x=i, in_place=True)
             layers.array_write(result, i=i, array=mem_array)
             paddle.assign(paddle.less_than(x=i, y=array_len), cond)
 
@@ -73,7 +73,7 @@ def simple_net(self):
                 prev2 = layers.array_read(array=mem_array, i=j)
                 result2 = layers.sums(input=[d2, prev2])
 
-                j = control_flow.increment(x=j, in_place=True)
+                j = increment(x=j, in_place=True)
                 layers.array_write(result2, i=j, array=mem_array)
                 paddle.assign(paddle.less_than(x=j, y=array_len2), cond2)
         sum_result = layers.array_read(array=mem_array, i=j)
@@ -135,7 +135,7 @@ def test_error(self):
 
             def test_bad_x():
                 x = [1, 2, 3]
-                control_flow.increment(x)
+                increment(x)
 
             self.assertRaises(TypeError, test_bad_x)
 
diff --git a/python/paddle/fluid/tests/unittests/xpu/test_while_op_xpu.py b/python/paddle/fluid/tests/unittests/xpu/test_while_op_xpu.py
index 33056f8c9b832..a869573160beb 100644
--- a/python/paddle/fluid/tests/unittests/xpu/test_while_op_xpu.py
+++ b/python/paddle/fluid/tests/unittests/xpu/test_while_op_xpu.py
@@ -19,9 +19,9 @@
 import paddle
 import paddle.fluid as fluid
 import paddle.fluid.layers as layers
-import paddle.static.nn.control_flow as control_flow
 from paddle.fluid.backward import append_backward
 from paddle.fluid.executor import Executor
+from paddle.static.nn import increment
 
 paddle.enable_static()
 
@@ -42,9 +42,9 @@ def simple_net(self):
         init = layers.zeros(shape=[10], dtype='float32')
         mem_array = layers.array_write(x=init, i=i)
         data_array = layers.array_write(x=d0, i=i)
-        i = control_flow.increment(i)
+        i = increment(i)
         layers.array_write(d1, i, array=data_array)
-        i = control_flow.increment(i)
+        i = increment(i)
         layers.array_write(d2, i, array=data_array)
         i = layers.zeros(shape=[1], dtype='int64')
         i.stop_gradient = True
@@ -63,7 +63,7 @@ def simple_net(self):
             prev = layers.array_read(array=mem_array, i=i)
             result = layers.sums(input=[d, prev])
 
-            i = control_flow.increment(x=i, in_place=True)
+            i = increment(x=i, in_place=True)
             layers.array_write(result, i=i, array=mem_array)
             paddle.assign(paddle.less_than(x=i, y=array_len), cond)
 
@@ -72,7 +72,7 @@ def simple_net(self):
                 prev2 = layers.array_read(array=mem_array, i=j)
                 result2 = layers.sums(input=[d2, prev2])
 
-                j = control_flow.increment(x=j, in_place=True)
+                j = increment(x=j, in_place=True)
                 layers.array_write(result2, i=j, array=mem_array)
                 paddle.assign(paddle.less_than(x=j, y=array_len2), cond2)
         sum_result = layers.array_read(array=mem_array, i=j)
diff --git a/python/paddle/jit/dy2static/convert_operators.py b/python/paddle/jit/dy2static/convert_operators.py
index a5d2ca14cd9b6..6db2bdafb4bf7 100644
--- a/python/paddle/jit/dy2static/convert_operators.py
+++ b/python/paddle/jit/dy2static/convert_operators.py
@@ -784,7 +784,8 @@ def cond(i, new_array):
     def body(i, new_array):
         item = array_read(array=array, i=i)
         array_write(item, paddle.tensor.array_length(new_array), new_array)
-        from paddle.static.nn.control_flow import increment
+
+        from paddle.static.nn import increment
 
         i = increment(i)
         return i, new_array
diff --git a/python/paddle/static/nn/__init__.py b/python/paddle/static/nn/__init__.py
index 9635811f6a818..b005c6f556097 100755
--- a/python/paddle/static/nn/__init__.py
+++ b/python/paddle/static/nn/__init__.py
@@ -33,6 +33,7 @@
 from ...fluid.layers import multi_box_head  # noqa: F401
 from .loss import nce  # noqa: F401
 from .common import prelu  # noqa: F401
+from .control_flow import Assert, increment  # noqa: F401
 from ...fluid.layers import row_conv  # noqa: F401
 from ...fluid.layers import spectral_norm  # noqa: F401
 from ...fluid.layers import switch_case  # noqa: F401
@@ -100,4 +101,6 @@
     'sequence_reverse',
     'StaticRNN',
     'prelu',
+    'Assert',
+    'increment',
 ]

From 1823bcf2352539136b168284cb851e8dea547366 Mon Sep 17 00:00:00 2001
From: wuzhanfei <wuzhanfei@baidu.com>
Date: Tue, 6 Dec 2022 12:37:24 +0000
Subject: [PATCH 05/16] fixed import statments

---
 .../collective/fleet/hybrid_parallel_inference_helper.py        | 2 +-
 python/paddle/fluid/tests/unittests/test_assert_op.py           | 2 +-
 python/paddle/jit/dy2static/convert_operators.py                | 2 +-
 3 files changed, 3 insertions(+), 3 deletions(-)

diff --git a/python/paddle/fluid/tests/unittests/collective/fleet/hybrid_parallel_inference_helper.py b/python/paddle/fluid/tests/unittests/collective/fleet/hybrid_parallel_inference_helper.py
index c903359f68239..1edb4bdfccbd8 100644
--- a/python/paddle/fluid/tests/unittests/collective/fleet/hybrid_parallel_inference_helper.py
+++ b/python/paddle/fluid/tests/unittests/collective/fleet/hybrid_parallel_inference_helper.py
@@ -89,7 +89,7 @@ def test_hybrid_parallel_inference_helper_mp1pp2(self):
             with while_op.block():
                 with paddle.fluid.device_guard(f'{device}:all'):
                     input = layers.array_read(array=data, i=step_idx)
-                    paddle.static.nn.control_flow.increment(
+                    paddle.static.nn.increment(
                         x=step_idx, value=1.0, in_place=True
                     )
                     layers.array_write(input, i=step_idx, array=data)
diff --git a/python/paddle/fluid/tests/unittests/test_assert_op.py b/python/paddle/fluid/tests/unittests/test_assert_op.py
index f62ad38d459d1..a6ce260fb9435 100644
--- a/python/paddle/fluid/tests/unittests/test_assert_op.py
+++ b/python/paddle/fluid/tests/unittests/test_assert_op.py
@@ -17,7 +17,7 @@
 import paddle
 import paddle.fluid as fluid
 import paddle.fluid.layers as layers
-from paddle.static.nn.control_flow import Assert
+from paddle.static.nn import Assert
 
 
 class TestAssertOp(unittest.TestCase):
diff --git a/python/paddle/jit/dy2static/convert_operators.py b/python/paddle/jit/dy2static/convert_operators.py
index 6db2bdafb4bf7..8d99b534cb721 100644
--- a/python/paddle/jit/dy2static/convert_operators.py
+++ b/python/paddle/jit/dy2static/convert_operators.py
@@ -730,7 +730,7 @@ def convert_assert(cond, message=""):
     if isinstance(cond, Variable):
         cond = cast(cond, "bool")
         # NOTE: message is not used because Paddle Assert has no corresponding parameter to use.
-        from paddle.static.nn.control_flow import Assert
+        from paddle.static.nn import Assert
 
         return Assert(cond)
     else:

From 39287d774ab74d4f2c99fce3caf4189b41c98869 Mon Sep 17 00:00:00 2001
From: wuzhanfei <wuzhanfei@baidu.com>
Date: Wed, 7 Dec 2022 03:00:02 +0000
Subject: [PATCH 06/16] fix static err

---
 python/paddle/static/nn/control_flow.py | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/python/paddle/static/nn/control_flow.py b/python/paddle/static/nn/control_flow.py
index 1eecde1cb6c86..edf90babea525 100644
--- a/python/paddle/static/nn/control_flow.py
+++ b/python/paddle/static/nn/control_flow.py
@@ -58,7 +58,7 @@ def Assert(cond, data=None, summarize=20, name=None):
             import paddle
             from paddle.static.nn.control_flow import Assert
 
-            x = paddle.full(shape=[2, 3], dtype='float32', value=2.0)
+            x = paddle.full([2, 3], 2.0, 'float32')
             condition = paddle.max(x) < 1.0 # False
             Assert(condition, [x], 10, "example_assert_layer")
 

From 75e1b0ea26a390017ee930eb5282479c1cf268d1 Mon Sep 17 00:00:00 2001
From: wuzhanfei <wuzhanfei@baidu.com>
Date: Wed, 7 Dec 2022 06:09:30 +0000
Subject: [PATCH 07/16] migrate cond api

---
 .../passes/auto_parallel_gradient_merge.py    |   2 +-
 .../contrib/slim/quantization/adaround.py     |   2 +-
 python/paddle/fluid/layers/control_flow.py    | 501 +-----------------
 python/paddle/fluid/optimizer.py              |   2 +-
 .../dygraph_to_static/ifelse_simple_func.py   |   2 +-
 .../unittests/dygraph_to_static/test_dict.py  |   2 +-
 .../dygraph_to_static/test_ifelse.py          |   4 +-
 .../dygraph_to_static/test_warning.py         |   2 +-
 .../ir/test_ir_subgraph_python_interface.py   |   2 +-
 .../test_standalone_controlflow.py            |   2 +-
 .../paddle/fluid/tests/unittests/test_cond.py |  58 +-
 .../fluid/tests/unittests/test_desc_clone.py  |   4 +-
 .../fluid/tests/unittests/test_layers.py      |  18 +-
 .../tests/unittests/test_math_op_patch.py     |   2 +-
 .../tests/unittests/test_optimizer_grad.py    |   2 +-
 .../tests/unittests/test_program_code.py      |   2 +-
 python/paddle/fluid/variable_index.py         |   4 +-
 python/paddle/static/nn/__init__.py           |   9 +-
 python/paddle/static/nn/control_flow.py       | 478 ++++++++++++++++-
 19 files changed, 557 insertions(+), 541 deletions(-)

diff --git a/python/paddle/distributed/passes/auto_parallel_gradient_merge.py b/python/paddle/distributed/passes/auto_parallel_gradient_merge.py
index 8ac3492c2b14d..592db6767c0fa 100644
--- a/python/paddle/distributed/passes/auto_parallel_gradient_merge.py
+++ b/python/paddle/distributed/passes/auto_parallel_gradient_merge.py
@@ -286,7 +286,7 @@ def true_apply_gradient():
             )
             new_grad.op._set_attr(OP_ROLE_KEY, op_maker.OpRole.Optimize)
 
-    layers.cond(cond_var, true_fn=true_apply_gradient, false_fn=None)
+    paddle.static.nn.cond(cond_var, true_fn=true_apply_gradient, false_fn=None)
     cond_op = main_program.global_block().ops[-1]
     cond_op._set_attr(OP_ROLE_KEY, OpRole.Optimize)
 
diff --git a/python/paddle/fluid/contrib/slim/quantization/adaround.py b/python/paddle/fluid/contrib/slim/quantization/adaround.py
index b024c0d773996..f4bbccd7f1d86 100644
--- a/python/paddle/fluid/contrib/slim/quantization/adaround.py
+++ b/python/paddle/fluid/contrib/slim/quantization/adaround.py
@@ -83,7 +83,7 @@ def round_loss_fn():
 
             return round_loss
 
-        round_loss = fluid.layers.cond(
+        round_loss = paddle.static.nn.cond(
             warm_start,
             lambda: fluid.layers.fill_constant(
                 shape=[1], dtype='float32', value=0.0
diff --git a/python/paddle/fluid/layers/control_flow.py b/python/paddle/fluid/layers/control_flow.py
index ef448f357b87a..0d7770773a0d8 100755
--- a/python/paddle/fluid/layers/control_flow.py
+++ b/python/paddle/fluid/layers/control_flow.py
@@ -56,7 +56,6 @@
     'Switch',
     'array_write',
     'array_read',
-    'cond',
     'IfElse',
     'StaticRNN',
     'reorder_lod_tensor_by_rank',
@@ -97,26 +96,6 @@ def select_output(input, outputs, mask):
     return outputs
 
 
-def _select_input_infer_shape(first_shape, second_shape):
-    """
-    This function infer the output shape by following algorithm:
-    1. if the dims is different, raise a error.
-    2. compare axis one by one:
-        if a == b: we set axis to a
-        if a != b: we set axis to -1
-    for compatibility，non declarative mode, we just return second_shape.
-    """
-    if len(first_shape) != len(second_shape):
-        warnings.warn(
-            f"the input shapes of select_input should have the same rank, but get {first_shape}, {second_shape}"
-        )
-        return second_shape
-    out_shape = list(
-        map(lambda a, b: a if a == b else -1, first_shape, second_shape)
-    )
-    return out_shape
-
-
 def select_input(inputs, mask):
     """
     **select_input**
@@ -138,6 +117,25 @@ def select_input(inputs, mask):
 
     # Select input should expand the shape. If it is - 1 and valid number, use - 1 first. If the dim is different, an error will be reported directly
     # assert inputs[0].dtype == inputs[1].dtype, f"Expect the inputs should have the same dtype, but get {inputs[0].dtype} and {inputs[1].dtype}"
+    def _select_input_infer_shape(first_shape, second_shape):
+        """
+        This function infer the output shape by following algorithm:
+        1. if the dims is different, raise a error.
+        2. compare axis one by one:
+            if a == b: we set axis to a
+            if a != b: we set axis to -1
+        for compatibility，non declarative mode, we just return second_shape.
+        """
+        if len(first_shape) != len(second_shape):
+            warnings.warn(
+                f"the input shapes of select_input should have the same rank, but get {first_shape}, {second_shape}"
+            )
+            return second_shape
+        out_shape = list(
+            map(lambda a, b: a if a == b else -1, first_shape, second_shape)
+        )
+        return out_shape
+
     output_shape = _select_input_infer_shape(inputs[0].shape, inputs[1].shape)
     output_dtype = inputs[1].dtype
     output_type = inputs[1].type
@@ -153,84 +151,6 @@ def select_input(inputs, mask):
     return out
 
 
-def select_input_with_buildin_type(inputs, mask, name):
-    from paddle.jit.dy2static.variable_trans_func import (
-        to_static_variable,
-    )
-    from paddle.jit.dy2static.utils import UndefinedVar
-
-    false_var, true_var = inputs
-
-    if isinstance(false_var, UndefinedVar) and isinstance(
-        true_var, UndefinedVar
-    ):
-        """None -> UndefinedVar, so the real value is a [None, UndefinedVar] or [None, None], we just return None."""
-        return None
-
-    if isinstance(false_var, Variable) and isinstance(true_var, Variable):
-        try:
-            return select_input(inputs, mask)
-        except Exception as e:
-            raise RuntimeError(
-                f"Exceptions throwed while doing select_input on {name}:\n{e}"
-            )
-
-    elif isinstance(false_var, support_ret_buildin_type) and isinstance(
-        false_var, type(true_var)
-    ):
-        if false_var == true_var:
-            return false_var
-        else:
-            inputs = [
-                to_static_variable(false_var),
-                to_static_variable(true_var),
-            ]
-    # Deal with the situations like this: false_var is int and true_var is Variable
-    elif (
-        isinstance(false_var, support_ret_buildin_type)
-        and isinstance(true_var, Variable)
-    ) or (
-        isinstance(true_var, support_ret_buildin_type)
-        and isinstance(false_var, Variable)
-    ):
-        inputs = [to_static_variable(false_var), to_static_variable(true_var)]
-        warnings.warn(
-            "Return results from different branches in cond are not same type: "
-            "false_var returned by false_fn is '{}' and true_var of true_fn is "
-            "'{}'".format(type(false_var), type(true_var))
-        )
-    elif (
-        isinstance(false_var, UndefinedVar)
-        and isinstance(true_var, (Variable,) + support_ret_buildin_type)
-    ) or (
-        isinstance(true_var, UndefinedVar)
-        and isinstance(false_var, (Variable,) + support_ret_buildin_type)
-    ):
-
-        def create_var_if_not_undefined_var(a):
-            if isinstance(a, UndefinedVar):
-                return a
-            return to_static_variable(a)
-
-        true_var, false_var = to_static_variable(true_var), to_static_variable(
-            false_var
-        )
-        inputs = [false_var, true_var]
-    else:
-        raise TypeError(
-            "Unsupported return type of true_fn and false_fn in cond: false_var "
-            "returned by false_fn is '{}' and true_var of true_fn is '{}'".format(
-                type(false_var), type(true_var)
-            )
-        )
-    try:
-        return select_input(inputs, mask)
-    except Exception as e:
-        raise RuntimeError(
-            f"Exceptions throwed while doing select_input on {name}:\n{e}"
-        )
-
-
 def split_lod_tensor(input, mask, level=0):
     """
     This function takes in an input that contains the complete lod information,
@@ -1915,315 +1835,6 @@ def append_conditional_block_grad(
         self.helper.main_program._sync_with_cpp()
 
 
-def copy_var_to_parent_block(var, layer_helper):
-    if not isinstance(var, Variable):
-        return var
-    prog = layer_helper.main_program
-    parent_idx = prog.current_block().parent_idx
-    assert (
-        parent_idx >= 0
-    ), "Got wrong parent block index when assigning var to parent scope in control_flow"
-    parent_block = prog.block(parent_idx)
-
-    if (
-        var.type == core.VarDesc.VarType.LOD_TENSOR_ARRAY
-        and parent_block._find_var_recursive(var.name)
-    ):
-        parent_block_var = var
-    else:
-        parent_block_var = parent_block.create_var(
-            dtype=var.dtype, shape=var.shape, type=var.type
-        )
-        assign(var, parent_block_var)
-    return parent_block_var
-
-
-def cond(pred, true_fn=None, false_fn=None, name=None, return_names=None):
-    """
-    This API returns ``true_fn()`` if the predicate ``pred`` is true else
-    ``false_fn()`` . Users could also set ``true_fn`` or ``false_fn`` to
-    ``None`` if do nothing and this API will treat the callable simply returns
-    ``None`` in this case.
-
-    ``true_fn`` and ``false_fn`` should return same nest structure of tensors
-    or both return ``None`` if user doens't like to return anything. A nest
-    structure of tensors in PaddlePaddle is tensor(s), or tuple of tensors, or
-    list of tensors.
-
-    Note:
-        1. The tuples or lists returned by ``true_fn`` and ``false_fn`` must have
-        the same shape because of dataflow model of PaddlePaddle while the
-        tensors in the tuples or the lists can have different shapes.
-
-        2. This API could be used under both static mode or dygraph mode. If it
-        is in dygraph mode, the API only runs one branch based on condition.
-
-        3. If it is in static mode, any tensors or operations created outside
-        or inside of ``true_fn`` and ``false_fn`` will be in net building
-        regardless of which branch is selected at runtime. This has frequently
-        surprised users who expected a lazy semantics. For example:
-
-        .. code-block:: python
-
-            import paddle
-
-            a = paddle.zeros((1, 1))
-            b = paddle.zeros((1, 1))
-            c = a * b
-            out = paddle.static.nn.cond(a < b, lambda: a + c, lambda: b * b)
-
-        No matter whether ``a < b`` , ``c = a * b`` will be in net building and
-        run. ``a + c`` and ``b * b`` will be in net building, but only one
-        branch will be executed during runtime.
-
-    Args:
-        pred(Tensor): A boolean tensor whose numel should be 1. The boolean
-            value determines whether to return the result of ``true_fn`` or
-            ``false_fn`` .
-        true_fn(callable, optional): A callable to be performed if ``pred`` is
-            true. The default value is ``None`` .
-        false_fn(callable, optional): A callable to be performed if ``pred`` is
-            false. The default value is ``None`` .
-        name(str, optional): The default value is ``None`` . Normally users
-             don't have to set this parameter. For more information, please
-             refer to :ref:`api_guide_Name` .
-        return_names(sequence of string, optional): The default value is ``None`` .
-             Normally users don't have to set this parameters.  A sequence of strings
-             to represents the name of returned vars.  The structure of sequence must
-             be same with return values of true_fn and false_fn.
-
-    Returns:
-        Tensor|list(Tensor)|tuple(Tensor): returns ``true_fn()`` if the
-        predicate ``pred`` is true else ``false_fn()`` .
-
-    Raises:
-        TypeError: if ``true_fn`` or ``false_fn`` is not callable.
-        ValueError: if ``true_fn`` and ``false_fn`` don't return the same nest
-            structure of tensors.
-
-    Examples:
-        .. code-block:: python
-
-            import paddle
-
-            #
-            # pseudocode:
-            # if 0.1 < 0.23:
-            #     return 1, True
-            # else:
-            #     return 3, 2
-            #
-
-            def true_func():
-                return paddle.full(shape=[1, 2], dtype='int32',
-                                   fill_value=1), paddle.full(shape=[2, 3],
-                                                              dtype='bool',
-                                                              fill_value=True)
-
-
-            def false_func():
-                return paddle.full(shape=[3, 4], dtype='float32',
-                                   fill_value=3), paddle.full(shape=[4, 5],
-                                                              dtype='int64',
-                                                              fill_value=2)
-
-
-            x = paddle.full(shape=[1], dtype='float32', fill_value=0.1)
-            y = paddle.full(shape=[1], dtype='float32', fill_value=0.23)
-            pred = paddle.less_than(x=x, y=y, name=None)
-            ret = paddle.static.nn.cond(pred, true_func, false_func)
-            # ret is a tuple containing 2 tensors
-            # ret[0] = [[1 1]]
-            # ret[1] = [[ True  True  True]
-            #           [ True  True  True]]
-
-    """
-    if _non_static_mode():
-        assert isinstance(pred, Variable), "The pred in cond must be Variable"
-        assert pred.size == 1, "condition input's numel should be 1"
-        pred = pred.numpy()[0]
-        if pred:
-            if true_fn is not None:
-                if not callable(true_fn):
-                    raise TypeError(
-                        "The true_fn in cond must be callable, but received {}".format(
-                            type(true_fn).__name__
-                        )
-                    )
-                return true_fn()
-        else:
-            if false_fn is not None:
-                if not callable(false_fn):
-                    raise TypeError(
-                        "The false_fn in cond must be callable, but received {}".format(
-                            type(false_fn).__name__
-                        )
-                    )
-                return false_fn()
-        return None
-
-    check_variable_and_dtype(pred, "pred", ['bool'], "fluid.layers.cond")
-    check_type(name, "name", (str, type(None)), "fluid.layers.cond")
-    helper = LayerHelper('cond', **locals())
-    true_output = None
-    false_output = None
-    copy_to_parent_func = lambda var: copy_var_to_parent_block(var, helper)
-    if true_fn is not None:
-        if not callable(true_fn):
-            raise TypeError(
-                "The true_fn in cond must be callable, but received {}".format(
-                    type(true_fn).__name__
-                )
-            )
-        true_cond_block = ConditionalBlock([pred], is_scalar_condition=True)
-        with true_cond_block.block():
-            origin_true_output = true_fn()
-            if origin_true_output is not None:
-                true_output = map_structure(
-                    copy_to_parent_func, origin_true_output
-                )
-    if false_fn is not None:
-        if not callable(false_fn):
-            raise TypeError(
-                "The false_fn in cond must be callable, but received {}".format(
-                    type(false_fn).__name__
-                )
-            )
-        false_cond_block = ConditionalBlock(
-            [paddle.logical_not(pred)], is_scalar_condition=True
-        )
-        with false_cond_block.block():
-            origin_false_output = false_fn()
-            if origin_false_output is not None:
-                false_output = map_structure(
-                    copy_to_parent_func, origin_false_output
-                )
-
-    if true_output is None and false_output is None:
-        return None
-
-    if true_output is None:
-        raise ValueError(
-            "Incompatible return values of true_fn and false_fn in cond: "
-            "true_fn returns None while false_fn returns non-None"
-        )
-    if false_output is None:
-        raise ValueError(
-            "Incompatible return values of true_fn and false_fn in cond: "
-            "true_fn returns non-None while false_fn returns None"
-        )
-
-    # Merge true and false output if they are not None
-    if return_names is None:
-        is_dy2staic = False
-        return_names = ["no name"] * len(_to_sequence_except_dict(true_output))
-    else:
-        """
-        dy2static will set the return_names and expand the return values to UndefinedVar.
-        """
-        is_dy2staic = True
-
-        # TODO:  expand_undefined_var will replace None to Undefinedvar(), to fix cases like:
-        #       a = None
-        #       if condition:
-        #           a = 1
-        # Because we can not use variable to express 'None'
-        true_output, false_output = expand_undefined_var(
-            true_output, false_output, return_names
-        )
-
-    if len(_to_sequence_except_dict(true_output)) != len(
-        _to_sequence_except_dict(false_output)
-    ):
-        raise ValueError(
-            "true fn returns {} vars, but false fn returns {} vars, which is not equals".format(
-                len(_to_sequence_except_dict(true_output)),
-                len(_to_sequence_except_dict(false_output)),
-            )
-        )
-    for true_out, false_out, return_name in zip(
-        _to_sequence_except_dict(true_output),
-        _to_sequence_except_dict(false_output),
-        _to_sequence_except_dict(return_names),
-    ):
-        try:
-            assert_same_structure(true_out, false_out, check_types=False)
-        except ValueError as e:
-            raise ValueError(
-                "Incompatible return values of `{}` in true_fn and false_fn in cond: {}".format(
-                    return_name, e
-                )
-            )
-
-    def check_ret_none(seq_true, seq_false, seq_names):
-        for f_true, f_false, f_name in zip(seq_true, seq_false, seq_names):
-            f_true = flatten(f_true)
-            f_false = flatten(f_false)
-            for idx in range(len(f_true)):
-                if (
-                    f_true[idx] is None
-                    and f_false[idx] is not None
-                    or f_false[idx] is None
-                    and f_true[idx] is not None
-                ):
-                    warnings.warn(
-                        "In cond : Var '{}' or part of it is set differently in ifelse branchs, "
-                        "<{}, {}> in true branch and <{}, {}> in false branch. Set var to "
-                        "'None' in ifelse block might lead to error.".format(
-                            f_name,
-                            type(f_true[idx]),
-                            f_true[idx],
-                            type(f_false[idx]),
-                            f_false[idx],
-                        )
-                    )
-
-    check_ret_none(
-        _to_sequence_except_dict(true_output),
-        _to_sequence_except_dict(false_output),
-        _to_sequence_except_dict(return_names),
-    )
-
-    if is_dy2staic:
-        true_output, false_output = change_none_to_undefinedvar(
-            true_output, false_output
-        )
-
-    mask = cast(pred, dtype='int32')
-    merge_func = (
-        lambda name, false_var, true_var: select_input_with_buildin_type(
-            [false_var, true_var], mask, name
-        )
-    )
-
-    def merge_every_var_list(false_vars, true_vars, name):
-        return map_structure(partial(merge_func, name), false_vars, true_vars)
-
-    merged_output = list(
-        map(
-            merge_every_var_list,
-            _to_sequence_except_dict(false_output),
-            _to_sequence_except_dict(true_output),
-            _to_sequence_except_dict(return_names),
-        )
-    )
-    merged_output = pack_sequence_as(false_output, flatten(merged_output))
-    return merged_output
-
-
-def change_none_to_undefinedvar(nest1, nest2):
-    from paddle.jit.dy2static.utils import UndefinedVar
-
-    def map_fn(x):
-        if x is None:
-            return UndefinedVar("padding")
-        return x
-
-    nest1_out = pack_sequence_as(nest1, list(map(map_fn, flatten(nest1))))
-    nest2_out = pack_sequence_as(nest2, list(map(map_fn, flatten(nest2))))
-    return nest1_out, nest2_out
-
-
 def _to_sequence_except_dict(x):
     """
     In this function, dict is not viewed as sequence.
@@ -2233,80 +1844,6 @@ def _to_sequence_except_dict(x):
     return to_sequence(x)
 
 
-def _is_sequence_except_dict(x):
-    """
-    In this function, dict is not viewed as sequence.
-    """
-    if isinstance(x, dict):
-        return False
-    return is_sequence(x)
-
-
-def expand_undefined_var(nest1, nest2, names):
-    """TODO: make this function recursively.
-    nest1: Var1, (UndefinedVar, [1,2,3])
-    nest2: Var2, ([1,2,3,4], UndefinedVar)
-    In this case, we should not expand recursively.
-    """
-    from paddle.jit.dy2static.utils import UndefinedVar
-    from paddle.jit.dy2static.return_transformer import (
-        RETURN_VALUE_PREFIX,
-    )
-
-    def pack_undefined_var_as(seq):
-        return pack_sequence_as(
-            seq, [UndefinedVar("padding") for i in flatten(seq)]
-        )
-
-    def map_fn(n1, n2, name, order):
-        if not name.startswith(RETURN_VALUE_PREFIX) and (
-            isinstance(n1, UndefinedVar) or n1 is None
-        ):
-            if n1 is None and n2 is not None:
-                if order == 0:
-                    warnings.warn(
-                        "In cond : Var '{}' or part of it is set differently in ifelse branchs, "
-                        "<{}, {}> in true branch and <{}, {}> in false branch. Set var to "
-                        "'None' in ifelse block might lead to error.".format(
-                            name, type(n1), n1, type(n2), n2
-                        )
-                    )
-                else:
-                    warnings.warn(
-                        "In cond : Var '{}' or part of it is set differently in ifelse branchs, "
-                        "<{}, {}> in true branch and <{}, {}> in false branch. Set var to "
-                        "'None' in ifelse block might lead to error.".format(
-                            name, type(n2), n2, type(n1), n1
-                        )
-                    )
-            return pack_undefined_var_as(n2)
-        return n1
-
-    nest1_out = list(
-        map(
-            map_fn,
-            _to_sequence_except_dict(nest1),
-            _to_sequence_except_dict(nest2),
-            _to_sequence_except_dict(names),
-            [0 for i in _to_sequence_except_dict(names)],
-        )
-    )
-    nest2_out = list(
-        map(
-            map_fn,
-            _to_sequence_except_dict(nest2),
-            _to_sequence_except_dict(nest1),
-            _to_sequence_except_dict(names),
-            [1 for i in _to_sequence_except_dict(names)],
-        )
-    )
-    if not _is_sequence_except_dict(nest1):
-        nest1_out = nest1_out[0]
-    if not _is_sequence_except_dict(nest2):
-        nest2_out = nest2_out[0]
-    return nest1_out, nest2_out
-
-
 def _error_message(what, arg_name, op_name, right_value, error_value):
     error_message = (
         "{what} of '{arg_name}' in {op_name} must be "
diff --git a/python/paddle/fluid/optimizer.py b/python/paddle/fluid/optimizer.py
index d49fe0c70159d..444658ac35aa0 100755
--- a/python/paddle/fluid/optimizer.py
+++ b/python/paddle/fluid/optimizer.py
@@ -7647,7 +7647,7 @@ def true_apply_gradient():
                 )
 
         # step3. apply gradient
-        layers.cond(cond, true_fn=true_apply_gradient, false_fn=None)
+        paddle.static.nn.cond(cond, true_fn=true_apply_gradient, false_fn=None)
 
         return self._optimize_ops
 
diff --git a/python/paddle/fluid/tests/unittests/dygraph_to_static/ifelse_simple_func.py b/python/paddle/fluid/tests/unittests/dygraph_to_static/ifelse_simple_func.py
index 985d091d6b9c4..a4584e6ad5a0d 100644
--- a/python/paddle/fluid/tests/unittests/dygraph_to_static/ifelse_simple_func.py
+++ b/python/paddle/fluid/tests/unittests/dygraph_to_static/ifelse_simple_func.py
@@ -85,7 +85,7 @@ def false_fn_0(q, x, y):
         m = x + 2
         n = x + 3
         return q, x, y, z
-    q, x, y, z = fluid.layers.cond(paddle.mean(x)[0] < 5, lambda :
+    q, x, y, z = paddle.static.nn.cond(paddle.mean(x)[0] < 5, lambda :
         paddle.jit.dy2static.convert_call(true_fn_0)(q, x, y),
         lambda : paddle.jit.dy2static.convert_call(false_fn_0)(q,
         x, y))
diff --git a/python/paddle/fluid/tests/unittests/dygraph_to_static/test_dict.py b/python/paddle/fluid/tests/unittests/dygraph_to_static/test_dict.py
index 597580eedc765..3c60ef52bb87f 100644
--- a/python/paddle/fluid/tests/unittests/dygraph_to_static/test_dict.py
+++ b/python/paddle/fluid/tests/unittests/dygraph_to_static/test_dict.py
@@ -97,7 +97,7 @@ def forward(self, input, max_len=4):
             ),
         }
         # TODO(Aurelius84): The following code will be converted into:
-        # max_len = layers.cond(paddle.shape(input)[0] != max_len,
+        # max_len = paddle.static.nn.cond(paddle.shape(input)[0] != max_len,
         #                       lambda: paddle.shape(input)[0], lambda: max_len)
         # But max_len should be wrapped into tensor, which is not supported.
 
diff --git a/python/paddle/fluid/tests/unittests/dygraph_to_static/test_ifelse.py b/python/paddle/fluid/tests/unittests/dygraph_to_static/test_ifelse.py
index 8cc543a19f94d..72c648e675b3f 100644
--- a/python/paddle/fluid/tests/unittests/dygraph_to_static/test_ifelse.py
+++ b/python/paddle/fluid/tests/unittests/dygraph_to_static/test_ifelse.py
@@ -154,7 +154,7 @@ def map_func(func, tensor_list):
 
     def body(i, ten, y):
         # It will be converted into `layers.cond` as followed.
-        # map_func(lambda x: fluid.layers.cond(i==0, lambda: x, lambda: add_fn(x), y)
+        # map_func(lambda x: paddle.static.nn.cond(i==0, lambda: x, lambda: add_fn(x), y)
         y = map_func(lambda x: x if (i == 0) is not None else add_fn(x), y)
         i += 1
         return [i, ten, y]
@@ -183,7 +183,7 @@ def map_func(func, tensor_list):
 
     i = fluid.layers.fill_constant(shape=[1], dtype='int64', value=0)
     # It will be converted into `layers.cond` as followed.
-    # map_func(lambda x: fluid.layers.cond(i==1, lambda: x, lambda: add_fn(x), y)
+    # map_func(lambda x: paddle.static.nn.cond(i==1, lambda: x, lambda: add_fn(x), y)
     # `if (Tensor) == 1` is supported in dygraph.
     y = map_func(lambda x: x if i == 1 else add_fn(x), y)
     return y[0]
diff --git a/python/paddle/fluid/tests/unittests/dygraph_to_static/test_warning.py b/python/paddle/fluid/tests/unittests/dygraph_to_static/test_warning.py
index e74e06f8f9b68..3ba8caea85099 100644
--- a/python/paddle/fluid/tests/unittests/dygraph_to_static/test_warning.py
+++ b/python/paddle/fluid/tests/unittests/dygraph_to_static/test_warning.py
@@ -16,7 +16,7 @@
 import warnings
 
 import paddle
-from paddle.fluid.layers.control_flow import cond
+from paddle.static.nn import cond
 
 
 @paddle.jit.to_static
diff --git a/python/paddle/fluid/tests/unittests/ir/test_ir_subgraph_python_interface.py b/python/paddle/fluid/tests/unittests/ir/test_ir_subgraph_python_interface.py
index 2a7ebae071073..33efd9b7c54c2 100644
--- a/python/paddle/fluid/tests/unittests/ir/test_ir_subgraph_python_interface.py
+++ b/python/paddle/fluid/tests/unittests/ir/test_ir_subgraph_python_interface.py
@@ -52,7 +52,7 @@ def false_func():
             x = layers.fill_constant(shape=[1], dtype='float32', value=0.1)
             y = layers.fill_constant(shape=[1], dtype='float32', value=0.23)
             pred = paddle.less_than(y, x)
-            out = layers.cond(pred, true_func, false_func)
+            out = paddle.static.nn.cond(pred, true_func, false_func)
 
         core_graph = core.Graph(main_program.desc)
         # We should create graph for test, otherwise it will throw a
diff --git a/python/paddle/fluid/tests/unittests/standalone_executor/test_standalone_controlflow.py b/python/paddle/fluid/tests/unittests/standalone_executor/test_standalone_controlflow.py
index c989ff866e8c0..000f1b7d4b856 100644
--- a/python/paddle/fluid/tests/unittests/standalone_executor/test_standalone_controlflow.py
+++ b/python/paddle/fluid/tests/unittests/standalone_executor/test_standalone_controlflow.py
@@ -57,7 +57,7 @@ def false_func():
             x = layers.fill_constant(shape=[1], dtype='float32', value=0.1)
             y = layers.fill_constant(shape=[1], dtype='float32', value=0.23)
             pred = paddle.less_than(x, y)
-            out = layers.cond(pred, true_func, false_func)
+            out = paddle.static.nn.cond(pred, true_func, false_func)
             # out is a tuple containing 2 tensors
             return main_program, startup_program, out
 
diff --git a/python/paddle/fluid/tests/unittests/test_cond.py b/python/paddle/fluid/tests/unittests/test_cond.py
index bc5a73d048dfa..c96cece765b24 100644
--- a/python/paddle/fluid/tests/unittests/test_cond.py
+++ b/python/paddle/fluid/tests/unittests/test_cond.py
@@ -54,7 +54,7 @@ def false_func():
             x = layers.fill_constant(shape=[1], dtype='float32', value=0.1)
             y = layers.fill_constant(shape=[1], dtype='float32', value=0.23)
             pred = paddle.less_than(y, x)
-            out = layers.cond(pred, true_func, false_func)
+            out = paddle.static.nn.cond(pred, true_func, false_func)
             # out is one tensor
 
         place = (
@@ -94,7 +94,7 @@ def false_func():
         startup_program = Program()
         with program_guard(main_program, startup_program):
             pred = layers.fill_constant(shape=[1], dtype='bool', value=True)
-            out = layers.cond(pred, true_func, false_func)
+            out = paddle.static.nn.cond(pred, true_func, false_func)
             # out is a tuple containing 2 tensors
 
         place = (
@@ -138,7 +138,7 @@ def false_func(a, i):
             a = layers.fill_constant(shape=[3, 2, 1], dtype='int32', value=7)
             i = fluid.data(name="i", shape=[1], dtype='int32')
             pred = (i % 2) == 0
-            a = layers.cond(
+            a = paddle.static.nn.cond(
                 pred, lambda: true_func(a, i), lambda: false_func(a, i)
             )
         place = (
@@ -183,9 +183,9 @@ def false_func():
         with program_guard(main_program, startup_program):
             i = fluid.data(name="i", shape=[1], dtype='int32')
             pred = (i % 2) == 0
-            out1 = layers.cond(pred, true_func, false_func)
-            out2 = layers.cond(pred, None, false_func)
-            out3 = layers.cond(pred, true_func, None)
+            out1 = paddle.static.nn.cond(pred, true_func, false_func)
+            out2 = paddle.static.nn.cond(pred, None, false_func)
+            out3 = paddle.static.nn.cond(pred, true_func, None)
         place = (
             fluid.CUDAPlace(0)
             if core.is_compiled_with_cuda()
@@ -223,13 +223,15 @@ def func_return_two_tensors():
             i = fluid.data(name="i", shape=[1], dtype='int32')
             pred = (i % 2) == 0
             with self.assertRaises(TypeError):
-                out = layers.cond(pred, i, func_return_one_tensor)
+                out = paddle.static.nn.cond(pred, i, func_return_one_tensor)
 
             with self.assertRaises(TypeError):
-                out = layers.cond(pred, func_return_one_tensor, np.asarray([3]))
+                out = paddle.static.nn.cond(
+                    pred, func_return_one_tensor, np.asarray([3])
+                )
 
             with self.assertRaises(Exception) as e:
-                out = layers.cond(
+                out = paddle.static.nn.cond(
                     pred, func_return_none, func_return_one_tensor
                 )
             self.assertTrue(
@@ -238,7 +240,7 @@ def func_return_two_tensors():
             )
 
             with self.assertRaises(Exception) as e:
-                out = layers.cond(
+                out = paddle.static.nn.cond(
                     pred, func_return_two_tensors, func_return_none
                 )
             self.assertTrue(
@@ -247,7 +249,7 @@ def func_return_two_tensors():
             )
 
             with self.assertRaises(Exception) as e:
-                out = layers.cond(
+                out = paddle.static.nn.cond(
                     pred, func_return_one_tensor, func_return_two_tensors
                 )
             self.assertTrue(
@@ -268,7 +270,7 @@ def test_extremely_simple_net_with_op_in_condition(self):
                 shape=[1], dtype='float32', value=1.25
             )
             b.stop_gradient = False
-            out = layers.cond(a - b < -1.0, lambda: a, lambda: b)
+            out = paddle.static.nn.cond(a - b < -1.0, lambda: a, lambda: b)
         append_backward(out)
 
         place = (
@@ -308,14 +310,14 @@ def test_cond_inside_cond(self):
         paddle.enable_static()
 
         def less_than_branch(i, a):
-            return layers.cond(
+            return paddle.static.nn.cond(
                 i >= 3.0,
                 lambda: paddle.add(a, a),
                 lambda: paddle.subtract(a, a),
             )
 
         def greater_equal_branch(i, a):
-            return layers.cond(
+            return paddle.static.nn.cond(
                 i < 8.0,
                 lambda: paddle.multiply(a, a),
                 lambda: paddle.divide(a, a),
@@ -326,7 +328,7 @@ def greater_equal_branch(i, a):
         with program_guard(main_program, startup_program):
             i = fluid.data(name="i", shape=[1], dtype='float32')
             a = 2.0 * i
-            out = layers.cond(
+            out = paddle.static.nn.cond(
                 i < 5.0,
                 lambda: less_than_branch(i, a),
                 lambda: greater_equal_branch(i, a),
@@ -370,14 +372,14 @@ def test_cond_op_in_condition(self):
                 shape=[1], dtype='float32', value=1.24
             )
             b.stop_gradient = False
-            out = fluid.layers.cond(
+            out = fluid.paddle.static.nn.cond(
                 a < b,
-                lambda: fluid.layers.cond(
+                lambda: fluid.paddle.static.nn.cond(
                     a - b < -1.0,
                     lambda: paddle.add(a, b),
                     lambda: paddle.multiply(a, b),
                 ),
-                lambda: fluid.layers.cond(
+                lambda: fluid.paddle.static.nn.cond(
                     a == b,
                     lambda: paddle.subtract(a, b),
                     lambda: paddle.pow(a, b),
@@ -550,7 +552,7 @@ def test_cond_backward(self):
 
         def cond_func(i, img, label):
             predicate = (i % 2) == 0
-            return layers.cond(
+            return paddle.static.nn.cond(
                 predicate,
                 lambda: simple_fc_net_with_inputs(img, label, class_num=10),
                 lambda: batchnorm_fc_with_inputs(img, label, class_num=10),
@@ -574,19 +576,19 @@ def test_half_nested_cond_backward(self):
         paddle.enable_static()
 
         def branch(i, img, label):
-            return layers.cond(
+            return paddle.static.nn.cond(
                 (i % 2) == 0,
                 lambda: simple_fc_net_with_inputs(img, label, class_num=10),
                 lambda: batchnorm_fc_with_inputs(img, label, class_num=10),
             )
 
         def cond_func_simple_net_at_true(i, img, label):
-            return layers.cond(
+            return paddle.static.nn.cond(
                 i < 5, lambda: branch(i, img, label), lambda: paddle.mean(img)
             )
 
         def cond_func_simple_net_at_false(i, img, label):
-            return layers.cond(
+            return paddle.static.nn.cond(
                 i < 5, lambda: paddle.mean(img), lambda: branch(i, img, label)
             )
 
@@ -626,14 +628,14 @@ def branch(i, img, label, mod_two):
                 predicate = (i % 2) == 0
             else:
                 predicate = (i % 2) != 0
-            return layers.cond(
+            return paddle.static.nn.cond(
                 predicate,
                 lambda: simple_fc_net_with_inputs(img, label, class_num=10),
                 lambda: batchnorm_fc_with_inputs(img, label, class_num=10),
             )
 
         def cond_func(i, img, label):
-            return layers.cond(
+            return paddle.static.nn.cond(
                 i < 5,
                 lambda: branch(i, img, label, True),
                 lambda: branch(i, img, label, False),
@@ -665,16 +667,16 @@ def func():
                 return pred
 
             with self.assertRaises(TypeError):
-                layers.cond(None, func, func)
+                paddle.static.nn.cond(None, func, func)
 
             with self.assertRaises(TypeError):
-                layers.cond(pred, func, set())
+                paddle.static.nn.cond(pred, func, set())
 
             with self.assertRaises(TypeError):
-                layers.cond(pred, set(), func)
+                paddle.static.nn.cond(pred, set(), func)
 
             with self.assertRaises(TypeError):
-                layers.cond(pred, func, func, set())
+                paddle.static.nn.cond(pred, func, func, set())
 
 
 class TestCondWithDict(unittest.TestCase):
diff --git a/python/paddle/fluid/tests/unittests/test_desc_clone.py b/python/paddle/fluid/tests/unittests/test_desc_clone.py
index 477910f53d59d..0e8ee6b90a288 100644
--- a/python/paddle/fluid/tests/unittests/test_desc_clone.py
+++ b/python/paddle/fluid/tests/unittests/test_desc_clone.py
@@ -224,7 +224,7 @@ def false_fn():
                 hidden2 = fluid.layers.dropout(hidden1, dropout_prob=0.6)
                 return hidden2
 
-            hidden2 = fluid.layers.cond(cond, true_fn, false_fn)
+            hidden2 = paddle.static.nn.cond(cond, true_fn, false_fn)
 
             loss = fluid.layers.cross_entropy(
                 input=fluid.layers.fc(hidden2, size=10, act='softmax'),
@@ -265,7 +265,7 @@ def false_fn():
                 hidden2 = fluid.layers.dropout(hidden1, dropout_prob=0.6)
                 return hidden2
 
-            hidden2 = fluid.layers.cond(cond, true_fn, false_fn)
+            hidden2 = paddle.static.nn.cond(cond, true_fn, false_fn)
             loss = fluid.layers.cross_entropy(
                 input=fluid.layers.fc(hidden2, size=10, act='softmax'),
                 label=fluid.layers.data(name='label', shape=[1], dtype='int64'),
diff --git a/python/paddle/fluid/tests/unittests/test_layers.py b/python/paddle/fluid/tests/unittests/test_layers.py
index dcf442200d178..2bdd871013a5a 100644
--- a/python/paddle/fluid/tests/unittests/test_layers.py
+++ b/python/paddle/fluid/tests/unittests/test_layers.py
@@ -2473,7 +2473,7 @@ def greater_equal_branch(a, b):
             b = fluid.layers.fill_constant(
                 shape=[1], dtype='float32', value=0.23
             )
-            out = fluid.layers.cond(
+            out = fluid.paddle.static.nn.cond(
                 a >= b,
                 lambda: greater_equal_branch(a, b),
                 lambda: less_than_branch(a, b),
@@ -2493,12 +2493,12 @@ def greater_equal_branch(a, b):
                 b = fluid.dygraph.to_variable(
                     np.array([0.23]).astype('float32')
                 )
-                out = layers.cond(
+                out = paddle.static.nn.cond(
                     a < b,
                     lambda: less_than_branch(a, b),
                     lambda: greater_equal_branch(a, b),
                 )
-                out2 = layers.cond(
+                out2 = paddle.static.nn.cond(
                     a >= b,
                     lambda: greater_equal_branch(a, b),
                     lambda: less_than_branch(a, b),
@@ -2509,18 +2509,18 @@ def greater_equal_branch(a, b):
                     eager_dynamic_res, eager_dynamic_res2
                 )
                 with self.assertRaises(TypeError):
-                    layers.cond(a < b, 'str', 'str')
+                    paddle.static.nn.cond(a < b, 'str', 'str')
                 with self.assertRaises(TypeError):
-                    layers.cond(a >= b, 'str', 'str')
+                    paddle.static.nn.cond(a >= b, 'str', 'str')
 
             a = fluid.dygraph.to_variable(np.array([0.1]).astype('float32'))
             b = fluid.dygraph.to_variable(np.array([0.23]).astype('float32'))
-            out = layers.cond(
+            out = paddle.static.nn.cond(
                 a < b,
                 lambda: less_than_branch(a, b),
                 lambda: greater_equal_branch(a, b),
             )
-            out2 = layers.cond(
+            out2 = paddle.static.nn.cond(
                 a >= b,
                 lambda: greater_equal_branch(a, b),
                 lambda: less_than_branch(a, b),
@@ -2529,9 +2529,9 @@ def greater_equal_branch(a, b):
             dynamic_res2 = out2.numpy()
             np.testing.assert_array_equal(dynamic_res, dynamic_res2)
             with self.assertRaises(TypeError):
-                layers.cond(a < b, 'str', 'str')
+                paddle.static.nn.cond(a < b, 'str', 'str')
             with self.assertRaises(TypeError):
-                layers.cond(a >= b, 'str', 'str')
+                paddle.static.nn.cond(a >= b, 'str', 'str')
 
         np.testing.assert_array_equal(static_res, dynamic_res)
         np.testing.assert_array_equal(static_res, eager_dynamic_res)
diff --git a/python/paddle/fluid/tests/unittests/test_math_op_patch.py b/python/paddle/fluid/tests/unittests/test_math_op_patch.py
index a582b4951b117..298b6d4ed0af7 100644
--- a/python/paddle/fluid/tests/unittests/test_math_op_patch.py
+++ b/python/paddle/fluid/tests/unittests/test_math_op_patch.py
@@ -237,7 +237,7 @@ def test_equal_and_cond(self):
         one = paddle.ones(shape=[1], dtype='int32')
         zero = fluid.layers.zeros(shape=[1], dtype='int32')
         cond = one == zero
-        c = fluid.layers.cond(cond, lambda: a + b, lambda: a - b)
+        c = paddle.static.nn.cond(cond, lambda: a + b, lambda: a - b)
 
         place = fluid.CPUPlace()
         exe = fluid.Executor(place)
diff --git a/python/paddle/fluid/tests/unittests/test_optimizer_grad.py b/python/paddle/fluid/tests/unittests/test_optimizer_grad.py
index e20d563ebdc03..99c4d79bb3168 100644
--- a/python/paddle/fluid/tests/unittests/test_optimizer_grad.py
+++ b/python/paddle/fluid/tests/unittests/test_optimizer_grad.py
@@ -115,7 +115,7 @@ def cond_false():
             return cond_res
 
         cond_i = fluid.layers.assign(np.array([cond_i], dtype='float32'))
-        sum_cond = fluid.layers.cond(cond_i > 1.0, cond_true, cond_false)
+        sum_cond = paddle.static.nn.cond(cond_i > 1.0, cond_true, cond_false)
         sum_all = paddle.add_n([sum_xy, sub_yz, sum_cond])
         mean_out = paddle.mean(sum_all)
         if use_bf16:
diff --git a/python/paddle/fluid/tests/unittests/test_program_code.py b/python/paddle/fluid/tests/unittests/test_program_code.py
index e60706794f5b1..3ecd2619c15fc 100644
--- a/python/paddle/fluid/tests/unittests/test_program_code.py
+++ b/python/paddle/fluid/tests/unittests/test_program_code.py
@@ -46,7 +46,7 @@ def false_func():
             x = layers.fill_constant(shape=[1], dtype='float32', value=0.1)
             y = layers.fill_constant(shape=[1], dtype='float32', value=0.23)
             pred = paddle.less_than(y, x)
-            out = layers.cond(pred, true_func, false_func)
+            out = paddle.static.nn.cond(pred, true_func, false_func)
 
     def test_program_code(self):
         self.var._to_readable_code()
diff --git a/python/paddle/fluid/variable_index.py b/python/paddle/fluid/variable_index.py
index 31d587269db71..6081f3d9a3e57 100644
--- a/python/paddle/fluid/variable_index.py
+++ b/python/paddle/fluid/variable_index.py
@@ -341,7 +341,7 @@ def idx_empty(var):
         var_shape[0] = 0
         return paddle.empty(var_shape, dtype=var.dtype)
 
-    from .layers.control_flow import cond
+    from paddle.static.nn import cond
 
     return cond(
         item.any(), lambda: idx_not_empty(var, item), lambda: idx_empty(var)
@@ -874,7 +874,7 @@ def idx_not_empty(var, item, value):
         out = scatter_nd_add(var, idx, gather_val_new)
         var[:] = out
 
-    from .layers.control_flow import cond
+    from paddle.static.nn import cond
 
     # If all the bool index is False, just do nothing
     cond(item.any(), lambda: idx_not_empty(var, item, value))
diff --git a/python/paddle/static/nn/__init__.py b/python/paddle/static/nn/__init__.py
index b005c6f556097..772ffdcc61454 100755
--- a/python/paddle/static/nn/__init__.py
+++ b/python/paddle/static/nn/__init__.py
@@ -22,18 +22,19 @@
 from .common import conv3d_transpose  # noqa: F401
 from .common import bilinear_tensor_product  # noqa: F401
 from .common import py_func  # noqa: F401
+from .common import prelu  # noqa: F401
+from .control_flow import Assert  # noqa: F401
+from .control_flow import increment  # noqa: F401
+from .control_flow import cond  # noqa: F401
+from .loss import nce  # noqa: F401
 
 from ...tensor.creation import create_parameter  # noqa: F401
 from ...fluid.layers import batch_norm  # noqa: F401
 from ...fluid.layers import case  # noqa: F401
-from ...fluid.layers import cond  # noqa: F401
 from ...fluid.layers import conv2d  # noqa: F401
 from ...fluid.layers import crf_decoding  # noqa: F401
 from ...fluid.layers import layer_norm  # noqa: F401
 from ...fluid.layers import multi_box_head  # noqa: F401
-from .loss import nce  # noqa: F401
-from .common import prelu  # noqa: F401
-from .control_flow import Assert, increment  # noqa: F401
 from ...fluid.layers import row_conv  # noqa: F401
 from ...fluid.layers import spectral_norm  # noqa: F401
 from ...fluid.layers import switch_case  # noqa: F401
diff --git a/python/paddle/static/nn/control_flow.py b/python/paddle/static/nn/control_flow.py
index edf90babea525..0753d9a504e1a 100644
--- a/python/paddle/static/nn/control_flow.py
+++ b/python/paddle/static/nn/control_flow.py
@@ -13,14 +13,33 @@
 # limitations under the License.
 
 
+import warnings
+from functools import partial
+
+import paddle
 from paddle import _C_ops
 from paddle.common_ops_import import LayerHelper
+from paddle.fluid.control_flow import (
+    ConditionalBlock,
+    _to_sequence_except_dict,
+    select_input,
+    support_ret_buildin_type,
+)
 from paddle.fluid.data_feeder import check_type, check_variable_and_dtype
-from paddle.framework import in_dygraph_mode
+from paddle.fluid.tensor import assign
+from paddle.fluid.utils import (
+    assert_same_structure,
+    flatten,
+    is_sequence,
+    map_structure,
+    pack_sequence_as,
+)
+from paddle.framework import Variable, _non_static_mode, core, in_dygraph_mode
 
 __all__ = [
     'Assert',
     'increment',
+    'cond',
 ]
 
 
@@ -141,3 +160,460 @@ def increment(x, value=1.0, in_place=True):
         attrs={'step': float(value)},
     )
     return out
+
+
+def copy_var_to_parent_block(var, layer_helper):
+    if not isinstance(var, Variable):
+        return var
+    prog = layer_helper.main_program
+    parent_idx = prog.current_block().parent_idx
+    assert (
+        parent_idx >= 0
+    ), "Got wrong parent block index when assigning var to parent scope in control_flow"
+    parent_block = prog.block(parent_idx)
+
+    if (
+        var.type == core.VarDesc.VarType.LOD_TENSOR_ARRAY
+        and parent_block._find_var_recursive(var.name)
+    ):
+        parent_block_var = var
+    else:
+        parent_block_var = parent_block.create_var(
+            dtype=var.dtype, shape=var.shape, type=var.type
+        )
+        assign(var, parent_block_var)
+    return parent_block_var
+
+
+def expand_undefined_var(nest1, nest2, names):
+    """TODO: make this function recursively.
+    nest1: Var1, (UndefinedVar, [1,2,3])
+    nest2: Var2, ([1,2,3,4], UndefinedVar)
+    In this case, we should not expand recursively.
+    """
+    from paddle.jit.dy2static.return_transformer import RETURN_VALUE_PREFIX
+    from paddle.jit.dy2static.utils import UndefinedVar
+
+    def pack_undefined_var_as(seq):
+        return pack_sequence_as(
+            seq, [UndefinedVar("padding") for i in flatten(seq)]
+        )
+
+    def map_fn(n1, n2, name, order):
+        if not name.startswith(RETURN_VALUE_PREFIX) and (
+            isinstance(n1, UndefinedVar) or n1 is None
+        ):
+            if n1 is None and n2 is not None:
+                if order == 0:
+                    warnings.warn(
+                        "In cond : Var '{}' or part of it is set differently in ifelse branchs, "
+                        "<{}, {}> in true branch and <{}, {}> in false branch. Set var to "
+                        "'None' in ifelse block might lead to error.".format(
+                            name, type(n1), n1, type(n2), n2
+                        )
+                    )
+                else:
+                    warnings.warn(
+                        "In cond : Var '{}' or part of it is set differently in ifelse branchs, "
+                        "<{}, {}> in true branch and <{}, {}> in false branch. Set var to "
+                        "'None' in ifelse block might lead to error.".format(
+                            name, type(n2), n2, type(n1), n1
+                        )
+                    )
+            return pack_undefined_var_as(n2)
+        return n1
+
+    nest1_out = list(
+        map(
+            map_fn,
+            _to_sequence_except_dict(nest1),
+            _to_sequence_except_dict(nest2),
+            _to_sequence_except_dict(names),
+            [0 for i in _to_sequence_except_dict(names)],
+        )
+    )
+    nest2_out = list(
+        map(
+            map_fn,
+            _to_sequence_except_dict(nest2),
+            _to_sequence_except_dict(nest1),
+            _to_sequence_except_dict(names),
+            [1 for i in _to_sequence_except_dict(names)],
+        )
+    )
+
+    def _is_sequence_except_dict(x):
+        """
+        In this function, dict is not viewed as sequence.
+        """
+        if isinstance(x, dict):
+            return False
+        return is_sequence(x)
+
+    if not _is_sequence_except_dict(nest1):
+        nest1_out = nest1_out[0]
+    if not _is_sequence_except_dict(nest2):
+        nest2_out = nest2_out[0]
+    return nest1_out, nest2_out
+
+
+def change_none_to_undefinedvar(nest1, nest2):
+    from paddle.jit.dy2static.utils import UndefinedVar
+
+    def map_fn(x):
+        if x is None:
+            return UndefinedVar("padding")
+        return x
+
+    nest1_out = pack_sequence_as(nest1, list(map(map_fn, flatten(nest1))))
+    nest2_out = pack_sequence_as(nest2, list(map(map_fn, flatten(nest2))))
+    return nest1_out, nest2_out
+
+
+def select_input_with_buildin_type(inputs, mask, name):
+    from paddle.jit.dy2static.utils import UndefinedVar
+    from paddle.jit.dy2static.variable_trans_func import to_static_variable
+
+    false_var, true_var = inputs
+
+    if isinstance(false_var, UndefinedVar) and isinstance(
+        true_var, UndefinedVar
+    ):
+        """None -> UndefinedVar, so the real value is a [None, UndefinedVar] or [None, None], we just return None."""
+        return None
+
+    if isinstance(false_var, Variable) and isinstance(true_var, Variable):
+        try:
+            return select_input(inputs, mask)
+        except Exception as e:
+            raise RuntimeError(
+                f"Exceptions throwed while doing select_input on {name}:\n{e}"
+            )
+
+    elif isinstance(false_var, support_ret_buildin_type) and isinstance(
+        false_var, type(true_var)
+    ):
+        if false_var == true_var:
+            return false_var
+        else:
+            inputs = [
+                to_static_variable(false_var),
+                to_static_variable(true_var),
+            ]
+    # Deal with the situations like this: false_var is int and true_var is Variable
+    elif (
+        isinstance(false_var, support_ret_buildin_type)
+        and isinstance(true_var, Variable)
+    ) or (
+        isinstance(true_var, support_ret_buildin_type)
+        and isinstance(false_var, Variable)
+    ):
+        inputs = [to_static_variable(false_var), to_static_variable(true_var)]
+        warnings.warn(
+            "Return results from different branches in cond are not same type: "
+            "false_var returned by false_fn is '{}' and true_var of true_fn is "
+            "'{}'".format(type(false_var), type(true_var))
+        )
+    elif (
+        isinstance(false_var, UndefinedVar)
+        and isinstance(true_var, (Variable,) + support_ret_buildin_type)
+    ) or (
+        isinstance(true_var, UndefinedVar)
+        and isinstance(false_var, (Variable,) + support_ret_buildin_type)
+    ):
+
+        def create_var_if_not_undefined_var(a):
+            if isinstance(a, UndefinedVar):
+                return a
+            return to_static_variable(a)
+
+        true_var, false_var = to_static_variable(true_var), to_static_variable(
+            false_var
+        )
+        inputs = [false_var, true_var]
+    else:
+        raise TypeError(
+            "Unsupported return type of true_fn and false_fn in cond: false_var "
+            "returned by false_fn is '{}' and true_var of true_fn is '{}'".format(
+                type(false_var), type(true_var)
+            )
+        )
+    try:
+        return select_input(inputs, mask)
+    except Exception as e:
+        raise RuntimeError(
+            f"Exceptions throwed while doing select_input on {name}:\n{e}"
+        )
+
+
+def cond(pred, true_fn=None, false_fn=None, name=None, return_names=None):
+    """
+    This API returns ``true_fn()`` if the predicate ``pred`` is true else
+    ``false_fn()`` . Users could also set ``true_fn`` or ``false_fn`` to
+    ``None`` if do nothing and this API will treat the callable simply returns
+    ``None`` in this case.
+
+    ``true_fn`` and ``false_fn`` should return same nest structure of tensors
+    or both return ``None`` if user doens't like to return anything. A nest
+    structure of tensors in PaddlePaddle is tensor(s), or tuple of tensors, or
+    list of tensors.
+
+    Note:
+        1. The tuples or lists returned by ``true_fn`` and ``false_fn`` must have
+        the same shape because of dataflow model of PaddlePaddle while the
+        tensors in the tuples or the lists can have different shapes.
+
+        2. This API could be used under both static mode or dygraph mode. If it
+        is in dygraph mode, the API only runs one branch based on condition.
+
+        3. If it is in static mode, any tensors or operations created outside
+        or inside of ``true_fn`` and ``false_fn`` will be in net building
+        regardless of which branch is selected at runtime. This has frequently
+        surprised users who expected a lazy semantics. For example:
+
+        .. code-block:: python
+
+            import paddle
+
+            a = paddle.zeros((1, 1))
+            b = paddle.zeros((1, 1))
+            c = a * b
+            out = paddle.static.nn.cond(a < b, lambda: a + c, lambda: b * b)
+
+        No matter whether ``a < b`` , ``c = a * b`` will be in net building and
+        run. ``a + c`` and ``b * b`` will be in net building, but only one
+        branch will be executed during runtime.
+
+    Args:
+        pred(Tensor): A boolean tensor whose numel should be 1. The boolean
+            value determines whether to return the result of ``true_fn`` or
+            ``false_fn`` .
+        true_fn(callable, optional): A callable to be performed if ``pred`` is
+            true. The default value is ``None`` .
+        false_fn(callable, optional): A callable to be performed if ``pred`` is
+            false. The default value is ``None`` .
+        name(str, optional): The default value is ``None`` . Normally users
+             don't have to set this parameter. For more information, please
+             refer to :ref:`api_guide_Name` .
+        return_names(sequence of string, optional): The default value is ``None`` .
+             Normally users don't have to set this parameters.  A sequence of strings
+             to represents the name of returned vars.  The structure of sequence must
+             be same with return values of true_fn and false_fn.
+
+    Returns:
+        Tensor|list(Tensor)|tuple(Tensor): returns ``true_fn()`` if the
+        predicate ``pred`` is true else ``false_fn()`` .
+
+    Raises:
+        TypeError: if ``true_fn`` or ``false_fn`` is not callable.
+        ValueError: if ``true_fn`` and ``false_fn`` don't return the same nest
+            structure of tensors.
+
+    Examples:
+        .. code-block:: python
+
+            import paddle
+
+            #
+            # pseudocode:
+            # if 0.1 < 0.23:
+            #     return 1, True
+            # else:
+            #     return 3, 2
+            #
+
+            def true_func():
+                return paddle.full(shape=[1, 2], dtype='int32',
+                                   fill_value=1), paddle.full(shape=[2, 3],
+                                                              dtype='bool',
+                                                              fill_value=True)
+
+
+            def false_func():
+                return paddle.full(shape=[3, 4], dtype='float32',
+                                   fill_value=3), paddle.full(shape=[4, 5],
+                                                              dtype='int64',
+                                                              fill_value=2)
+
+
+            x = paddle.full(shape=[1], dtype='float32', fill_value=0.1)
+            y = paddle.full(shape=[1], dtype='float32', fill_value=0.23)
+            pred = paddle.less_than(x=x, y=y, name=None)
+            ret = paddle.static.nn.cond(pred, true_func, false_func)
+            # ret is a tuple containing 2 tensors
+            # ret[0] = [[1 1]]
+            # ret[1] = [[ True  True  True]
+            #           [ True  True  True]]
+
+    """
+    if _non_static_mode():
+        assert isinstance(pred, Variable), "The pred in cond must be Variable"
+        assert pred.size == 1, "condition input's numel should be 1"
+        pred = pred.numpy()[0]
+        if pred:
+            if true_fn is not None:
+                if not callable(true_fn):
+                    raise TypeError(
+                        "The true_fn in cond must be callable, but received {}".format(
+                            type(true_fn).__name__
+                        )
+                    )
+                return true_fn()
+        else:
+            if false_fn is not None:
+                if not callable(false_fn):
+                    raise TypeError(
+                        "The false_fn in cond must be callable, but received {}".format(
+                            type(false_fn).__name__
+                        )
+                    )
+                return false_fn()
+        return None
+
+    check_variable_and_dtype(pred, "pred", ['bool'], "fluid.layers.cond")
+    check_type(name, "name", (str, type(None)), "fluid.layers.cond")
+    helper = LayerHelper('cond', **locals())
+    true_output = None
+    false_output = None
+    copy_to_parent_func = lambda var: copy_var_to_parent_block(var, helper)
+    if true_fn is not None:
+        if not callable(true_fn):
+            raise TypeError(
+                "The true_fn in cond must be callable, but received {}".format(
+                    type(true_fn).__name__
+                )
+            )
+        true_cond_block = ConditionalBlock([pred], is_scalar_condition=True)
+        with true_cond_block.block():
+            origin_true_output = true_fn()
+            if origin_true_output is not None:
+                true_output = map_structure(
+                    copy_to_parent_func, origin_true_output
+                )
+    if false_fn is not None:
+        if not callable(false_fn):
+            raise TypeError(
+                "The false_fn in cond must be callable, but received {}".format(
+                    type(false_fn).__name__
+                )
+            )
+        false_cond_block = ConditionalBlock(
+            [paddle.logical_not(pred)], is_scalar_condition=True
+        )
+        with false_cond_block.block():
+            origin_false_output = false_fn()
+            if origin_false_output is not None:
+                false_output = map_structure(
+                    copy_to_parent_func, origin_false_output
+                )
+
+    if true_output is None and false_output is None:
+        return None
+
+    if true_output is None:
+        raise ValueError(
+            "Incompatible return values of true_fn and false_fn in cond: "
+            "true_fn returns None while false_fn returns non-None"
+        )
+    if false_output is None:
+        raise ValueError(
+            "Incompatible return values of true_fn and false_fn in cond: "
+            "true_fn returns non-None while false_fn returns None"
+        )
+
+    # Merge true and false output if they are not None
+    if return_names is None:
+        is_dy2staic = False
+        return_names = ["no name"] * len(_to_sequence_except_dict(true_output))
+    else:
+        """
+        dy2static will set the return_names and expand the return values to UndefinedVar.
+        """
+        is_dy2staic = True
+
+        # TODO:  expand_undefined_var will replace None to Undefinedvar(), to fix cases like:
+        #       a = None
+        #       if condition:
+        #           a = 1
+        # Because we can not use variable to express 'None'
+        true_output, false_output = expand_undefined_var(
+            true_output, false_output, return_names
+        )
+
+    if len(_to_sequence_except_dict(true_output)) != len(
+        _to_sequence_except_dict(false_output)
+    ):
+        raise ValueError(
+            "true fn returns {} vars, but false fn returns {} vars, which is not equals".format(
+                len(_to_sequence_except_dict(true_output)),
+                len(_to_sequence_except_dict(false_output)),
+            )
+        )
+    for true_out, false_out, return_name in zip(
+        _to_sequence_except_dict(true_output),
+        _to_sequence_except_dict(false_output),
+        _to_sequence_except_dict(return_names),
+    ):
+        try:
+            assert_same_structure(true_out, false_out, check_types=False)
+        except ValueError as e:
+            raise ValueError(
+                "Incompatible return values of `{}` in true_fn and false_fn in cond: {}".format(
+                    return_name, e
+                )
+            )
+
+    def check_ret_none(seq_true, seq_false, seq_names):
+        for f_true, f_false, f_name in zip(seq_true, seq_false, seq_names):
+            f_true = flatten(f_true)
+            f_false = flatten(f_false)
+            for idx in range(len(f_true)):
+                if (
+                    f_true[idx] is None
+                    and f_false[idx] is not None
+                    or f_false[idx] is None
+                    and f_true[idx] is not None
+                ):
+                    warnings.warn(
+                        "In cond : Var '{}' or part of it is set differently in ifelse branchs, "
+                        "<{}, {}> in true branch and <{}, {}> in false branch. Set var to "
+                        "'None' in ifelse block might lead to error.".format(
+                            f_name,
+                            type(f_true[idx]),
+                            f_true[idx],
+                            type(f_false[idx]),
+                            f_false[idx],
+                        )
+                    )
+
+    check_ret_none(
+        _to_sequence_except_dict(true_output),
+        _to_sequence_except_dict(false_output),
+        _to_sequence_except_dict(return_names),
+    )
+
+    if is_dy2staic:
+        true_output, false_output = change_none_to_undefinedvar(
+            true_output, false_output
+        )
+
+    mask = paddle.cast(pred, dtype='int32')
+    merge_func = (
+        lambda name, false_var, true_var: select_input_with_buildin_type(
+            [false_var, true_var], mask, name
+        )
+    )
+
+    def merge_every_var_list(false_vars, true_vars, name):
+        return map_structure(partial(merge_func, name), false_vars, true_vars)
+
+    merged_output = list(
+        map(
+            merge_every_var_list,
+            _to_sequence_except_dict(false_output),
+            _to_sequence_except_dict(true_output),
+            _to_sequence_except_dict(return_names),
+        )
+    )
+    merged_output = pack_sequence_as(false_output, flatten(merged_output))
+    return merged_output

From 85548f8476e73a0caf9a1ccae1ba28a25fde4230 Mon Sep 17 00:00:00 2001
From: wuzhanfei <wuzhanfei@baidu.com>
Date: Wed, 7 Dec 2022 06:33:40 +0000
Subject: [PATCH 08/16] fix example codes

---
 python/paddle/fluid/layers/control_flow.py | 1 +
 python/paddle/static/nn/control_flow.py    | 7 ++++---
 2 files changed, 5 insertions(+), 3 deletions(-)

diff --git a/python/paddle/fluid/layers/control_flow.py b/python/paddle/fluid/layers/control_flow.py
index ef448f357b87a..2d3d8ea5d3407 100755
--- a/python/paddle/fluid/layers/control_flow.py
+++ b/python/paddle/fluid/layers/control_flow.py
@@ -1159,6 +1159,7 @@ class While:
             import paddle.fluid as fluid
             import numpy as np
 
+            paddle.enable_static()
             i = fluid.layers.fill_constant(shape=[1], dtype='int64', value=0)
             loop_len = fluid.layers.fill_constant(shape=[1], dtype='int64', value=10)
             one = fluid.layers.fill_constant(shape=[1], dtype='float32', value=1)
diff --git a/python/paddle/static/nn/control_flow.py b/python/paddle/static/nn/control_flow.py
index edf90babea525..2d17acb100dca 100644
--- a/python/paddle/static/nn/control_flow.py
+++ b/python/paddle/static/nn/control_flow.py
@@ -56,8 +56,9 @@ def Assert(cond, data=None, summarize=20, name=None):
         .. code-block:: python
 
             import paddle
-            from paddle.static.nn.control_flow import Assert
+            from paddle.static.nn import Assert
 
+            paddle.enable_static()
             x = paddle.full([2, 3], 2.0, 'float32')
             condition = paddle.max(x) < 1.0 # False
             Assert(condition, [x], 10, "example_assert_layer")
@@ -65,7 +66,7 @@ def Assert(cond, data=None, summarize=20, name=None):
             exe = paddle.static.Executor()
             try:
                 exe.run(paddle.static.default_main_program())
-                # Print x and throws paddle.framework.core.EnforceNotMet exception
+                # Print x and throws ValueError
                 # Example printed message for x:
                 #
                 # Variable: fill_constant_0.tmp_0
@@ -75,7 +76,7 @@ def Assert(cond, data=None, summarize=20, name=None):
                 #   - layout: NCHW
                 #   - dtype: float
                 #   - data: [2 2 2 2 2 2]
-            except paddle.framework.core.EnforceNotMet as e:
+            except ValueError as e:
                 print("Assert Exception Example")
 
     '''

From 3401e6fc55a782326b15c295156eec30fc65d139 Mon Sep 17 00:00:00 2001
From: feifei-111 <2364819892@qq.com>
Date: Thu, 8 Dec 2022 10:09:14 +0000
Subject: [PATCH 09/16] fix api path

---
 python/paddle/fluid/layers/control_flow.py |   1 -
 python/paddle/static/nn/control_flow.py    | 481 ++++++++++++++++++++-
 2 files changed, 472 insertions(+), 10 deletions(-)

diff --git a/python/paddle/fluid/layers/control_flow.py b/python/paddle/fluid/layers/control_flow.py
index 3dad7cf268d93..a2dcdc642942a 100755
--- a/python/paddle/fluid/layers/control_flow.py
+++ b/python/paddle/fluid/layers/control_flow.py
@@ -55,7 +55,6 @@
     'Switch',
     'array_write',
     'array_read',
-    'cond',
     'StaticRNN',
     'Print',
     'while_loop',
diff --git a/python/paddle/static/nn/control_flow.py b/python/paddle/static/nn/control_flow.py
index 32bb548bfef4d..f5896db4ea140 100644
--- a/python/paddle/static/nn/control_flow.py
+++ b/python/paddle/static/nn/control_flow.py
@@ -28,13 +28,17 @@
 from paddle.fluid.framework import Operator, Program, Variable
 
 # Temporary solution, it will be deleted later
-from paddle.fluid.layers.control_flow import cond
+from paddle.fluid.layers.control_flow import ConditionalBlock, select_input
+from paddle.fluid.layers.tensor import assign, cast
 from paddle.fluid.layers.utils import (
     assert_same_structure,
     copy_mutable_vars,
+    flatten,
     hold_mutable_vars,
     is_sequence,
     map_structure,
+    pack_sequence_as,
+    to_sequence,
 )
 from paddle.framework import in_dygraph_mode
 
@@ -66,13 +70,6 @@ def Assert(cond, data=None, summarize=20, name=None):
     Returns:
         Operator: the created operation.
 
-    Raises:
-        TypeError: If ``cond`` is not boolean Variable.
-        TypeError: If ``data`` is not a list or tuple or ``None``.
-        TypeError: If ``summarize`` is not int.
-        TypeError: If ``name`` is not a string or ``None`` .
-        framework.core.EnforceNotMet: If the condition is False in running time.
-
     Examples:
         .. code-block:: python
 
@@ -124,7 +121,7 @@ def Assert(cond, data=None, summarize=20, name=None):
 
 def increment(x, value=1.0, in_place=True):
     """
-    The OP is usually used for control flow to increment the data of :attr:`x` by an amount :attr:`value`.
+    This API is usually used for control flow to increment the data of :attr:`x` by an amount :attr:`value`.
     Notice that the number of elements in :attr:`x` must be equal to 1.
 
     Parameters:
@@ -923,3 +920,469 @@ def _check_args(branch_index, branch_fns, default):
 
     final_fn = false_fn
     return final_fn()
+
+
+def cond(pred, true_fn=None, false_fn=None, name=None, return_names=None):
+    """
+    This API returns ``true_fn()`` if the predicate ``pred`` is true else
+    ``false_fn()`` . Users could also set ``true_fn`` or ``false_fn`` to
+    ``None`` if do nothing and this API will treat the callable simply returns
+    ``None`` in this case.
+
+    ``true_fn`` and ``false_fn`` should return same nest structure of tensors
+    or both return ``None`` if user doens't like to return anything. A nest
+    structure of tensors in PaddlePaddle is tensor(s), or tuple of tensors, or
+    list of tensors.
+
+    Note:
+        1. The tuples or lists returned by ``true_fn`` and ``false_fn`` must have
+        the same shape because of dataflow model of PaddlePaddle while the
+        tensors in the tuples or the lists can have different shapes.
+
+        2. This API could be used under both static mode or dygraph mode. If it
+        is in dygraph mode, the API only runs one branch based on condition.
+
+        3. If it is in static mode, any tensors or operations created outside
+        or inside of ``true_fn`` and ``false_fn`` will be in net building
+        regardless of which branch is selected at runtime. This has frequently
+        surprised users who expected a lazy semantics. For example:
+
+        .. code-block:: python
+
+            import paddle
+
+            a = paddle.zeros((1, 1))
+            b = paddle.zeros((1, 1))
+            c = a * b
+            out = paddle.static.nn.cond(a < b, lambda: a + c, lambda: b * b)
+
+        No matter whether ``a < b`` , ``c = a * b`` will be in net building and
+        run. ``a + c`` and ``b * b`` will be in net building, but only one
+        branch will be executed during runtime.
+
+    Args:
+        pred(Tensor): A boolean tensor whose numel should be 1. The boolean
+            value determines whether to return the result of ``true_fn`` or
+            ``false_fn`` .
+        true_fn(callable, optional): A callable to be performed if ``pred`` is
+            true. The default value is ``None`` .
+        false_fn(callable, optional): A callable to be performed if ``pred`` is
+            false. The default value is ``None`` .
+        name(str, optional): The default value is ``None`` . Normally users
+             don't have to set this parameter. For more information, please
+             refer to :ref:`api_guide_Name` .
+        return_names(sequence of string, optional): The default value is ``None`` .
+             Normally users don't have to set this parameters.  A sequence of strings
+             to represents the name of returned vars.  The structure of sequence must
+             be same with return values of true_fn and false_fn.
+
+    Returns:
+        Tensor|list(Tensor)|tuple(Tensor): returns ``true_fn()`` if the
+        predicate ``pred`` is true else ``false_fn()`` .
+
+    Raises:
+        TypeError: if ``true_fn`` or ``false_fn`` is not callable.
+        ValueError: if ``true_fn`` and ``false_fn`` don't return the same nest
+            structure of tensors.
+
+    Examples:
+        .. code-block:: python
+
+            import paddle
+
+            #
+            # pseudocode:
+            # if 0.1 < 0.23:
+            #     return 1, True
+            # else:
+            #     return 3, 2
+            #
+
+            def true_func():
+                return paddle.full(shape=[1, 2], dtype='int32',
+                                   fill_value=1), paddle.full(shape=[2, 3],
+                                                              dtype='bool',
+                                                              fill_value=True)
+
+
+            def false_func():
+                return paddle.full(shape=[3, 4], dtype='float32',
+                                   fill_value=3), paddle.full(shape=[4, 5],
+                                                              dtype='int64',
+                                                              fill_value=2)
+
+
+            x = paddle.full(shape=[1], dtype='float32', fill_value=0.1)
+            y = paddle.full(shape=[1], dtype='float32', fill_value=0.23)
+            pred = paddle.less_than(x=x, y=y, name=None)
+            ret = paddle.static.nn.cond(pred, true_func, false_func)
+            # ret is a tuple containing 2 tensors
+            # ret[0] = [[1 1]]
+            # ret[1] = [[ True  True  True]
+            #           [ True  True  True]]
+
+    """
+    if _non_static_mode():
+        assert isinstance(pred, Variable), "The pred in cond must be Variable"
+        assert pred.size == 1, "condition input's numel should be 1"
+        pred = pred.numpy()[0]
+        if pred:
+            if true_fn is not None:
+                if not callable(true_fn):
+                    raise TypeError(
+                        "The true_fn in cond must be callable, but received {}".format(
+                            type(true_fn).__name__
+                        )
+                    )
+                return true_fn()
+        else:
+            if false_fn is not None:
+                if not callable(false_fn):
+                    raise TypeError(
+                        "The false_fn in cond must be callable, but received {}".format(
+                            type(false_fn).__name__
+                        )
+                    )
+                return false_fn()
+        return None
+
+    check_variable_and_dtype(pred, "pred", ['bool'], "fluid.layers.cond")
+    check_type(name, "name", (str, type(None)), "fluid.layers.cond")
+    helper = LayerHelper('cond', **locals())
+    true_output = None
+    false_output = None
+    copy_to_parent_func = lambda var: copy_var_to_parent_block(var, helper)
+    if true_fn is not None:
+        if not callable(true_fn):
+            raise TypeError(
+                "The true_fn in cond must be callable, but received {}".format(
+                    type(true_fn).__name__
+                )
+            )
+        true_cond_block = ConditionalBlock([pred], is_scalar_condition=True)
+        with true_cond_block.block():
+            origin_true_output = true_fn()
+            if origin_true_output is not None:
+                true_output = map_structure(
+                    copy_to_parent_func, origin_true_output
+                )
+    if false_fn is not None:
+        if not callable(false_fn):
+            raise TypeError(
+                "The false_fn in cond must be callable, but received {}".format(
+                    type(false_fn).__name__
+                )
+            )
+        false_cond_block = ConditionalBlock(
+            [paddle.logical_not(pred)], is_scalar_condition=True
+        )
+        with false_cond_block.block():
+            origin_false_output = false_fn()
+            if origin_false_output is not None:
+                false_output = map_structure(
+                    copy_to_parent_func, origin_false_output
+                )
+
+    if true_output is None and false_output is None:
+        return None
+
+    if true_output is None:
+        raise ValueError(
+            "Incompatible return values of true_fn and false_fn in cond: "
+            "true_fn returns None while false_fn returns non-None"
+        )
+    if false_output is None:
+        raise ValueError(
+            "Incompatible return values of true_fn and false_fn in cond: "
+            "true_fn returns non-None while false_fn returns None"
+        )
+
+    # Merge true and false output if they are not None
+    if return_names is None:
+        is_dy2staic = False
+        return_names = ["no name"] * len(_to_sequence_except_dict(true_output))
+    else:
+        """
+        dy2static will set the return_names and expand the return values to UndefinedVar.
+        """
+        is_dy2staic = True
+
+        # TODO:  expand_undefined_var will replace None to Undefinedvar(), to fix cases like:
+        #       a = None
+        #       if condition:
+        #           a = 1
+        # Because we can not use variable to express 'None'
+        true_output, false_output = expand_undefined_var(
+            true_output, false_output, return_names
+        )
+
+    if len(_to_sequence_except_dict(true_output)) != len(
+        _to_sequence_except_dict(false_output)
+    ):
+        raise ValueError(
+            "true fn returns {} vars, but false fn returns {} vars, which is not equals".format(
+                len(_to_sequence_except_dict(true_output)),
+                len(_to_sequence_except_dict(false_output)),
+            )
+        )
+    for true_out, false_out, return_name in zip(
+        _to_sequence_except_dict(true_output),
+        _to_sequence_except_dict(false_output),
+        _to_sequence_except_dict(return_names),
+    ):
+        try:
+            assert_same_structure(true_out, false_out, check_types=False)
+        except ValueError as e:
+            raise ValueError(
+                "Incompatible return values of `{}` in true_fn and false_fn in cond: {}".format(
+                    return_name, e
+                )
+            )
+
+    def check_ret_none(seq_true, seq_false, seq_names):
+        for f_true, f_false, f_name in zip(seq_true, seq_false, seq_names):
+            f_true = flatten(f_true)
+            f_false = flatten(f_false)
+            for idx in range(len(f_true)):
+                if (
+                    f_true[idx] is None
+                    and f_false[idx] is not None
+                    or f_false[idx] is None
+                    and f_true[idx] is not None
+                ):
+                    warnings.warn(
+                        "In cond : Var '{}' or part of it is set differently in ifelse branchs, "
+                        "<{}, {}> in true branch and <{}, {}> in false branch. Set var to "
+                        "'None' in ifelse block might lead to error.".format(
+                            f_name,
+                            type(f_true[idx]),
+                            f_true[idx],
+                            type(f_false[idx]),
+                            f_false[idx],
+                        )
+                    )
+
+    check_ret_none(
+        _to_sequence_except_dict(true_output),
+        _to_sequence_except_dict(false_output),
+        _to_sequence_except_dict(return_names),
+    )
+
+    if is_dy2staic:
+        true_output, false_output = change_none_to_undefinedvar(
+            true_output, false_output
+        )
+
+    mask = cast(pred, dtype='int32')
+    merge_func = (
+        lambda name, false_var, true_var: select_input_with_buildin_type(
+            [false_var, true_var], mask, name
+        )
+    )
+
+    def merge_every_var_list(false_vars, true_vars, name):
+        return map_structure(partial(merge_func, name), false_vars, true_vars)
+
+    merged_output = list(
+        map(
+            merge_every_var_list,
+            _to_sequence_except_dict(false_output),
+            _to_sequence_except_dict(true_output),
+            _to_sequence_except_dict(return_names),
+        )
+    )
+    merged_output = pack_sequence_as(false_output, flatten(merged_output))
+    return merged_output
+
+
+def copy_var_to_parent_block(var, layer_helper):
+    if not isinstance(var, Variable):
+        return var
+    prog = layer_helper.main_program
+    parent_idx = prog.current_block().parent_idx
+    assert (
+        parent_idx >= 0
+    ), "Got wrong parent block index when assigning var to parent scope in control_flow"
+    parent_block = prog.block(parent_idx)
+
+    if (
+        var.type == core.VarDesc.VarType.LOD_TENSOR_ARRAY
+        and parent_block._find_var_recursive(var.name)
+    ):
+        parent_block_var = var
+    else:
+        parent_block_var = parent_block.create_var(
+            dtype=var.dtype, shape=var.shape, type=var.type
+        )
+        assign(var, parent_block_var)
+    return parent_block_var
+
+
+def select_input_with_buildin_type(inputs, mask, name):
+    from paddle.jit.dy2static.utils import UndefinedVar
+    from paddle.jit.dy2static.variable_trans_func import to_static_variable
+
+    false_var, true_var = inputs
+
+    if isinstance(false_var, UndefinedVar) and isinstance(
+        true_var, UndefinedVar
+    ):
+        """None -> UndefinedVar, so the real value is a [None, UndefinedVar] or [None, None], we just return None."""
+        return None
+
+    if isinstance(false_var, Variable) and isinstance(true_var, Variable):
+        try:
+            return select_input(inputs, mask)
+        except Exception as e:
+            raise RuntimeError(
+                f"Exceptions throwed while doing select_input on {name}:\n{e}"
+            )
+
+    elif isinstance(false_var, support_ret_buildin_type) and isinstance(
+        false_var, type(true_var)
+    ):
+        if false_var == true_var:
+            return false_var
+        else:
+            inputs = [
+                to_static_variable(false_var),
+                to_static_variable(true_var),
+            ]
+    # Deal with the situations like this: false_var is int and true_var is Variable
+    elif (
+        isinstance(false_var, support_ret_buildin_type)
+        and isinstance(true_var, Variable)
+    ) or (
+        isinstance(true_var, support_ret_buildin_type)
+        and isinstance(false_var, Variable)
+    ):
+        inputs = [to_static_variable(false_var), to_static_variable(true_var)]
+        warnings.warn(
+            "Return results from different branches in cond are not same type: "
+            "false_var returned by false_fn is '{}' and true_var of true_fn is "
+            "'{}'".format(type(false_var), type(true_var))
+        )
+    elif (
+        isinstance(false_var, UndefinedVar)
+        and isinstance(true_var, (Variable,) + support_ret_buildin_type)
+    ) or (
+        isinstance(true_var, UndefinedVar)
+        and isinstance(false_var, (Variable,) + support_ret_buildin_type)
+    ):
+
+        def create_var_if_not_undefined_var(a):
+            if isinstance(a, UndefinedVar):
+                return a
+            return to_static_variable(a)
+
+        true_var, false_var = to_static_variable(true_var), to_static_variable(
+            false_var
+        )
+        inputs = [false_var, true_var]
+    else:
+        raise TypeError(
+            "Unsupported return type of true_fn and false_fn in cond: false_var "
+            "returned by false_fn is '{}' and true_var of true_fn is '{}'".format(
+                type(false_var), type(true_var)
+            )
+        )
+    try:
+        return select_input(inputs, mask)
+    except Exception as e:
+        raise RuntimeError(
+            f"Exceptions throwed while doing select_input on {name}:\n{e}"
+        )
+
+
+def _is_sequence_except_dict(x):
+    """
+    In this function, dict is not viewed as sequence.
+    """
+    if isinstance(x, dict):
+        return False
+    return is_sequence(x)
+
+
+def _to_sequence_except_dict(x):
+    """
+    In this function, dict is not viewed as sequence.
+    """
+    if isinstance(x, dict):
+        return [x]
+    return to_sequence(x)
+
+
+def expand_undefined_var(nest1, nest2, names):
+    """TODO: make this function recursively.
+    nest1: Var1, (UndefinedVar, [1,2,3])
+    nest2: Var2, ([1,2,3,4], UndefinedVar)
+    In this case, we should not expand recursively.
+    """
+    from paddle.jit.dy2static.return_transformer import RETURN_VALUE_PREFIX
+    from paddle.jit.dy2static.utils import UndefinedVar
+
+    def pack_undefined_var_as(seq):
+        return pack_sequence_as(
+            seq, [UndefinedVar("padding") for i in flatten(seq)]
+        )
+
+    def map_fn(n1, n2, name, order):
+        if not name.startswith(RETURN_VALUE_PREFIX) and (
+            isinstance(n1, UndefinedVar) or n1 is None
+        ):
+            if n1 is None and n2 is not None:
+                if order == 0:
+                    warnings.warn(
+                        "In cond : Var '{}' or part of it is set differently in ifelse branchs, "
+                        "<{}, {}> in true branch and <{}, {}> in false branch. Set var to "
+                        "'None' in ifelse block might lead to error.".format(
+                            name, type(n1), n1, type(n2), n2
+                        )
+                    )
+                else:
+                    warnings.warn(
+                        "In cond : Var '{}' or part of it is set differently in ifelse branchs, "
+                        "<{}, {}> in true branch and <{}, {}> in false branch. Set var to "
+                        "'None' in ifelse block might lead to error.".format(
+                            name, type(n2), n2, type(n1), n1
+                        )
+                    )
+            return pack_undefined_var_as(n2)
+        return n1
+
+    nest1_out = list(
+        map(
+            map_fn,
+            _to_sequence_except_dict(nest1),
+            _to_sequence_except_dict(nest2),
+            _to_sequence_except_dict(names),
+            [0 for i in _to_sequence_except_dict(names)],
+        )
+    )
+    nest2_out = list(
+        map(
+            map_fn,
+            _to_sequence_except_dict(nest2),
+            _to_sequence_except_dict(nest1),
+            _to_sequence_except_dict(names),
+            [1 for i in _to_sequence_except_dict(names)],
+        )
+    )
+    if not _is_sequence_except_dict(nest1):
+        nest1_out = nest1_out[0]
+    if not _is_sequence_except_dict(nest2):
+        nest2_out = nest2_out[0]
+    return nest1_out, nest2_out
+
+
+def change_none_to_undefinedvar(nest1, nest2):
+    from paddle.jit.dy2static.utils import UndefinedVar
+
+    def map_fn(x):
+        if x is None:
+            return UndefinedVar("padding")
+        return x
+
+    nest1_out = pack_sequence_as(nest1, list(map(map_fn, flatten(nest1))))
+    nest2_out = pack_sequence_as(nest2, list(map(map_fn, flatten(nest2))))
+    return nest1_out, nest2_out

From cd5329732aa9400dbddcf27497d9a62b396988b8 Mon Sep 17 00:00:00 2001
From: feifei-111 <2364819892@qq.com>
Date: Thu, 8 Dec 2022 11:09:52 +0000
Subject: [PATCH 10/16] fix fix circular import

---
 python/paddle/fluid/layers/control_flow.py       | 1 +
 python/paddle/jit/dy2static/convert_operators.py | 7 ++++---
 python/paddle/static/nn/control_flow.py          | 2 +-
 3 files changed, 6 insertions(+), 4 deletions(-)

diff --git a/python/paddle/fluid/layers/control_flow.py b/python/paddle/fluid/layers/control_flow.py
index a2dcdc642942a..34b009df67e18 100755
--- a/python/paddle/fluid/layers/control_flow.py
+++ b/python/paddle/fluid/layers/control_flow.py
@@ -387,6 +387,7 @@ def Print(
     return output
 
 
+# (TODO: Mine) There exists dependency. It will be removed later.
 class BlockGuard:
     """
     BlockGuard class.
diff --git a/python/paddle/jit/dy2static/convert_operators.py b/python/paddle/jit/dy2static/convert_operators.py
index c3e3163588bdc..fe9cd827d728c 100644
--- a/python/paddle/jit/dy2static/convert_operators.py
+++ b/python/paddle/jit/dy2static/convert_operators.py
@@ -33,7 +33,6 @@
     control_flow,
 )
 from paddle.fluid.layers.control_flow import (
-    cond,
     while_loop,
 )
 from .return_transformer import (
@@ -394,7 +393,7 @@ def new_false_fn():
             return ret
 
     try:
-        cond_outs = control_flow.cond(
+        cond_outs = paddle.static.nn.cond(
             pred, new_true_fn, new_false_fn, None, return_name_ids
         )
     except Exception as e:
@@ -820,7 +819,9 @@ def false_fn(array, start, end):
         new_array = paddle.slice(array, starts=[start], ends=[end], axes=[0])
         return new_array
 
-    new_array = cond(start == end, true_fn, lambda: false_fn(array, start, end))
+    new_array = paddle.static.nn.cond(
+        start == end, true_fn, lambda: false_fn(array, start, end)
+    )
     return new_array
 
 
diff --git a/python/paddle/static/nn/control_flow.py b/python/paddle/static/nn/control_flow.py
index f5896db4ea140..705f4eea2967f 100644
--- a/python/paddle/static/nn/control_flow.py
+++ b/python/paddle/static/nn/control_flow.py
@@ -24,6 +24,7 @@
     check_type,
     check_variable_and_dtype,
     convert_dtype,
+    in_dygraph_mode,
 )
 from paddle.fluid.framework import Operator, Program, Variable
 
@@ -40,7 +41,6 @@
     pack_sequence_as,
     to_sequence,
 )
-from paddle.framework import in_dygraph_mode
 
 __all__ = [
     'Assert',

From 14629e5b110acf2e00a6a1b60dd7985bffcc7f9f Mon Sep 17 00:00:00 2001
From: feifei-111 <2364819892@qq.com>
Date: Thu, 8 Dec 2022 11:18:38 +0000
Subject: [PATCH 11/16] del unused function

---
 python/paddle/fluid/layers/control_flow.py | 20 --------------------
 1 file changed, 20 deletions(-)

diff --git a/python/paddle/fluid/layers/control_flow.py b/python/paddle/fluid/layers/control_flow.py
index 34b009df67e18..7152116131644 100755
--- a/python/paddle/fluid/layers/control_flow.py
+++ b/python/paddle/fluid/layers/control_flow.py
@@ -90,26 +90,6 @@ def select_output(input, outputs, mask):
     return outputs
 
 
-def _select_input_infer_shape(first_shape, second_shape):
-    """
-    This function infer the output shape by following algorithm:
-    1. if the dims is different, raise a error.
-    2. compare axis one by one:
-        if a == b: we set axis to a
-        if a != b: we set axis to -1
-    for compatibility, non declarative mode, we just return second_shape.
-    """
-    if len(first_shape) != len(second_shape):
-        warnings.warn(
-            f"the input shapes of select_input should have the same rank, but get {first_shape}, {second_shape}"
-        )
-        return second_shape
-    out_shape = list(
-        map(lambda a, b: a if a == b else -1, first_shape, second_shape)
-    )
-    return out_shape
-
-
 def select_input(inputs, mask):
     """
     **select_input**

From 9f7f6cfb556177e479a82bdabbd4c1308c445c4a Mon Sep 17 00:00:00 2001
From: feifei-111 <2364819892@qq.com>
Date: Thu, 8 Dec 2022 11:20:59 +0000
Subject: [PATCH 12/16] remove unused function

---
 python/paddle/fluid/layers/control_flow.py | 38 ++++++++++++----------
 1 file changed, 20 insertions(+), 18 deletions(-)

diff --git a/python/paddle/fluid/layers/control_flow.py b/python/paddle/fluid/layers/control_flow.py
index 7152116131644..b19a077307a14 100755
--- a/python/paddle/fluid/layers/control_flow.py
+++ b/python/paddle/fluid/layers/control_flow.py
@@ -90,6 +90,26 @@ def select_output(input, outputs, mask):
     return outputs
 
 
+def _select_input_infer_shape(first_shape, second_shape):
+    """
+    This function infer the output shape by following algorithm:
+    1. if the dims is different, raise a error.
+    2. compare axis one by one:
+        if a == b: we set axis to a
+        if a != b: we set axis to -1
+    for compatibility，non declarative mode, we just return second_shape.
+    """
+    if len(first_shape) != len(second_shape):
+        warnings.warn(
+            f"the input shapes of select_input should have the same rank, but get {first_shape}, {second_shape}"
+        )
+        return second_shape
+    out_shape = list(
+        map(lambda a, b: a if a == b else -1, first_shape, second_shape)
+    )
+    return out_shape
+
+
 def select_input(inputs, mask):
     """
     **select_input**
@@ -111,24 +131,6 @@ def select_input(inputs, mask):
 
     # Select input should expand the shape. If it is - 1 and valid number, use - 1 first. If the dim is different, an error will be reported directly
     # assert inputs[0].dtype == inputs[1].dtype, f"Expect the inputs should have the same dtype, but get {inputs[0].dtype} and {inputs[1].dtype}"
-    def _select_input_infer_shape(first_shape, second_shape):
-        """
-        This function infer the output shape by following algorithm:
-        1. if the dims is different, raise a error.
-        2. compare axis one by one:
-            if a == b: we set axis to a
-            if a != b: we set axis to -1
-        for compatibility，non declarative mode, we just return second_shape.
-        """
-        if len(first_shape) != len(second_shape):
-            warnings.warn(
-                f"the input shapes of select_input should have the same rank, but get {first_shape}, {second_shape}"
-            )
-            return second_shape
-        out_shape = list(
-            map(lambda a, b: a if a == b else -1, first_shape, second_shape)
-        )
-        return out_shape
 
     output_shape = _select_input_infer_shape(inputs[0].shape, inputs[1].shape)
     output_dtype = inputs[1].dtype

From 3ca8284de33558de41f20ef5a4a785131848ccbf Mon Sep 17 00:00:00 2001
From: feifei-111 <2364819892@qq.com>
Date: Thu, 8 Dec 2022 11:29:32 +0000
Subject: [PATCH 13/16] update doc

---
 python/paddle/static/nn/control_flow.py | 5 -----
 1 file changed, 5 deletions(-)

diff --git a/python/paddle/static/nn/control_flow.py b/python/paddle/static/nn/control_flow.py
index 705f4eea2967f..e7f1903ccf0aa 100644
--- a/python/paddle/static/nn/control_flow.py
+++ b/python/paddle/static/nn/control_flow.py
@@ -980,11 +980,6 @@ def cond(pred, true_fn=None, false_fn=None, name=None, return_names=None):
         Tensor|list(Tensor)|tuple(Tensor): returns ``true_fn()`` if the
         predicate ``pred`` is true else ``false_fn()`` .
 
-    Raises:
-        TypeError: if ``true_fn`` or ``false_fn`` is not callable.
-        ValueError: if ``true_fn`` and ``false_fn`` don't return the same nest
-            structure of tensors.
-
     Examples:
         .. code-block:: python
 

From c93df21ba8b460a29c21b1bcf03fb6c2a30cb5ce Mon Sep 17 00:00:00 2001
From: feifei-111 <2364819892@qq.com>
Date: Thu, 8 Dec 2022 15:41:18 +0000
Subject: [PATCH 14/16] fix type err

---
 python/paddle/fluid/tests/unittests/test_cond.py   | 6 +++---
 python/paddle/fluid/tests/unittests/test_layers.py | 2 +-
 2 files changed, 4 insertions(+), 4 deletions(-)

diff --git a/python/paddle/fluid/tests/unittests/test_cond.py b/python/paddle/fluid/tests/unittests/test_cond.py
index c96cece765b24..3176ace0a3813 100644
--- a/python/paddle/fluid/tests/unittests/test_cond.py
+++ b/python/paddle/fluid/tests/unittests/test_cond.py
@@ -372,14 +372,14 @@ def test_cond_op_in_condition(self):
                 shape=[1], dtype='float32', value=1.24
             )
             b.stop_gradient = False
-            out = fluid.paddle.static.nn.cond(
+            out = paddle.static.nn.cond(
                 a < b,
-                lambda: fluid.paddle.static.nn.cond(
+                lambda: paddle.static.nn.cond(
                     a - b < -1.0,
                     lambda: paddle.add(a, b),
                     lambda: paddle.multiply(a, b),
                 ),
-                lambda: fluid.paddle.static.nn.cond(
+                lambda: paddle.static.nn.cond(
                     a == b,
                     lambda: paddle.subtract(a, b),
                     lambda: paddle.pow(a, b),
diff --git a/python/paddle/fluid/tests/unittests/test_layers.py b/python/paddle/fluid/tests/unittests/test_layers.py
index 868b6bc217b63..8074a5de52a8b 100644
--- a/python/paddle/fluid/tests/unittests/test_layers.py
+++ b/python/paddle/fluid/tests/unittests/test_layers.py
@@ -1577,7 +1577,7 @@ def greater_equal_branch(a, b):
             b = fluid.layers.fill_constant(
                 shape=[1], dtype='float32', value=0.23
             )
-            out = fluid.paddle.static.nn.cond(
+            out = paddle.static.nn.cond(
                 a >= b,
                 lambda: greater_equal_branch(a, b),
                 lambda: less_than_branch(a, b),

From b503a427f870d83c2d20c76ef025ea734a27714a Mon Sep 17 00:00:00 2001
From: feifei-111 <2364819892@qq.com>
Date: Fri, 9 Dec 2022 03:41:19 +0000
Subject: [PATCH 15/16] rm __all__ to hide api interface

---
 .../fleet/utils/hybrid_parallel_inference.py  |  4 +-
 python/paddle/fluid/layers/control_flow.py    | 47 ++---------------
 python/paddle/fluid/layers/rnn.py             |  4 +-
 python/paddle/fluid/optimizer.py              |  4 +-
 .../auto_parallel/test_while_op_partition.py  |  2 +-
 .../fleet/hybrid_parallel_inference_helper.py |  4 +-
 .../fluid/tests/unittests/dist_transformer.py |  6 +--
 .../unittests/dygraph_to_static/test_loop.py  |  9 ++--
 .../unittests/npu/test_increment_op_npu.py    |  2 +-
 .../tests/unittests/npu/test_while_op_npu.py  |  9 ++--
 .../unittests/test_array_read_write_op.py     |  9 ++--
 .../test_async_ssa_graph_executor_mnist.py    |  2 +-
 .../test_dynamic_rnn_stop_gradient.py         |  2 +-
 .../unittests/test_eager_deletion_while_op.py |  9 ++--
 .../tests/unittests/test_executor_and_mul.py  |  6 +--
 .../unittests/test_fetch_lod_tensor_array.py  |  6 +--
 .../fluid/tests/unittests/test_profiler.py    |  2 +-
 .../tests/unittests/test_while_loop_op.py     | 31 ++++++-----
 .../fluid/tests/unittests/test_while_op.py    | 11 ++--
 .../tests/unittests/xpu/test_while_op_xpu.py  |  9 ++--
 .../paddle/jit/dy2static/convert_operators.py |  4 +-
 python/paddle/static/nn/__init__.py           |  4 +-
 python/paddle/static/nn/control_flow.py       | 51 -------------------
 23 files changed, 65 insertions(+), 172 deletions(-)

diff --git a/python/paddle/distributed/fleet/utils/hybrid_parallel_inference.py b/python/paddle/distributed/fleet/utils/hybrid_parallel_inference.py
index 3cccbd782ecc9..f5010e815c37b 100644
--- a/python/paddle/distributed/fleet/utils/hybrid_parallel_inference.py
+++ b/python/paddle/distributed/fleet/utils/hybrid_parallel_inference.py
@@ -64,7 +64,7 @@ class HybridParallelInferenceHelper:
                 element_in_arr = layers.array_read(array=arr, i=step_idx)
                 # write placehold data to global lod_tensor_array,
                 # it need for send_v2 of lod_tensor_array
-                paddle.static.nn.increment(x=step_idx, value=1.0, in_place=True)
+                paddle.increment(x=step_idx, value=1.0, in_place=True)
                 layers.array_write(element_in_arr, i=step_idx, array=arr)
 
             with paddle.fluid.device_guard(f'{device}:0'):
@@ -137,7 +137,7 @@ class HybridParallelInferenceHelper:
             with while_op.block():
                 with paddle.fluid.device_guard(f'{device}:all'):
                     input = layers.array_read(array=data, i=step_idx)
-                    paddle.static.nn.increment(x=step_idx, value=1.0, in_place=True)
+                    paddle.increment(x=step_idx, value=1.0, in_place=True)
                     layers.array_write(input, i=step_idx, array=data)
 
                 with paddle.fluid.device_guard(f'{device}:0'):
diff --git a/python/paddle/fluid/layers/control_flow.py b/python/paddle/fluid/layers/control_flow.py
index b19a077307a14..1bf5de42e7e95 100755
--- a/python/paddle/fluid/layers/control_flow.py
+++ b/python/paddle/fluid/layers/control_flow.py
@@ -97,7 +97,7 @@ def _select_input_infer_shape(first_shape, second_shape):
     2. compare axis one by one:
         if a == b: we set axis to a
         if a != b: we set axis to -1
-    for compatibility，non declarative mode, we just return second_shape.
+    for compatibility, non declarative mode, we just return second_shape.
     """
     if len(first_shape) != len(second_shape):
         warnings.warn(
@@ -1063,7 +1063,7 @@ class While:
             cond = paddle.less_than(x=i, y=loop_len)
             while_op = fluid.layers.While(cond=cond)
             with while_op.block():
-                i = paddle.static.nn.increment(x=i, value=1, in_place=True)
+                i = paddle.increment(x=i, value=1, in_place=True)
                 paddle.assign(paddle.less_than(x=i, y=loop_len), cond)
 
             exe = fluid.Executor(fluid.CPUPlace())
@@ -1092,7 +1092,7 @@ class While:
             with while_op.block():
                 sums_tensor = fluid.layers.elementwise_add(x=data, y=data)
                 fluid.layers.assign(sums_tensor, sums)  # Update the value of sums_tensor defined in While to the sums which defined outside of While through layers.assign
-                i = paddle.static.nn.increment(x=i, value=1, in_place=True)
+                i = paddle.increment(x=i, value=1, in_place=True)
                 data = fluid.layers.elementwise_add(x=data, y=one)
                 paddle.assign(paddle.less_than(x=i, y=loop_len), cond)
 
@@ -1362,47 +1362,6 @@ def create_var_like(o_var):
     return results
 
 
-def increment(x, value=1.0, in_place=True):
-    """
-    The OP is usually used for control flow to increment the data of :attr:`x` by an amount :attr:`value`.
-    Notice that the number of elements in :attr:`x` must be equal to 1.
-
-    Parameters:
-        x (Variable): A tensor that must always contain only one element, its data type supports
-            float32, float64, int32 and int64.
-        value (float, optional): The amount to increment the data of :attr:`x`. Default: 1.0.
-        in_place (bool, optional): Whether the OP should be performed in-place. Default: True.
-
-    Returns:
-        Variable: The elementwise-incremented tensor with the same shape and data type as :attr:`x`.
-
-    Examples:
-        .. code-block:: python
-
-          import paddle.fluid as fluid
-          counter = fluid.layers.zeros(shape=[1], dtype='float32') # [0.]
-          fluid.layers.increment(counter) # [1.]
-    """
-    if in_dygraph_mode():
-        return _C_ops.increment_(x, value)
-
-    check_variable_and_dtype(
-        x, 'x', ['float32', 'float64', 'int32', 'int64'], 'increment'
-    )
-    helper = LayerHelper("increment", **locals())
-    if not in_place:
-        out = helper.create_variable_for_type_inference(dtype=x.dtype)
-    else:
-        out = x
-    helper.append_op(
-        type='increment',
-        inputs={'X': [x]},
-        outputs={'Out': [out]},
-        attrs={'step': float(value)},
-    )
-    return out
-
-
 def array_write(x, i, array=None):
     """
     This OP writes the input ``x`` into the i-th position of the ``array``
diff --git a/python/paddle/fluid/layers/rnn.py b/python/paddle/fluid/layers/rnn.py
index 97dedf2b4681c..2fd45e9f084d2 100644
--- a/python/paddle/fluid/layers/rnn.py
+++ b/python/paddle/fluid/layers/rnn.py
@@ -1536,7 +1536,7 @@ def _maybe_copy(state, new_state, step_mask):
             next_sequence_lengths,
         )
 
-        paddle.static.nn.increment(x=step_idx_tensor, value=1.0, in_place=True)
+        paddle.increment(x=step_idx_tensor, value=1.0, in_place=True)
         step_idx += 1
 
         cond = paddle.logical_not(paddle.all(finished))
@@ -1696,7 +1696,7 @@ def _create_array_out_of_while(dtype):
             outputs_arrays,
         )
 
-        paddle.static.nn.increment(x=step_idx, value=1.0, in_place=True)
+        paddle.increment(x=step_idx, value=1.0, in_place=True)
         # update the global_finished first, since it might be also in states of
         # decoder, which otherwise would write a stale finished status to array
         tensor.assign(next_finished, global_finished)
diff --git a/python/paddle/fluid/optimizer.py b/python/paddle/fluid/optimizer.py
index 444658ac35aa0..a61913e8c2e18 100755
--- a/python/paddle/fluid/optimizer.py
+++ b/python/paddle/fluid/optimizer.py
@@ -7283,7 +7283,7 @@ def minimize(self, loss, startup_program=None):
                 dtype='int32',
                 persistable=True,
             )
-            paddle.static.nn.increment(x=step, value=1.0, in_place=True)
+            paddle.increment(x=step, value=1.0, in_place=True)
 
             # lookahead
             zero_var = layers.fill_constant(
@@ -7517,7 +7517,7 @@ def _get_gm_cond_var(self, main_block):
 
         with device_guard("cpu"):
             # step_var = (step_var + 1) % k_step
-            paddle.static.nn.increment(x=step_var, value=1.0, in_place=True)
+            paddle.increment(x=step_var, value=1.0, in_place=True)
             main_block.append_op(
                 type='elementwise_mod',
                 inputs={'X': step_var, 'Y': k_step_var},
diff --git a/python/paddle/fluid/tests/unittests/auto_parallel/test_while_op_partition.py b/python/paddle/fluid/tests/unittests/auto_parallel/test_while_op_partition.py
index 6384684a8c24d..6e59cf16eafe8 100644
--- a/python/paddle/fluid/tests/unittests/auto_parallel/test_while_op_partition.py
+++ b/python/paddle/fluid/tests/unittests/auto_parallel/test_while_op_partition.py
@@ -189,7 +189,7 @@ def get_program():
             cur_pred = mlp_while(pre_input)
 
             # 更新循环条件
-            i = paddle.static.nn.increment(x=i, value=1, in_place=True)
+            i = paddle.increment(x=i, value=1, in_place=True)
             fluid.layers.array_write(cur_pred, array=input_array, i=i)
             paddle.assign(paddle.less_than(x=i, y=loop_len), cond)
 
diff --git a/python/paddle/fluid/tests/unittests/collective/fleet/hybrid_parallel_inference_helper.py b/python/paddle/fluid/tests/unittests/collective/fleet/hybrid_parallel_inference_helper.py
index 921bed35bd756..75e7e708ef4a9 100644
--- a/python/paddle/fluid/tests/unittests/collective/fleet/hybrid_parallel_inference_helper.py
+++ b/python/paddle/fluid/tests/unittests/collective/fleet/hybrid_parallel_inference_helper.py
@@ -91,9 +91,7 @@ def test_hybrid_parallel_inference_helper_mp1pp2(self):
             with while_op.block():
                 with paddle.fluid.device_guard(f'{device}:all'):
                     input = layers.array_read(array=data, i=step_idx)
-                    paddle.static.nn.increment(
-                        x=step_idx, value=1.0, in_place=True
-                    )
+                    paddle.increment(x=step_idx, value=1.0, in_place=True)
                     layers.array_write(input, i=step_idx, array=data)
 
                 with paddle.fluid.device_guard(f'{device}:0'):
diff --git a/python/paddle/fluid/tests/unittests/dist_transformer.py b/python/paddle/fluid/tests/unittests/dist_transformer.py
index db95b2062816c..a24319fca8a01 100644
--- a/python/paddle/fluid/tests/unittests/dist_transformer.py
+++ b/python/paddle/fluid/tests/unittests/dist_transformer.py
@@ -1813,9 +1813,7 @@ def beam_search():
                     shape=[-1, 1, 1],
                     dtype=pre_ids.dtype,
                 ),
-                y=paddle.static.nn.increment(
-                    x=step_idx, value=1.0, in_place=False
-                ),
+                y=paddle.increment(x=step_idx, value=1.0, in_place=False),
                 axis=0,
             )
             logits = wrap_decoder(
@@ -1854,7 +1852,7 @@ def beam_search():
                 end_id=eos_idx,
             )
 
-            paddle.static.nn.increment(x=step_idx, value=1.0, in_place=True)
+            paddle.increment(x=step_idx, value=1.0, in_place=True)
             # update states
             layers.array_write(selected_ids, i=step_idx, array=ids)
             layers.array_write(selected_scores, i=step_idx, array=scores)
diff --git a/python/paddle/fluid/tests/unittests/dygraph_to_static/test_loop.py b/python/paddle/fluid/tests/unittests/dygraph_to_static/test_loop.py
index 1efb3669d26e9..1f74b508d911a 100644
--- a/python/paddle/fluid/tests/unittests/dygraph_to_static/test_loop.py
+++ b/python/paddle/fluid/tests/unittests/dygraph_to_static/test_loop.py
@@ -21,7 +21,6 @@
 import paddle.fluid as fluid
 from paddle.jit.api import declarative
 from paddle.jit.dy2static.loop_transformer import NameVisitor
-from paddle.static.nn import increment
 from paddle.utils import gast
 
 SEED = 2020
@@ -84,7 +83,7 @@ def while_loop_dyfunc_with_none(x):
 def for_loop_dyfunc(max_len):
     for i in range(max_len):
         ret = fluid.layers.zeros(shape=[1], dtype='float32')
-        increment(ret, value=2.0, in_place=True)
+        paddle.increment(ret, value=2.0, in_place=True)
     return ret
 
 
@@ -105,14 +104,14 @@ def for_loop_dyfunc2(max_len):
 def for_loop_dyfunc3(max_len):
     ret = fluid.layers.zeros(shape=[1], dtype='float32')
     for i in range(1, 10, 2):
-        increment(ret, value=2.0, in_place=True)
+        paddle.increment(ret, value=2.0, in_place=True)
     return ret
 
 
 def for_loop_dyfunc4(max_len):
     ret = fluid.layers.zeros(shape=[1], dtype='float32')
     for i in range(10, 1, -2):
-        increment(ret, value=2.0, in_place=True)
+        paddle.increment(ret, value=2.0, in_place=True)
     return ret
 
 
@@ -120,7 +119,7 @@ def for_loop_dyfunc_not_support(max_len):
     ret = fluid.layers.zeros(shape=[1], dtype='float32')
     a = -2
     for i in range(10, 1, a):
-        increment(ret, value=2.0, in_place=True)
+        paddle.increment(ret, value=2.0, in_place=True)
     return ret
 
 
diff --git a/python/paddle/fluid/tests/unittests/npu/test_increment_op_npu.py b/python/paddle/fluid/tests/unittests/npu/test_increment_op_npu.py
index 542ced34256ae..2467860bd93a7 100644
--- a/python/paddle/fluid/tests/unittests/npu/test_increment_op_npu.py
+++ b/python/paddle/fluid/tests/unittests/npu/test_increment_op_npu.py
@@ -121,7 +121,7 @@ def test_npu(self):
 
         with paddle.static.program_guard(main_prog, startup_prog):
             a = paddle.static.data(name="a", shape=[1], dtype='float32')
-            b = paddle.static.nn.increment(a)
+            b = paddle.increment(a)
 
         place = paddle.NPUPlace(NPUPlace)
 
diff --git a/python/paddle/fluid/tests/unittests/npu/test_while_op_npu.py b/python/paddle/fluid/tests/unittests/npu/test_while_op_npu.py
index 09186b24e5482..3cbed5d5d683f 100644
--- a/python/paddle/fluid/tests/unittests/npu/test_while_op_npu.py
+++ b/python/paddle/fluid/tests/unittests/npu/test_while_op_npu.py
@@ -21,7 +21,6 @@
 from paddle.fluid.backward import append_backward
 import numpy
 from paddle.fluid import compiler, Program, program_guard
-from paddle.static.nn import increment
 
 paddle.enable_static()
 
@@ -44,9 +43,9 @@ def simple_net(self):
         init = layers.zeros(shape=[10], dtype='float32')
         mem_array = layers.array_write(x=init, i=i)
         data_array = layers.array_write(x=d0, i=i)
-        i = increment(i)
+        i = paddle.increment(i)
         layers.array_write(d1, i, array=data_array)
-        i = increment(i)
+        i = paddle.increment(i)
         layers.array_write(d2, i, array=data_array)
         i = layers.zeros(shape=[1], dtype='int32')
         i = layers.cast(i, 'int64')
@@ -72,7 +71,7 @@ def simple_net(self):
             prev = layers.array_read(array=mem_array, i=i)
             result = layers.sums(input=[d, prev])
 
-            i = increment(x=i, in_place=True)
+            i = paddle.increment(x=i, in_place=True)
             layers.array_write(result, i=i, array=mem_array)
             paddle.assign(paddle.less_than(x=i, y=array_len), cond)
 
@@ -81,7 +80,7 @@ def simple_net(self):
                 prev2 = layers.array_read(array=mem_array, i=j)
                 result2 = layers.sums(input=[d2, prev2])
 
-                j = increment(x=j, in_place=True)
+                j = paddle.increment(x=j, in_place=True)
                 layers.array_write(result2, i=j, array=mem_array)
                 paddle.assign(paddle.less_than(x=j, y=array_len2), cond2)
         sum_result = layers.array_read(array=mem_array, i=j)
diff --git a/python/paddle/fluid/tests/unittests/test_array_read_write_op.py b/python/paddle/fluid/tests/unittests/test_array_read_write_op.py
index 8f6f3bd58dd49..d60239a8ad169 100644
--- a/python/paddle/fluid/tests/unittests/test_array_read_write_op.py
+++ b/python/paddle/fluid/tests/unittests/test_array_read_write_op.py
@@ -24,24 +24,23 @@
 from paddle.fluid.backward import append_backward
 from paddle.fluid.executor import Executor
 from paddle.fluid.framework import default_main_program
-from paddle.static.nn import increment
 
 
 def _test_read_write(x):
     i = layers.zeros(shape=[1], dtype='int64')
     i.stop_gradient = False
     arr = layers.array_write(x=x[0], i=i)
-    i = increment(x=i)
+    i = paddle.increment(x=i)
     arr = layers.array_write(x=x[1], i=i, array=arr)
-    i = increment(x=i)
+    i = paddle.increment(x=i)
     arr = layers.array_write(x=x[2], i=i, array=arr)
 
     i = layers.zeros(shape=[1], dtype='int64')
     i.stop_gradient = False
     a0 = layers.array_read(array=arr, i=i)
-    i = increment(x=i)
+    i = paddle.increment(x=i)
     a1 = layers.array_read(array=arr, i=i)
-    i = increment(x=i)
+    i = paddle.increment(x=i)
     a2 = layers.array_read(array=arr, i=i)
 
     mean_a0 = paddle.mean(a0)
diff --git a/python/paddle/fluid/tests/unittests/test_async_ssa_graph_executor_mnist.py b/python/paddle/fluid/tests/unittests/test_async_ssa_graph_executor_mnist.py
index 45a743ed7e44d..4ccff076149df 100644
--- a/python/paddle/fluid/tests/unittests/test_async_ssa_graph_executor_mnist.py
+++ b/python/paddle/fluid/tests/unittests/test_async_ssa_graph_executor_mnist.py
@@ -62,7 +62,7 @@ def convolutional_neural_network(use_py_reader):
         acc = paddle.static.accuracy(input=prediction, label=label)
         i = fluid.layers.zeros(shape=[1], dtype='int64')
         array = fluid.layers.array_write(x=prediction, i=i)
-        paddle.static.nn.increment(i)
+        paddle.increment(i)
         fluid.layers.array_write(x=acc, i=i, array=array)
         return array, img, label, prediction, avg_loss, acc, py_reader
 
diff --git a/python/paddle/fluid/tests/unittests/test_dynamic_rnn_stop_gradient.py b/python/paddle/fluid/tests/unittests/test_dynamic_rnn_stop_gradient.py
index 1fb0ffe7015a0..89b7a2d73f294 100644
--- a/python/paddle/fluid/tests/unittests/test_dynamic_rnn_stop_gradient.py
+++ b/python/paddle/fluid/tests/unittests/test_dynamic_rnn_stop_gradient.py
@@ -53,7 +53,7 @@ def build_and_run_program(place, batch_size, beam_size, stop_gradient=False):
         topk_coordinates = paddle.stack([batch_pos, indices], axis=2)
         topk_coordinates.stop_gradient = stop_gradient
         score = paddle.gather_nd(x, topk_coordinates)
-        paddle.static.nn.increment(x=step_idx, value=1.0, in_place=True)
+        paddle.increment(x=step_idx, value=1.0, in_place=True)
         layers.array_write(score, i=step_idx, array=scores)
         length_cond = paddle.less_than(x=step_idx, y=max_len)
         layers.assign(length_cond, cond)
diff --git a/python/paddle/fluid/tests/unittests/test_eager_deletion_while_op.py b/python/paddle/fluid/tests/unittests/test_eager_deletion_while_op.py
index 5f304acc733fb..385f61388db94 100644
--- a/python/paddle/fluid/tests/unittests/test_eager_deletion_while_op.py
+++ b/python/paddle/fluid/tests/unittests/test_eager_deletion_while_op.py
@@ -27,7 +27,6 @@
 import paddle.fluid.core as core
 import paddle.fluid.layers as layers
 from paddle.fluid.executor import Executor
-from paddle.static.nn import increment
 
 paddle.enable_static()
 fluid.core._set_eager_deletion_mode(0.0, 1.0, True)
@@ -84,10 +83,10 @@ def run_main(self, place, with_data_parallel):
         mem_array = layers.array_write(x=init, i=i)
         data_array = layers.array_write(x=d0, i=i)
 
-        i = increment(i)
+        i = paddle.increment(i)
         layers.array_write(d1, i, array=data_array)
 
-        i = increment(i)
+        i = paddle.increment(i)
         layers.array_write(d2, i, array=data_array)
 
         i = layers.zeros(shape=[1], dtype='int64')
@@ -113,7 +112,7 @@ def run_main(self, place, with_data_parallel):
             prev = paddle.reshape(prev, shape=[10])
             result = layers.sums(input=[d, prev])
 
-            i = increment(x=i, in_place=True)
+            i = paddle.increment(x=i, in_place=True)
             layers.array_write(result, i=i, array=mem_array)
             paddle.assign(paddle.less_than(x=i, y=array_len), cond)
             with while_op2.block():
@@ -123,7 +122,7 @@ def run_main(self, place, with_data_parallel):
                 prev2 = paddle.reshape(prev2, shape=[10])
                 result2 = layers.sums(input=[d2, prev2])
 
-                j = increment(x=j, in_place=True)
+                j = paddle.increment(x=j, in_place=True)
                 layers.array_write(result2, i=j, array=mem_array)
                 paddle.assign(paddle.less_than(x=j, y=array_len2), cond2)
 
diff --git a/python/paddle/fluid/tests/unittests/test_executor_and_mul.py b/python/paddle/fluid/tests/unittests/test_executor_and_mul.py
index 265e612a13a98..bf54cef74b988 100644
--- a/python/paddle/fluid/tests/unittests/test_executor_and_mul.py
+++ b/python/paddle/fluid/tests/unittests/test_executor_and_mul.py
@@ -16,9 +16,9 @@
 
 import numpy as np
 
+import paddle
 from paddle.fluid.executor import Executor
 from paddle.fluid.layers import array_write, data, mul, zeros
-from paddle.static.nn import increment
 
 
 class TestExecutor(unittest.TestCase):
@@ -27,13 +27,13 @@ def test_mul(self):
         a = data(name='a', shape=[784], dtype='float32')
         array = array_write(x=a, i=i)
 
-        i = increment(i)
+        i = paddle.increment(i)
         b = data(
             name='b', shape=[784, 100], dtype='float32', append_batch_size=False
         )
         array_write(x=b, i=i, array=array)
 
-        i = increment(i)
+        i = paddle.increment(i)
         out = mul(x=a, y=b)
         array_write(x=out, i=i, array=array)
 
diff --git a/python/paddle/fluid/tests/unittests/test_fetch_lod_tensor_array.py b/python/paddle/fluid/tests/unittests/test_fetch_lod_tensor_array.py
index 8ca1a4d118fda..a55ecbad35b40 100644
--- a/python/paddle/fluid/tests/unittests/test_fetch_lod_tensor_array.py
+++ b/python/paddle/fluid/tests/unittests/test_fetch_lod_tensor_array.py
@@ -18,9 +18,9 @@
 import numpy as np
 from simple_nets import simple_fc_net, simple_fc_net_with_inputs
 
+import paddle
 import paddle.fluid as fluid
 import paddle.fluid.layers as layers
-from paddle.static.nn import increment
 
 
 class TestFetchLoDTensorArray(unittest.TestCase):
@@ -36,9 +36,9 @@ def build_program(self, main_program, startup_program):
                 opt.minimize(loss)
 
                 array = layers.array_write(x=img, i=i)
-                i = increment(i)
+                i = paddle.increment(i)
                 layers.array_write(x=label, i=i, array=array)
-                i = increment(i)
+                i = paddle.increment(i)
                 layers.array_write(x=loss, i=i, array=array)
 
                 return loss, array
diff --git a/python/paddle/fluid/tests/unittests/test_profiler.py b/python/paddle/fluid/tests/unittests/test_profiler.py
index d0bf4180e798a..3413b6b3d1def 100644
--- a/python/paddle/fluid/tests/unittests/test_profiler.py
+++ b/python/paddle/fluid/tests/unittests/test_profiler.py
@@ -50,7 +50,7 @@ def build_program(self, compile_program=True):
             with while_op.block():
                 hidden_n = fluid.layers.fc(input=hidden1, size=64, act='relu')
                 layers.array_write(hidden_n, i, data_arr)
-                paddle.static.nn.increment(x=counter, value=1, in_place=True)
+                paddle.increment(x=counter, value=1, in_place=True)
                 paddle.assign(paddle.less_than(x=counter, y=until), cond)
 
             hidden_n = layers.array_read(data_arr, i)
diff --git a/python/paddle/fluid/tests/unittests/test_while_loop_op.py b/python/paddle/fluid/tests/unittests/test_while_loop_op.py
index b36e595526c83..792387c76fd53 100644
--- a/python/paddle/fluid/tests/unittests/test_while_loop_op.py
+++ b/python/paddle/fluid/tests/unittests/test_while_loop_op.py
@@ -22,7 +22,6 @@
 import paddle.fluid.layers as layers
 from paddle.fluid.backward import append_backward
 from paddle.fluid.framework import Program, program_guard
-from paddle.static.nn import increment
 
 paddle.enable_static()
 
@@ -60,7 +59,7 @@ def cond(i, mem):
 
         def body(i, mem):
             mem = paddle.add(x=mem, y=one)
-            i = increment(i)
+            i = paddle.increment(i)
             return [i, mem]
 
         main_program = Program()
@@ -101,7 +100,7 @@ def body(i, ten, test_dict, test_list, test_list_dict):
                 test_list_dict[0]["test_key"]
             )
 
-            i = increment(i)
+            i = paddle.increment(i)
             return [i, ten, test_dict, test_list, test_list_dict]
 
         main_program = Program()
@@ -175,7 +174,7 @@ def internal_cond(j, init, sums):
             def internal_body(j, init, sums):
                 init = paddle.add(x=init, y=ones)
                 sums = paddle.add(x=init, y=sums)
-                j = increment(j)
+                j = paddle.increment(j)
                 return [j, init, sums]
 
             result = paddle.static.nn.while_loop(
@@ -185,7 +184,7 @@ def internal_body(j, init, sums):
             init = result[1]
             sums = result[2]
             sums = paddle.add(x=init, y=sums)
-            i = increment(i)
+            i = paddle.increment(i)
             return [i, j, init, sums]
 
         main_program = Program()
@@ -230,7 +229,7 @@ def cond(i, x):
 
         def body(i, x):
             x = paddle.multiply(x=i, y=i)
-            i = increment(i)
+            i = paddle.increment(i)
             return [i, x]
 
         main_program = Program()
@@ -325,7 +324,7 @@ def internal_body(j, x, mem_array):
                 inner_prev = layers.array_read(array=mem_array, i=j)
                 inner_sum_0 = paddle.add(x=inner_data, y=inner_prev)
                 inner_sum_1 = paddle.add(x=x, y=inner_sum_0)
-                j = increment(x=j, in_place=True)
+                j = paddle.increment(x=j, in_place=True)
                 layers.array_write(inner_sum_1, i=j, array=mem_array)
                 return [j, x, mem_array]
 
@@ -333,7 +332,7 @@ def internal_body(j, x, mem_array):
             outer_prev = layers.array_read(array=mem_array, i=i)
             outer_sum_0 = paddle.add(x=outer_data, y=outer_prev)
             outer_sum_1 = paddle.add(x=x, y=outer_sum_0)
-            i = increment(x=i, in_place=True)
+            i = paddle.increment(x=i, in_place=True)
             layers.array_write(outer_sum_1, i=i, array=mem_array)
             j, x, mem_array = paddle.static.nn.while_loop(
                 internal_cond, internal_body, [j, x, mem_array]
@@ -353,9 +352,9 @@ def internal_body(j, x, mem_array):
             init = layers.zeros(shape=[10], dtype='float32')
             mem_array = layers.array_write(x=init, i=i)
             data_array = layers.array_write(x=d0, i=i)
-            i = increment(i)
+            i = paddle.increment(i)
             layers.array_write(d1, i, array=data_array)
-            i = increment(i)
+            i = paddle.increment(i)
             layers.array_write(d2, i, array=data_array)
             i = layers.zeros(shape=[1], dtype='int64')
             i.stop_gradient = True
@@ -445,7 +444,7 @@ def cond_returns_constant(i):
             return 1
 
         def cond_returns_not_bool_tensor(i):
-            return increment(i)
+            return paddle.increment(i)
 
         def cond_returns_bool_tensor(i):
             return paddle.less_than(i, ten)
@@ -457,14 +456,14 @@ def cond_receives_two_args(i, ten):
             return paddle.less_than(i, ten)
 
         def body(i):
-            return increment(i)
+            return paddle.increment(i)
 
         def body_returns_error_length(i):
-            i = increment(i)
+            i = paddle.increment(i)
             return [i, i]
 
         def body_returns_error_type(i, ten):
-            return increment(i)
+            return paddle.increment(i)
 
         def cond_returns_with_mutable_dict(i, test_dict):
             return i > 0
@@ -473,7 +472,7 @@ def body_returns_with_mutable_dict(i, test_dict):
             test_dict['new_key'] = layers.fill_constant(
                 shape=[1], dtype='int64', value=1
             )
-            return increment(i), test_dict
+            return paddle.increment(i), test_dict
 
         def cond_returns_with_mutable_list(i, test_list):
             return i > 0
@@ -482,7 +481,7 @@ def body_returns_with_mutable_list(i, test_list):
             test_list.append(
                 layers.fill_constant(shape=[1], dtype='int64', value=1)
             )
-            return increment(i), test_list
+            return paddle.increment(i), test_list
 
         main_program = Program()
         startup_program = Program()
diff --git a/python/paddle/fluid/tests/unittests/test_while_op.py b/python/paddle/fluid/tests/unittests/test_while_op.py
index 638a89ef79d76..79e9ffbc2ffd8 100644
--- a/python/paddle/fluid/tests/unittests/test_while_op.py
+++ b/python/paddle/fluid/tests/unittests/test_while_op.py
@@ -22,7 +22,6 @@
 import paddle.fluid.layers as layers
 from paddle.fluid.backward import append_backward
 from paddle.fluid.executor import Executor
-from paddle.static.nn import increment
 
 paddle.enable_static()
 
@@ -43,9 +42,9 @@ def simple_net(self):
         init = layers.zeros(shape=[10], dtype='float32')
         mem_array = layers.array_write(x=init, i=i)
         data_array = layers.array_write(x=d0, i=i)
-        i = increment(i)
+        i = paddle.increment(i)
         layers.array_write(d1, i, array=data_array)
-        i = increment(i)
+        i = paddle.increment(i)
         layers.array_write(d2, i, array=data_array)
         i = layers.zeros(shape=[1], dtype='int64')
         i.stop_gradient = True
@@ -64,7 +63,7 @@ def simple_net(self):
             prev = layers.array_read(array=mem_array, i=i)
             result = layers.sums(input=[d, prev])
 
-            i = increment(x=i, in_place=True)
+            i = paddle.increment(x=i, in_place=True)
             layers.array_write(result, i=i, array=mem_array)
             paddle.assign(paddle.less_than(x=i, y=array_len), cond)
 
@@ -73,7 +72,7 @@ def simple_net(self):
                 prev2 = layers.array_read(array=mem_array, i=j)
                 result2 = layers.sums(input=[d2, prev2])
 
-                j = increment(x=j, in_place=True)
+                j = paddle.increment(x=j, in_place=True)
                 layers.array_write(result2, i=j, array=mem_array)
                 paddle.assign(paddle.less_than(x=j, y=array_len2), cond2)
         sum_result = layers.array_read(array=mem_array, i=j)
@@ -135,7 +134,7 @@ def test_error(self):
 
             def test_bad_x():
                 x = [1, 2, 3]
-                increment(x)
+                paddle.increment(x)
 
             self.assertRaises(TypeError, test_bad_x)
 
diff --git a/python/paddle/fluid/tests/unittests/xpu/test_while_op_xpu.py b/python/paddle/fluid/tests/unittests/xpu/test_while_op_xpu.py
index 1f8c759d46b8c..2d3253295d0e7 100644
--- a/python/paddle/fluid/tests/unittests/xpu/test_while_op_xpu.py
+++ b/python/paddle/fluid/tests/unittests/xpu/test_while_op_xpu.py
@@ -21,7 +21,6 @@
 import paddle.fluid.layers as layers
 from paddle.fluid.backward import append_backward
 from paddle.fluid.executor import Executor
-from paddle.static.nn import increment
 
 paddle.enable_static()
 
@@ -42,9 +41,9 @@ def simple_net(self):
         init = layers.zeros(shape=[10], dtype='float32')
         mem_array = layers.array_write(x=init, i=i)
         data_array = layers.array_write(x=d0, i=i)
-        i = increment(i)
+        i = paddle.increment(i)
         layers.array_write(d1, i, array=data_array)
-        i = increment(i)
+        i = paddle.increment(i)
         layers.array_write(d2, i, array=data_array)
         i = layers.zeros(shape=[1], dtype='int64')
         i.stop_gradient = True
@@ -63,7 +62,7 @@ def simple_net(self):
             prev = layers.array_read(array=mem_array, i=i)
             result = layers.sums(input=[d, prev])
 
-            i = increment(x=i, in_place=True)
+            i = paddle.increment(x=i, in_place=True)
             layers.array_write(result, i=i, array=mem_array)
             paddle.assign(paddle.less_than(x=i, y=array_len), cond)
 
@@ -72,7 +71,7 @@ def simple_net(self):
                 prev2 = layers.array_read(array=mem_array, i=j)
                 result2 = layers.sums(input=[d2, prev2])
 
-                j = increment(x=j, in_place=True)
+                j = paddle.increment(x=j, in_place=True)
                 layers.array_write(result2, i=j, array=mem_array)
                 paddle.assign(paddle.less_than(x=j, y=array_len2), cond2)
         sum_result = layers.array_read(array=mem_array, i=j)
diff --git a/python/paddle/jit/dy2static/convert_operators.py b/python/paddle/jit/dy2static/convert_operators.py
index fe9cd827d728c..1a814086f5790 100644
--- a/python/paddle/jit/dy2static/convert_operators.py
+++ b/python/paddle/jit/dy2static/convert_operators.py
@@ -787,9 +787,7 @@ def body(i, new_array):
         item = array_read(array=array, i=i)
         array_write(item, paddle.tensor.array_length(new_array), new_array)
 
-        from paddle.static.nn import increment
-
-        i = increment(i)
+        i = paddle.increment(i)
         return i, new_array
 
     arr_len = paddle.tensor.array_length(array)
diff --git a/python/paddle/static/nn/__init__.py b/python/paddle/static/nn/__init__.py
index 7cefb7f5a5132..f446769837eb6 100755
--- a/python/paddle/static/nn/__init__.py
+++ b/python/paddle/static/nn/__init__.py
@@ -58,7 +58,7 @@
 from ...fluid.layers.sequence_lod import sequence_enumerate  # noqa: F401
 from ...fluid.layers.sequence_lod import sequence_reverse  # noqa: F401
 
-from .control_flow import Assert, increment, cond
+from .control_flow import cond
 
 __all__ = [  # noqa
     'fc',
@@ -102,6 +102,4 @@
     'sequence_reverse',
     'StaticRNN',
     'prelu',
-    'Assert',
-    'increment',
 ]
diff --git a/python/paddle/static/nn/control_flow.py b/python/paddle/static/nn/control_flow.py
index e7f1903ccf0aa..d331dd419fada 100644
--- a/python/paddle/static/nn/control_flow.py
+++ b/python/paddle/static/nn/control_flow.py
@@ -17,14 +17,12 @@
 
 import paddle
 import paddle.fluid.core as core
-from paddle import _C_ops
 from paddle.common_ops_import import (
     LayerHelper,
     _non_static_mode,
     check_type,
     check_variable_and_dtype,
     convert_dtype,
-    in_dygraph_mode,
 )
 from paddle.fluid.framework import Operator, Program, Variable
 
@@ -42,12 +40,6 @@
     to_sequence,
 )
 
-__all__ = [
-    'Assert',
-    'increment',
-    'cond',
-]
-
 
 def Assert(cond, data=None, summarize=20, name=None):
     '''
@@ -119,49 +111,6 @@ def Assert(cond, data=None, summarize=20, name=None):
     return op
 
 
-def increment(x, value=1.0, in_place=True):
-    """
-    This API is usually used for control flow to increment the data of :attr:`x` by an amount :attr:`value`.
-    Notice that the number of elements in :attr:`x` must be equal to 1.
-
-    Parameters:
-        x (Variable): A tensor that must always contain only one element, its data type supports
-            float32, float64, int32 and int64.
-        value (float, optional): The amount to increment the data of :attr:`x`. Default: 1.0.
-        in_place (bool, optional): Whether the OP should be performed in-place. Default: True.
-
-    Returns:
-        Variable: The elementwise-incremented tensor with the same shape and data type as :attr:`x`.
-
-    Examples:
-        .. code-block:: python
-
-          import paddle
-          from paddle.static.nn.control_flow import increment
-
-          counter = paddle.zeros(shape=[1], dtype='float32') # [0.]
-          increment(counter) # [1.]
-    """
-    if in_dygraph_mode():
-        return _C_ops.increment_(x, value)
-
-    check_variable_and_dtype(
-        x, 'x', ['float32', 'float64', 'int32', 'int64'], 'increment'
-    )
-    helper = LayerHelper("increment", **locals())
-    if not in_place:
-        out = helper.create_variable_for_type_inference(dtype=x.dtype)
-    else:
-        out = x
-    helper.append_op(
-        type='increment',
-        inputs={'X': [x]},
-        outputs={'Out': [out]},
-        attrs={'step': float(value)},
-    )
-    return out
-
-
 class BlockGuard:
     """
     BlockGuard class.

From afb67469f00e0209a64b3e20f3b6fcf721498b9d Mon Sep 17 00:00:00 2001
From: feifei-111 <2364819892@qq.com>
Date: Fri, 9 Dec 2022 08:30:15 +0000
Subject: [PATCH 16/16] fix conflict

---
 .../distributed/fleet/utils/hybrid_parallel_inference.py  | 4 ++--
 python/paddle/fluid/layers/control_flow.py                | 4 ++--
 python/paddle/fluid/layers/rnn.py                         | 4 ++--
 python/paddle/fluid/optimizer.py                          | 4 ++--
 .../unittests/auto_parallel/test_while_op_partition.py    | 2 +-
 .../collective/fleet/hybrid_parallel_inference_helper.py  | 2 +-
 python/paddle/fluid/tests/unittests/dist_transformer.py   | 4 ++--
 .../fluid/tests/unittests/dygraph_to_static/test_loop.py  | 8 ++++----
 .../paddle/fluid/tests/unittests/npu/test_while_op_npu.py | 4 ++--
 python/paddle/fluid/tests/unittests/test_assert_op.py     | 2 +-
 .../tests/unittests/test_dynamic_rnn_stop_gradient.py     | 2 +-
 .../fluid/tests/unittests/test_eager_deletion_while_op.py | 4 ++--
 python/paddle/fluid/tests/unittests/test_profiler.py      | 2 +-
 python/paddle/fluid/tests/unittests/test_while_loop_op.py | 4 ++--
 python/paddle/fluid/tests/unittests/test_while_op.py      | 4 ++--
 .../paddle/fluid/tests/unittests/xpu/test_while_op_xpu.py | 4 ++--
 python/paddle/jit/dy2static/convert_operators.py          | 2 +-
 python/paddle/static/nn/control_flow.py                   | 2 +-
 18 files changed, 31 insertions(+), 31 deletions(-)

diff --git a/python/paddle/distributed/fleet/utils/hybrid_parallel_inference.py b/python/paddle/distributed/fleet/utils/hybrid_parallel_inference.py
index f5010e815c37b..8085ac077c311 100644
--- a/python/paddle/distributed/fleet/utils/hybrid_parallel_inference.py
+++ b/python/paddle/distributed/fleet/utils/hybrid_parallel_inference.py
@@ -64,7 +64,7 @@ class HybridParallelInferenceHelper:
                 element_in_arr = layers.array_read(array=arr, i=step_idx)
                 # write placehold data to global lod_tensor_array,
                 # it need for send_v2 of lod_tensor_array
-                paddle.increment(x=step_idx, value=1.0, in_place=True)
+                paddle.increment(x=step_idx, value=1.0)
                 layers.array_write(element_in_arr, i=step_idx, array=arr)
 
             with paddle.fluid.device_guard(f'{device}:0'):
@@ -137,7 +137,7 @@ class HybridParallelInferenceHelper:
             with while_op.block():
                 with paddle.fluid.device_guard(f'{device}:all'):
                     input = layers.array_read(array=data, i=step_idx)
-                    paddle.increment(x=step_idx, value=1.0, in_place=True)
+                    paddle.increment(x=step_idx, value=1.0)
                     layers.array_write(input, i=step_idx, array=data)
 
                 with paddle.fluid.device_guard(f'{device}:0'):
diff --git a/python/paddle/fluid/layers/control_flow.py b/python/paddle/fluid/layers/control_flow.py
index 7ad3407db4330..ed15416fd35e3 100755
--- a/python/paddle/fluid/layers/control_flow.py
+++ b/python/paddle/fluid/layers/control_flow.py
@@ -1063,7 +1063,7 @@ class While:
             cond = paddle.less_than(x=i, y=loop_len)
             while_op = fluid.layers.While(cond=cond)
             with while_op.block():
-                i = paddle.increment(x=i, value=1, in_place=True)
+                i = paddle.increment(x=i, value=1)
                 paddle.assign(paddle.less_than(x=i, y=loop_len), cond)
 
             exe = fluid.Executor(fluid.CPUPlace())
@@ -1092,7 +1092,7 @@ class While:
             with while_op.block():
                 sums_tensor = fluid.layers.elementwise_add(x=data, y=data)
                 fluid.layers.assign(sums_tensor, sums)  # Update the value of sums_tensor defined in While to the sums which defined outside of While through layers.assign
-                i = paddle.increment(x=i, value=1, in_place=True)
+                i = paddle.increment(x=i, value=1)
                 data = fluid.layers.elementwise_add(x=data, y=one)
                 paddle.assign(paddle.less_than(x=i, y=loop_len), cond)
 
diff --git a/python/paddle/fluid/layers/rnn.py b/python/paddle/fluid/layers/rnn.py
index 0b03fb97183d9..f03ceb4b1bbf4 100644
--- a/python/paddle/fluid/layers/rnn.py
+++ b/python/paddle/fluid/layers/rnn.py
@@ -907,7 +907,7 @@ def _maybe_copy(state, new_state, step_mask):
             next_sequence_lengths,
         )
 
-        paddle.increment(x=step_idx_tensor, value=1.0, in_place=True)
+        paddle.increment(x=step_idx_tensor, value=1.0)
         step_idx += 1
 
         cond = paddle.logical_not(paddle.all(finished))
@@ -1067,7 +1067,7 @@ def _create_array_out_of_while(dtype):
             outputs_arrays,
         )
 
-        paddle.increment(x=step_idx, value=1.0, in_place=True)
+        paddle.increment(x=step_idx, value=1.0)
         # update the global_finished first, since it might be also in states of
         # decoder, which otherwise would write a stale finished status to array
         tensor.assign(next_finished, global_finished)
diff --git a/python/paddle/fluid/optimizer.py b/python/paddle/fluid/optimizer.py
index 2a2141a2e1ca1..550ea608c0465 100755
--- a/python/paddle/fluid/optimizer.py
+++ b/python/paddle/fluid/optimizer.py
@@ -7299,7 +7299,7 @@ def minimize(self, loss, startup_program=None):
                 dtype='int32',
                 persistable=True,
             )
-            paddle.increment(x=step, value=1.0, in_place=True)
+            paddle.increment(x=step, value=1.0)
 
             # lookahead
             zero_var = layers.fill_constant(
@@ -7533,7 +7533,7 @@ def _get_gm_cond_var(self, main_block):
 
         with device_guard("cpu"):
             # step_var = (step_var + 1) % k_step
-            paddle.increment(x=step_var, value=1.0, in_place=True)
+            paddle.increment(x=step_var, value=1.0)
             main_block.append_op(
                 type='elementwise_mod',
                 inputs={'X': step_var, 'Y': k_step_var},
diff --git a/python/paddle/fluid/tests/unittests/auto_parallel/test_while_op_partition.py b/python/paddle/fluid/tests/unittests/auto_parallel/test_while_op_partition.py
index 6e59cf16eafe8..8b0f746aa6576 100644
--- a/python/paddle/fluid/tests/unittests/auto_parallel/test_while_op_partition.py
+++ b/python/paddle/fluid/tests/unittests/auto_parallel/test_while_op_partition.py
@@ -189,7 +189,7 @@ def get_program():
             cur_pred = mlp_while(pre_input)
 
             # 更新循环条件
-            i = paddle.increment(x=i, value=1, in_place=True)
+            i = paddle.increment(x=i, value=1)
             fluid.layers.array_write(cur_pred, array=input_array, i=i)
             paddle.assign(paddle.less_than(x=i, y=loop_len), cond)
 
diff --git a/python/paddle/fluid/tests/unittests/collective/fleet/hybrid_parallel_inference_helper.py b/python/paddle/fluid/tests/unittests/collective/fleet/hybrid_parallel_inference_helper.py
index 75e7e708ef4a9..44bcbb6e70908 100644
--- a/python/paddle/fluid/tests/unittests/collective/fleet/hybrid_parallel_inference_helper.py
+++ b/python/paddle/fluid/tests/unittests/collective/fleet/hybrid_parallel_inference_helper.py
@@ -91,7 +91,7 @@ def test_hybrid_parallel_inference_helper_mp1pp2(self):
             with while_op.block():
                 with paddle.fluid.device_guard(f'{device}:all'):
                     input = layers.array_read(array=data, i=step_idx)
-                    paddle.increment(x=step_idx, value=1.0, in_place=True)
+                    paddle.increment(x=step_idx, value=1.0)
                     layers.array_write(input, i=step_idx, array=data)
 
                 with paddle.fluid.device_guard(f'{device}:0'):
diff --git a/python/paddle/fluid/tests/unittests/dist_transformer.py b/python/paddle/fluid/tests/unittests/dist_transformer.py
index d282ea7858c93..cb986dfabbbfd 100644
--- a/python/paddle/fluid/tests/unittests/dist_transformer.py
+++ b/python/paddle/fluid/tests/unittests/dist_transformer.py
@@ -1813,7 +1813,7 @@ def beam_search():
                     shape=[-1, 1, 1],
                     dtype=pre_ids.dtype,
                 ),
-                y=paddle.increment(x=step_idx, value=1.0, in_place=False),
+                y=paddle.increment(x=step_idx, value=1.0),
                 axis=0,
             )
             logits = wrap_decoder(
@@ -1852,7 +1852,7 @@ def beam_search():
                 end_id=eos_idx,
             )
 
-            paddle.increment(x=step_idx, value=1.0, in_place=True)
+            paddle.increment(x=step_idx, value=1.0)
             # update states
             layers.array_write(selected_ids, i=step_idx, array=ids)
             layers.array_write(selected_scores, i=step_idx, array=scores)
diff --git a/python/paddle/fluid/tests/unittests/dygraph_to_static/test_loop.py b/python/paddle/fluid/tests/unittests/dygraph_to_static/test_loop.py
index bde9aea72509d..4d2a70ee1cf4b 100644
--- a/python/paddle/fluid/tests/unittests/dygraph_to_static/test_loop.py
+++ b/python/paddle/fluid/tests/unittests/dygraph_to_static/test_loop.py
@@ -83,7 +83,7 @@ def while_loop_dyfunc_with_none(x):
 def for_loop_dyfunc(max_len):
     for i in range(max_len):
         ret = fluid.layers.zeros(shape=[1], dtype='float32')
-        paddle.increment(ret, value=2.0, in_place=True)
+        paddle.increment(ret, value=2.0)
     return ret
 
 
@@ -104,14 +104,14 @@ def for_loop_dyfunc2(max_len):
 def for_loop_dyfunc3(max_len):
     ret = fluid.layers.zeros(shape=[1], dtype='float32')
     for i in range(1, 10, 2):
-        paddle.increment(ret, value=2.0, in_place=True)
+        paddle.increment(ret, value=2.0)
     return ret
 
 
 def for_loop_dyfunc4(max_len):
     ret = fluid.layers.zeros(shape=[1], dtype='float32')
     for i in range(10, 1, -2):
-        paddle.increment(ret, value=2.0, in_place=True)
+        paddle.increment(ret, value=2.0)
     return ret
 
 
@@ -119,7 +119,7 @@ def for_loop_dyfunc_not_support(max_len):
     ret = fluid.layers.zeros(shape=[1], dtype='float32')
     a = -2
     for i in range(10, 1, a):
-        paddle.increment(ret, value=2.0, in_place=True)
+        paddle.increment(ret, value=2.0)
     return ret
 
 
diff --git a/python/paddle/fluid/tests/unittests/npu/test_while_op_npu.py b/python/paddle/fluid/tests/unittests/npu/test_while_op_npu.py
index 3cbed5d5d683f..c2b5d73b76d95 100644
--- a/python/paddle/fluid/tests/unittests/npu/test_while_op_npu.py
+++ b/python/paddle/fluid/tests/unittests/npu/test_while_op_npu.py
@@ -71,7 +71,7 @@ def simple_net(self):
             prev = layers.array_read(array=mem_array, i=i)
             result = layers.sums(input=[d, prev])
 
-            i = paddle.increment(x=i, in_place=True)
+            i = paddle.increment(x=i)
             layers.array_write(result, i=i, array=mem_array)
             paddle.assign(paddle.less_than(x=i, y=array_len), cond)
 
@@ -80,7 +80,7 @@ def simple_net(self):
                 prev2 = layers.array_read(array=mem_array, i=j)
                 result2 = layers.sums(input=[d2, prev2])
 
-                j = paddle.increment(x=j, in_place=True)
+                j = paddle.increment(x=j)
                 layers.array_write(result2, i=j, array=mem_array)
                 paddle.assign(paddle.less_than(x=j, y=array_len2), cond2)
         sum_result = layers.array_read(array=mem_array, i=j)
diff --git a/python/paddle/fluid/tests/unittests/test_assert_op.py b/python/paddle/fluid/tests/unittests/test_assert_op.py
index a6ce260fb9435..f62ad38d459d1 100644
--- a/python/paddle/fluid/tests/unittests/test_assert_op.py
+++ b/python/paddle/fluid/tests/unittests/test_assert_op.py
@@ -17,7 +17,7 @@
 import paddle
 import paddle.fluid as fluid
 import paddle.fluid.layers as layers
-from paddle.static.nn import Assert
+from paddle.static.nn.control_flow import Assert
 
 
 class TestAssertOp(unittest.TestCase):
diff --git a/python/paddle/fluid/tests/unittests/test_dynamic_rnn_stop_gradient.py b/python/paddle/fluid/tests/unittests/test_dynamic_rnn_stop_gradient.py
index 89b7a2d73f294..0fae895098083 100644
--- a/python/paddle/fluid/tests/unittests/test_dynamic_rnn_stop_gradient.py
+++ b/python/paddle/fluid/tests/unittests/test_dynamic_rnn_stop_gradient.py
@@ -53,7 +53,7 @@ def build_and_run_program(place, batch_size, beam_size, stop_gradient=False):
         topk_coordinates = paddle.stack([batch_pos, indices], axis=2)
         topk_coordinates.stop_gradient = stop_gradient
         score = paddle.gather_nd(x, topk_coordinates)
-        paddle.increment(x=step_idx, value=1.0, in_place=True)
+        paddle.increment(x=step_idx, value=1.0)
         layers.array_write(score, i=step_idx, array=scores)
         length_cond = paddle.less_than(x=step_idx, y=max_len)
         layers.assign(length_cond, cond)
diff --git a/python/paddle/fluid/tests/unittests/test_eager_deletion_while_op.py b/python/paddle/fluid/tests/unittests/test_eager_deletion_while_op.py
index 385f61388db94..ce23dcb54e5b0 100644
--- a/python/paddle/fluid/tests/unittests/test_eager_deletion_while_op.py
+++ b/python/paddle/fluid/tests/unittests/test_eager_deletion_while_op.py
@@ -112,7 +112,7 @@ def run_main(self, place, with_data_parallel):
             prev = paddle.reshape(prev, shape=[10])
             result = layers.sums(input=[d, prev])
 
-            i = paddle.increment(x=i, in_place=True)
+            i = paddle.increment(x=i)
             layers.array_write(result, i=i, array=mem_array)
             paddle.assign(paddle.less_than(x=i, y=array_len), cond)
             with while_op2.block():
@@ -122,7 +122,7 @@ def run_main(self, place, with_data_parallel):
                 prev2 = paddle.reshape(prev2, shape=[10])
                 result2 = layers.sums(input=[d2, prev2])
 
-                j = paddle.increment(x=j, in_place=True)
+                j = paddle.increment(x=j)
                 layers.array_write(result2, i=j, array=mem_array)
                 paddle.assign(paddle.less_than(x=j, y=array_len2), cond2)
 
diff --git a/python/paddle/fluid/tests/unittests/test_profiler.py b/python/paddle/fluid/tests/unittests/test_profiler.py
index 3413b6b3d1def..6e5c130ce954f 100644
--- a/python/paddle/fluid/tests/unittests/test_profiler.py
+++ b/python/paddle/fluid/tests/unittests/test_profiler.py
@@ -50,7 +50,7 @@ def build_program(self, compile_program=True):
             with while_op.block():
                 hidden_n = fluid.layers.fc(input=hidden1, size=64, act='relu')
                 layers.array_write(hidden_n, i, data_arr)
-                paddle.increment(x=counter, value=1, in_place=True)
+                paddle.increment(x=counter, value=1)
                 paddle.assign(paddle.less_than(x=counter, y=until), cond)
 
             hidden_n = layers.array_read(data_arr, i)
diff --git a/python/paddle/fluid/tests/unittests/test_while_loop_op.py b/python/paddle/fluid/tests/unittests/test_while_loop_op.py
index 792387c76fd53..533429d5b0fc8 100644
--- a/python/paddle/fluid/tests/unittests/test_while_loop_op.py
+++ b/python/paddle/fluid/tests/unittests/test_while_loop_op.py
@@ -324,7 +324,7 @@ def internal_body(j, x, mem_array):
                 inner_prev = layers.array_read(array=mem_array, i=j)
                 inner_sum_0 = paddle.add(x=inner_data, y=inner_prev)
                 inner_sum_1 = paddle.add(x=x, y=inner_sum_0)
-                j = paddle.increment(x=j, in_place=True)
+                j = paddle.increment(x=j)
                 layers.array_write(inner_sum_1, i=j, array=mem_array)
                 return [j, x, mem_array]
 
@@ -332,7 +332,7 @@ def internal_body(j, x, mem_array):
             outer_prev = layers.array_read(array=mem_array, i=i)
             outer_sum_0 = paddle.add(x=outer_data, y=outer_prev)
             outer_sum_1 = paddle.add(x=x, y=outer_sum_0)
-            i = paddle.increment(x=i, in_place=True)
+            i = paddle.increment(x=i)
             layers.array_write(outer_sum_1, i=i, array=mem_array)
             j, x, mem_array = paddle.static.nn.while_loop(
                 internal_cond, internal_body, [j, x, mem_array]
diff --git a/python/paddle/fluid/tests/unittests/test_while_op.py b/python/paddle/fluid/tests/unittests/test_while_op.py
index 79e9ffbc2ffd8..d9b8e52172871 100644
--- a/python/paddle/fluid/tests/unittests/test_while_op.py
+++ b/python/paddle/fluid/tests/unittests/test_while_op.py
@@ -63,7 +63,7 @@ def simple_net(self):
             prev = layers.array_read(array=mem_array, i=i)
             result = layers.sums(input=[d, prev])
 
-            i = paddle.increment(x=i, in_place=True)
+            i = paddle.increment(x=i)
             layers.array_write(result, i=i, array=mem_array)
             paddle.assign(paddle.less_than(x=i, y=array_len), cond)
 
@@ -72,7 +72,7 @@ def simple_net(self):
                 prev2 = layers.array_read(array=mem_array, i=j)
                 result2 = layers.sums(input=[d2, prev2])
 
-                j = paddle.increment(x=j, in_place=True)
+                j = paddle.increment(x=j)
                 layers.array_write(result2, i=j, array=mem_array)
                 paddle.assign(paddle.less_than(x=j, y=array_len2), cond2)
         sum_result = layers.array_read(array=mem_array, i=j)
diff --git a/python/paddle/fluid/tests/unittests/xpu/test_while_op_xpu.py b/python/paddle/fluid/tests/unittests/xpu/test_while_op_xpu.py
index 2d3253295d0e7..d627b8fb35d72 100644
--- a/python/paddle/fluid/tests/unittests/xpu/test_while_op_xpu.py
+++ b/python/paddle/fluid/tests/unittests/xpu/test_while_op_xpu.py
@@ -62,7 +62,7 @@ def simple_net(self):
             prev = layers.array_read(array=mem_array, i=i)
             result = layers.sums(input=[d, prev])
 
-            i = paddle.increment(x=i, in_place=True)
+            i = paddle.increment(x=i)
             layers.array_write(result, i=i, array=mem_array)
             paddle.assign(paddle.less_than(x=i, y=array_len), cond)
 
@@ -71,7 +71,7 @@ def simple_net(self):
                 prev2 = layers.array_read(array=mem_array, i=j)
                 result2 = layers.sums(input=[d2, prev2])
 
-                j = paddle.increment(x=j, in_place=True)
+                j = paddle.increment(x=j)
                 layers.array_write(result2, i=j, array=mem_array)
                 paddle.assign(paddle.less_than(x=j, y=array_len2), cond2)
         sum_result = layers.array_read(array=mem_array, i=j)
diff --git a/python/paddle/jit/dy2static/convert_operators.py b/python/paddle/jit/dy2static/convert_operators.py
index 1a814086f5790..9e39c6df445f3 100644
--- a/python/paddle/jit/dy2static/convert_operators.py
+++ b/python/paddle/jit/dy2static/convert_operators.py
@@ -732,7 +732,7 @@ def convert_assert(cond, message=""):
     if isinstance(cond, Variable):
         cond = cast(cond, "bool")
         # NOTE: message is not used because Paddle Assert has no corresponding parameter to use.
-        from paddle.static.nn import Assert
+        from paddle.static.nn.control_flow import Assert
 
         return Assert(cond)
     else:
diff --git a/python/paddle/static/nn/control_flow.py b/python/paddle/static/nn/control_flow.py
index d331dd419fada..3b2ae2537cded 100644
--- a/python/paddle/static/nn/control_flow.py
+++ b/python/paddle/static/nn/control_flow.py
@@ -66,7 +66,7 @@ def Assert(cond, data=None, summarize=20, name=None):
         .. code-block:: python
 
             import paddle
-            from paddle.static.nn import Assert
+            from paddle.static.nn.control_flow import Assert
 
             paddle.enable_static()
             x = paddle.full([2, 3], 2.0, 'float32')