Hidden ParallelDo. (#13454)

paddle/fluid/API.spec

@@ -237,12 +237,6 @@ paddle.fluid.layers.StaticRNN.step_input ArgSpec(args=['self', 'x'], varargs=Non
 paddle.fluid.layers.StaticRNN.step_output ArgSpec(args=['self', 'o'], varargs=None, keywords=None, defaults=None)
 paddle.fluid.layers.StaticRNN.update_memory ArgSpec(args=['self', 'mem', 'var'], varargs=None, keywords=None, defaults=None)
 paddle.fluid.layers.reorder_lod_tensor_by_rank ArgSpec(args=['x', 'rank_table'], varargs=None, keywords=None, defaults=None)
-paddle.fluid.layers.ParallelDo.__init__ ArgSpec(args=['self', 'places', 'use_nccl', 'name'], varargs=None, keywords=None, defaults=(False, None))
-paddle.fluid.layers.ParallelDo.do ArgSpec(args=['self'], varargs=None, keywords=None, defaults=None)
-paddle.fluid.layers.ParallelDo.get_parameters ArgSpec(args=['self'], varargs=None, keywords=None, defaults=None)
-paddle.fluid.layers.ParallelDo.parent_block ArgSpec(args=['self'], varargs=None, keywords=None, defaults=None)
-paddle.fluid.layers.ParallelDo.read_input ArgSpec(args=['self', 'var'], varargs=None, keywords=None, defaults=None)
-paddle.fluid.layers.ParallelDo.write_output ArgSpec(args=['self', 'var'], varargs=None, keywords=None, defaults=None)
 paddle.fluid.layers.Print ArgSpec(args=['input', 'first_n', 'message', 'summarize', 'print_tensor_name', 'print_tensor_type', 'print_tensor_shape', 'print_tensor_lod', 'print_phase'], varargs=None, keywords=None, defaults=(-1, None, -1, True, True, True, True, 'both'))
 paddle.fluid.layers.is_empty ArgSpec(args=['x', 'cond'], varargs=None, keywords='ignored', defaults=(None,))
 paddle.fluid.layers.mean ArgSpec(args=[], varargs='args', keywords='kwargs', defaults=None)

python/paddle/fluid/layers/control_flow.py

@@ -41,7 +41,6 @@ __all__ = [
     'DynamicRNN',
     'StaticRNN',
     'reorder_lod_tensor_by_rank',
-    'ParallelDo',
     'Print',
     'is_empty',
 ]
@@ -259,7 +258,7 @@ class ParallelDo(object):
       # ParallelDo version & Single-thread version
       if thread_num > 1:
           places = fluid.layers.get_places(thread_num)
-          pd = fluid.layers.ParallelDo(places)
+          pd = fluid.layers.control_flow.ParallelDo(places)
           with pd.do():
               images = pd.read_input(images)
               label = pd.read_input(label)

python/paddle/fluid/tests/book/notest_understand_sentiment.py

@@ -15,6 +15,7 @@
 from __future__ import print_function
 
 from paddle.fluid.layers.device import get_places
+from paddle.fluid.layers.control_flow import ParallelDo
 import unittest
 import paddle.fluid as fluid
 import paddle
@@ -147,7 +148,7 @@ def train(word_dict,
             data, label, input_dim=dict_dim, class_dim=class_dim)
     else:
         places = get_places()
-        pd = fluid.layers.ParallelDo(places)
+        pd = ParallelDo(places)
         with pd.do():
             cost, acc, _ = net_method(
                 pd.read_input(data),

python/paddle/fluid/tests/book/test_recognize_digits.py

@@ -25,6 +25,7 @@ import numpy
 import paddle
 import paddle.fluid as fluid
 from paddle.fluid.layers.device import get_places
+from paddle.fluid.layers.control_flow import ParallelDo
 
 BATCH_SIZE = 64
 
@@ -81,7 +82,7 @@ def train(nn_type,
 
     if parallel:
         places = get_places()
-        pd = fluid.layers.ParallelDo(places)
+        pd = ParallelDo(places)
         with pd.do():
             img_ = pd.read_input(img)
             label_ = pd.read_input(label)

python/paddle/fluid/tests/book/test_word2vec.py

@@ -17,6 +17,7 @@ from __future__ import print_function
 import paddle
 import paddle.fluid as fluid
 from paddle.fluid.layers.device import get_places
+from paddle.fluid.layers.control_flow import ParallelDo
 import unittest
 import os
 import numpy as np
@@ -84,7 +85,7 @@ def train(use_cuda, is_sparse, is_parallel, save_dirname, is_local=True):
             [first_word, second_word, third_word, forth_word, next_word])
     else:
         places = get_places()
-        pd = fluid.layers.ParallelDo(places)
+        pd = ParallelDo(places)
         with pd.do():
             avg_cost, predict_word = __network__(
                 list(

python/paddle/fluid/tests/book_memory_optimization/test_memopt_fit_a_line.py

@@ -20,6 +20,7 @@ import sys
 import paddle
 import paddle.fluid as fluid
 from paddle.fluid.layers.device import get_places
+from paddle.fluid.layers.control_flow import ParallelDo
 
 # need to fix random seed and training data to compare the loss
 # value accurately calculated by the default and the memory optimization
@@ -38,7 +39,7 @@ if fluid.core.is_compiled_with_cuda():
     place = fluid.CUDAPlace(0)
 
 places = get_places(device_count=0, device_type=device_type)
-pd = fluid.layers.ParallelDo(places, use_nccl=use_nccl)
+pd = ParallelDo(places, use_nccl=use_nccl)
 with pd.do():
     x_ = pd.read_input(x)
     y_ = pd.read_input(y)

python/paddle/fluid/tests/unittests/test_parallel_op.py

@@ -18,6 +18,7 @@ import unittest
 
 import paddle.fluid as fluid
 from paddle.fluid.layers.device import get_places
+from paddle.fluid.layers.control_flow import ParallelDo
 import paddle.fluid.profiler as profiler
 import numpy
 import six
@@ -120,7 +121,7 @@ class BaseParallelForTest(unittest.TestCase):
                 thread_num = fluid.core.get_cuda_device_count(
                 ) if use_gpu else 8
                 places = get_places(thread_num)
-                pd = fluid.layers.ParallelDo(places, use_nccl=use_nccl)
+                pd = ParallelDo(places, use_nccl=use_nccl)
                 data = next(generator)
 
                 if isinstance(data, fluid.framework.Variable):