Support list in control flow for dygraph_to_static (#22902)

* support list in control flow if. test=develop

* support list in for/while and supplement tests. test=develop

python/paddle/fluid/dygraph/dygraph_to_static/ast_transformer.py

@@ -30,6 +30,7 @@ from paddle.fluid.dygraph.dygraph_to_static.utils import ast_to_func
 from paddle.fluid.dygraph.dygraph_to_static.utils import is_paddle_api, is_dygraph_api, is_to_variable
 from paddle.fluid.dygraph.dygraph_to_static.utils import to_assign_node, to_static_ast, update_args_of_func
 from paddle.fluid.dygraph.dygraph_to_static.utils import dygraph_class_to_static_api, create_api_shape_node
+from paddle.fluid.dygraph.dygraph_to_static.list_transformer import ListTransformer
 from paddle.fluid.dygraph.dygraph_to_static.loop_transformer import LoopTransformer
 from paddle.fluid.dygraph.dygraph_to_static.ifelse_transformer import IfElseTransformer
 from paddle.fluid.dygraph.dygraph_to_static.static_analysis import AstNodeWrapper, NodeVarType
@@ -67,6 +68,7 @@ class DygraphToStaticAst(gast.NodeTransformer):
         basic_api_trans.ast_visit()
         self.feed_name_to_arg_name = basic_api_trans.get_feed_name_to_arg_id()
 
+        ListTransformer(node_wrapper).transform()
         # Transform all if/else statement of Dygraph into Static Graph.
         IfElseTransformer(node_wrapper).transform()
 

python/paddle/fluid/dygraph/dygraph_to_static/ifelse_transformer.py

@@ -424,7 +424,9 @@ def create_cond_node(return_name_ids, pred, true_func, false_func):
         func=cond_api,
         args=[pred, true_func_lambda, false_func_lambda],
         keywords=[])
+    if return_name_ids:
         targets = [generate_name_node(return_name_ids, ctx=gast.Store())]
         assign_node = gast.Assign(targets=targets, value=cond_layer)
-
         return assign_node
+    else:
+        return gast.Expr(value=cond_layer)

python/paddle/fluid/dygraph/dygraph_to_static/list_transformer.py

+#   Copyright (c) 2020 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 __future__ import print_function
+
+import gast
+import astor
+from paddle.fluid.dygraph.dygraph_to_static.static_analysis import AstNodeWrapper, NodeVarType, StaticAnalysisVisitor
+from paddle.fluid.dygraph.dygraph_to_static.utils import is_control_flow_to_transform
+
+
+class ListTransformer(gast.NodeTransformer):
+    """
+    This class transforms python list used in control flow into Static Graph Ast
+    """
+
+    def __init__(self, wrapper_root):
+        assert isinstance(
+            wrapper_root, AstNodeWrapper
+        ), "Input non-AstNodeWrapper node for the initialization of ListTransformer."
+        self.wrapper_root = wrapper_root
+        self.root = wrapper_root.node
+        self.name_of_list_set = set()
+        self.list_name_to_updated = dict()
+
+        self.static_analysis_visitor = StaticAnalysisVisitor(self.root)
+        self.node_to_wrapper_map = self.static_analysis_visitor.get_node_to_wrapper_map(
+        )
+        var_env = self.static_analysis_visitor.get_var_env()
+        var_env.cur_scope = var_env.cur_scope.sub_scopes[0]
+        self.scope_var_type_dict = var_env.get_scope_var_type()
+
+    def transform(self):
+        self.visit(self.root)
+        self.replace_list_with_tensor_array(self.root)
+
+    def visit_Assign(self, node):
+        self._update_list_name_to_updated(node)
+        return node
+
+    def visit_If(self, node):
+        self.generic_visit(node)
+        if is_control_flow_to_transform(node, self.scope_var_type_dict):
+            self._transform_list_append_in_control_flow(node)
+        return node
+
+    def visit_While(self, node):
+        self.generic_visit(node)
+        if is_control_flow_to_transform(node, self.scope_var_type_dict):
+            self._transform_list_append_in_control_flow(node)
+        return node
+
+    def visit_For(self, node):
+        self.generic_visit(node)
+        if is_control_flow_to_transform(node, self.scope_var_type_dict):
+            self._transform_list_append_in_control_flow(node)
+        return node
+
+    def replace_list_with_tensor_array(self, node):
+        for child_node in gast.walk(node):
+            if isinstance(child_node, gast.Assign):
+                if self._need_to_create_tensor_array(child_node):
+                    child_node.value = self._create_tensor_array()
+
+    def _transform_list_append_in_control_flow(self, node):
+        for child_node in gast.walk(node):
+            if self._need_to_array_write_node(child_node):
+                child_node.value = \
+                    self._to_array_write_node(child_node.value)
+
+    def _need_to_array_write_node(self, node):
+        if isinstance(node, gast.Expr):
+            if isinstance(node.value, gast.Call):
+                if self._is_list_append_tensor(node.value):
+                    return True
+
+        return False
+
+    def _is_list_append_tensor(self, node):
+        """
+        a.append(b): a is list, b is Tensor
+        self.x.append(b): self.x is list, b is Tensor
+        """
+        assert isinstance(node, gast.Call)
+        # 1. The func is `append`.
+        if not isinstance(node.func, gast.Attribute):
+            return False
+        if node.func.attr != 'append':
+            return False
+
+        # 2. It's a `python list` to call append().
+        value_name = astor.to_source(gast.gast_to_ast(node.func.value)).strip()
+        if value_name not in self.list_name_to_updated:
+            return False
+
+        # 3. The arg of append() is one `Tensor`
+        # Only one argument is supported in Python list.append()
+        if len(node.args) != 1:
+            return False
+        arg = node.args[0]
+        if isinstance(arg, gast.Name):
+            # TODO: `arg.id` may be not in scope_var_type_dict if `arg.id` is the arg of decorated function
+            # Need a better way to confirm whether `arg.id` is a Tensor.
+            try:
+                var_type_set = self.scope_var_type_dict[arg.id]
+            except KeyError:
+                return False
+
+            if NodeVarType.NUMPY_NDARRAY in var_type_set:
+                return False
+            if NodeVarType.TENSOR not in var_type_set and NodeVarType.PADDLE_RETURN_TYPES not in var_type_set:
+                return False
+        # else:
+        # Todo: Consider that `arg` may be a gast.Call about Paddle Api.
+        # eg: list_a.append(fluid.layers.reshape(x))
+        # return True
+        self.list_name_to_updated[value_name.strip()] = True
+        return True
+
+    def _need_to_create_tensor_array(self, node):
+        assert isinstance(node, gast.Assign)
+        target_node = node.targets[0]
+        try:
+            target_id = target_node.id
+        except AttributeError:
+            return False
+        if self.list_name_to_updated.get(target_id):
+            return True
+        return False
+
+    def _create_tensor_array(self):
+        # Although `dtype='float32'`, other types such as `int32` can also be supported
+        func_code = "fluid.layers.create_array(dtype='float32')"
+        func_node = gast.parse(func_code).body[0].value
+        return func_node
+
+    def _to_array_write_node(self, node):
+        assert isinstance(node, gast.Call)
+        array = astor.to_source(gast.gast_to_ast(node.func.value))
+        x = astor.to_source(gast.gast_to_ast(node.args[0]))
+        i = "fluid.layers.array_length({})".format(array)
+        func_code = "fluid.layers.array_write(x={}, i={}, array={})".format(
+            x, i, array)
+        return gast.parse(func_code).body[0].value
+
+    def _update_list_name_to_updated(self, node):
+        assert isinstance(node, gast.Assign)
+        target_node = node.targets[0]
+        # TODO: Consider node has more than one target. eg: x, y = a, []
+        try:
+            target_id = target_node.id
+        except AttributeError:
+            return False
+        value_node = node.value
+        if isinstance(value_node, gast.List):
+            self.list_name_to_updated[target_id] = False
+            return True
+        elif target_id in self.name_of_list_set:
+            del self.list_name_to_updated[target_id]
+        return False

python/paddle/fluid/dygraph/dygraph_to_static/utils.py

@@ -77,6 +77,29 @@ def is_numpy_api(node):
         return False
 
 
+def is_control_flow_to_transform(node, var_name_to_type):
+    """
+    Determines whether the node is a Paddle control flow statement which needs to
+    transform into a static graph control flow statement.
+    """
+    assert isinstance(node, gast.AST), \
+        "The type of input node must be gast.AST, but received %s." % type(node)
+
+    if isinstance(node, gast.If):
+        # TODO: make a better condition
+        return True
+
+    if isinstance(node, gast.For):
+        # TODO: make a better condition
+        return True
+
+    if isinstance(node, gast.While):
+        # TODO: make a better condition
+        return True
+
+    return False
+
+
 def _delete_keywords_from(node):
     assert isinstance(node, gast.Call)
     func_src = astor.to_source(gast.gast_to_ast(node.func))
@@ -255,6 +278,7 @@ def create_funcDef_node(nodes, name, input_args, return_name_ids):
     """
     nodes = copy.copy(nodes)
     # add return statement
+    if return_name_ids:
         nodes.append(gast.Return(value=generate_name_node(return_name_ids)))
     func_def_node = gast.FunctionDef(
         name=name,

python/paddle/fluid/tests/unittests/dygraph_to_static/test_list.py

+# Copyright (c) 2020 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 __future__ import print_function
+
+import unittest
+import numpy as np
+import paddle.fluid as fluid
+from paddle.fluid.dygraph.jit import dygraph_to_static_graph
+
+SEED = 2020
+np.random.seed(SEED)
+
+
+def test_list_without_control_flow(x):
+    # Python list will not be transformed.
+    x = fluid.dygraph.to_variable(x)
+    a = []
+    a.append(x)
+    return a
+
+
+def test_list_in_if(x):
+    x = fluid.dygraph.to_variable(x)
+    a = []
+    if x.numpy()[0] > 0:
+        a.append(x)
+    else:
+        a.append(
+            fluid.layers.fill_constant(
+                shape=[1, 2], value=9, dtype="int64"))
+    return a
+
+
+def test_list_in_for_loop(x, iter_num):
+    # Note: for_loop can't be transformed before PR22867 merged.
+    x = fluid.dygraph.to_variable(x)
+    a = []
+    for i in range(iter_num):
+        a.append(x)
+    return a
+
+
+def test_list_in_while_loop(x, iter_num):
+    x = fluid.dygraph.to_variable(x)
+    iter_num = fluid.layers.fill_constant(
+        shape=[1], value=iter_num, dtype="int32")
+    a = []
+    i = 0
+    # Note: `i < iter_num` can't be supported in dygraph mode now,
+    # but PR22892 is fixing it https://github.com/PaddlePaddle/Paddle/pull/22892.
+    # If PR22892 merged, change `i < iter_num.numpy()[0]` to `i < iter_num`.
+    while i < iter_num.numpy()[0]:
+        a.append(x)
+        i += 1
+    return a
+
+
+class TestListWithoutControlFlow(unittest.TestCase):
+    def setUp(self):
+        self.input = np.random.random((3)).astype('int32')
+        self.dygraph_func = test_list_without_control_flow
+        self.place = fluid.CUDAPlace(0) if fluid.is_compiled_with_cuda(
+        ) else fluid.CPUPlace()
+
+    def run_dygraph_mode(self):
+        with fluid.dygraph.guard():
+            res = self.dygraph_func(self.input)
+            if isinstance(res, (list, tuple)):
+                res = res[0]
+            return res.numpy()
+
+    def run_static_mode(self):
+        main_program = fluid.Program()
+        with fluid.program_guard(main_program):
+            tensor_list = dygraph_to_static_graph(self.dygraph_func)(self.input)
+        exe = fluid.Executor(self.place)
+        static_res = exe.run(main_program, fetch_list=tensor_list[0])
+
+        return static_res[0]
+
+    def test_transformed_static_result(self):
+        static_res = self.run_static_mode()
+        dygraph_res = self.run_dygraph_mode()
+        self.assertTrue(
+            np.allclose(dygraph_res, static_res),
+            msg='dygraph res is {}\nstatic_res is {}'.format(dygraph_res,
+                                                             static_res))
+
+
+class TestListInIf(TestListWithoutControlFlow):
+    def setUp(self):
+        self.input = np.random.random((3)).astype('int32')
+        self.dygraph_func = test_list_in_if
+        self.place = fluid.CUDAPlace(0) if fluid.is_compiled_with_cuda(
+        ) else fluid.CPUPlace()
+
+    def run_static_mode(self):
+        main_program = fluid.Program()
+        with fluid.program_guard(main_program):
+            tensor_array = dygraph_to_static_graph(self.dygraph_func)(
+                self.input)
+            static_out = fluid.layers.array_read(
+                tensor_array,
+                i=fluid.layers.fill_constant(
+                    shape=[1], value=0, dtype='int64'))
+        exe = fluid.Executor(self.place)
+        numpy_res = exe.run(main_program, fetch_list=static_out)
+        return numpy_res[0]
+
+
+class TestListInWhileLoop(unittest.TestCase):
+    def setUp(self):
+        self.iter_num = 3
+        self.input = np.random.random((3)).astype('int32')
+        self.dygraph_func = test_list_in_while_loop
+        self.place = fluid.CUDAPlace(0) if fluid.is_compiled_with_cuda(
+        ) else fluid.CPUPlace()
+
+    def run_dygraph_mode(self):
+        with fluid.dygraph.guard():
+            var_res = self.dygraph_func(self.input, self.iter_num)
+            numpy_res = [ele.numpy() for ele in var_res]
+            return numpy_res
+
+    def run_static_mode(self):
+        main_program = fluid.Program()
+        with fluid.program_guard(main_program):
+            tensor_array = dygraph_to_static_graph(self.dygraph_func)(
+                self.input, self.iter_num)
+            static_outs = []
+            for i in range(self.iter_num):
+                static_outs.append(
+                    fluid.layers.array_read(
+                        tensor_array,
+                        i=fluid.layers.fill_constant(
+                            shape=[1], value=i, dtype='int64')))
+        exe = fluid.Executor(self.place)
+        numpy_res = exe.run(main_program, fetch_list=static_outs)
+        return numpy_res
+
+    def test_transformed_static_result(self):
+        static_res = self.run_static_mode()
+        dygraph_res = self.run_dygraph_mode()
+        self.assertTrue(
+            np.array_equal(dygraph_res, static_res),
+            msg='dygraph res is {}\nstatic_res is {}'.format(dygraph_res,
+                                                             static_res))
+
+
+class TestListInForLoop(unittest.TestCase):
+    def setUp(self):
+        self.iter_num = 3
+        self.input = np.random.random((3)).astype('int32')
+        self.dygraph_func = test_list_in_for_loop
+        self.place = fluid.CUDAPlace(0) if fluid.is_compiled_with_cuda(
+        ) else fluid.CPUPlace()
+
+    def run_dygraph_mode(self):
+        with fluid.dygraph.guard():
+            var_res = self.dygraph_func(self.input, self.iter_num)
+            numpy_res = [ele.numpy() for ele in var_res]
+            return numpy_res
+
+    def run_static_mode(self):
+        main_program = fluid.Program()
+        with fluid.program_guard(main_program):
+            tensor_array = dygraph_to_static_graph(self.dygraph_func)(
+                self.input, self.iter_num)
+            static_outs = []
+            for i in range(self.iter_num):
+                static_outs.append(
+                    fluid.layers.array_read(
+                        tensor_array,
+                        i=fluid.layers.fill_constant(
+                            shape=[1], value=i, dtype='int64')))
+        exe = fluid.Executor(self.place)
+        numpy_res = exe.run(main_program, fetch_list=static_outs)
+        return numpy_res
+
+    def test_transformed_static_result(self):
+        static_res = self.run_static_mode()
+        dygraph_res = self.run_dygraph_mode()
+        self.assertTrue(
+            np.array_equal(dygraph_res, static_res),
+            msg='dygraph res is {}\nstatic_res is {}'.format(dygraph_res,
+                                                             static_res))
+
+
+if __name__ == '__main__':
+    unittest.main()