support to modify dict and list in while_loop (#23083)

* support to modify dict and list in while_loop test=develop

* polish code test=develop

python/paddle/fluid/layers/control_flow.py

@@ -21,7 +21,7 @@ from .. import core
 from ..framework import Program, Variable, Operator, in_dygraph_mode
 from ..layer_helper import LayerHelper, unique_name
 from .nn import logical_and, logical_not, logical_or
-from .utils import assert_same_structure, map_structure
+from .utils import assert_same_structure, map_structure, hold_mutable_vars, copy_mutable_vars
 import numpy
 import warnings
 import six
@@ -1018,7 +1018,16 @@ def while_loop(cond, body, loop_vars, is_test=False, name=None):
         return loop_vars
 
     while_loop_block = While(pre_cond, is_test, name)
+    has_mutable_vars_in_loop = hold_mutable_vars(loop_vars)
     with while_loop_block.block():
+        # If a variable with mutable type is included in loop_vars, like `dict/list`,
+        # modifying it in the body function will cause origin variable to be modified
+        # synchronously. This will raise an assignment error out of while block.
+        # Here we make a copy of the mutable vars to avoid this problem.
+        if has_mutable_vars_in_loop:
+            new_loop_vars = copy_mutable_vars(loop_vars)
+            output_vars = body(*new_loop_vars)
+        else:
             output_vars = body(*loop_vars)
         if not isinstance(output_vars, (list, tuple)):
             output_vars = [output_vars]

python/paddle/fluid/layers/utils.py

@@ -14,6 +14,7 @@
 
 from __future__ import print_function
 import collections
+import copy
 import six
 import numpy as np
 from ..framework import Variable
@@ -187,6 +188,24 @@ def map_structure(func, *structure):
     return pack_sequence_as(structure[0], [func(*x) for x in entries])
 
 
+def hold_mutable_vars(structure):
+    """
+    Returns whether structure holds sequence like `list/dict`.
+    """
+    for s in structure:
+        if is_sequence(s):
+            return True
+    return False
+
+
+def copy_mutable_vars(structure):
+    """
+    Returns vars copied from sequence without mutable property.
+    """
+    flat_structure = copy.copy(flatten(structure))
+    return pack_sequence_as(structure, flat_structure)
+
+
 def _recursive_assert_same_structure(nest1, nest2, check_types):
     """
     Helper function for `assert_same_structure`.

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

@@ -59,17 +59,7 @@ class SubNetWithDict(fluid.dygraph.Layer):
             cache_k, cache_v = cache["k"], cache["v"]
             k = 0.1 * cache_k + k
             v = 0.2 * cache_v + v
-            # TODO: currently while_loop can have a dict as loop_vars, but
+            cache["k"], cache["v"] = k, v
-            # to change the value in a dict, you have to use layers.assign
-            # because cache["k"] = k is putting k in dict without building
-            # network. So we cannot write:
-            #
-            # cache["k"], cache["v"] = k, v
-            #
-            # we have to support this kind of dict in loop in the future.
-            # For example, automatically change = to assign in AutoTracer
-            fluid.layers.assign(k, cache["k"])
-            fluid.layers.assign(v, cache["v"])
 
         weight = fluid.layers.matmul(x=q, y=k, transpose_y=True)
         weight = fluid.layers.softmax(weight)
@@ -108,16 +98,7 @@ class MainNetWithDict(fluid.dygraph.Layer):
 
     def update_cache(self, cache):
         for k, val in six.iteritems(cache):
-            # TODO: currently while_loop can have a dict as loop_vars, but
+            cache[k] = fluid.layers.softmax(val)
-            # to change the value in a dict, you have to use layers.assign
-            # because cache["k"] = k is putting k in dict without building
-            # network. So we cannot write:
-            #
-            # cache[k] = fluid.layers.softmax(val)
-            #
-            # we have to support this kind of dict in loop in the future.
-            # For example, automatically change = to assign in AutoTracer
-            fluid.layers.assign(fluid.layers.softmax(val), cache[k])
 
         return cache
 

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

@@ -78,13 +78,21 @@ class TestApiWhileLoop(unittest.TestCase):
         self.assertTrue(np.allclose(np.asarray(res[1]), data))
 
     def test_var_dict(self):
-        def cond(i, ten, test_dict):
+        def cond(i, ten, test_dict, test_list, test_list_dict):
             return layers.less_than(i, ten)
 
-        def body(i, ten, test_dict):
+        def body(i, ten, test_dict, test_list, test_list_dict):
-            layers.assign(i, test_dict["test_key"])
+            test_dict["test_key"] = i
+            test_dict["test_key"] += 1
+
+            test_list[0] = fluid.layers.reshape(test_list[0], [2, -1]) + 1
+
+            test_list_dict[0]["test_key"] += 1
+            test_list_dict[0]["test_key"] = fluid.layers.relu(test_list_dict[0][
+                "test_key"])
+
             i = layers.increment(i)
-            return [i, ten, test_dict]
+            return [i, ten, test_dict, test_list, test_list_dict]
 
         main_program = Program()
         startup_program = Program()
@@ -92,18 +100,42 @@ class TestApiWhileLoop(unittest.TestCase):
             i = layers.zeros(shape=[1], dtype='int64')
             ten = layers.fill_constant(shape=[1], dtype='int64', value=10)
             test_data = layers.fill_constant(shape=[1], dtype='int64', value=0)
+
             test_dict = {"test_key": test_data}
-            i, ten, test_dict = layers.while_loop(cond, body,
+            test_list = [
-                                                  [i, ten, test_dict])
+                layers.fill_constant(
+                    shape=[1, 2], dtype='int64', value=0)
+            ]
+            test_list_dict = [{
+                "test_key": layers.fill_constant(
+                    shape=[1], dtype='float32', value=0)
+            }]
+
+            i, ten, test_dict, test_list, test_list_dict = layers.while_loop(
+                cond, body, [i, ten, test_dict, test_list, test_list_dict])
         place = fluid.CUDAPlace(0) if core.is_compiled_with_cuda(
         ) else fluid.CPUPlace()
         exe = fluid.Executor(place)
-        res = exe.run(main_program, fetch_list=[test_dict["test_key"]])
+        res = exe.run(main_program,
+                      fetch_list=[
+                          test_dict["test_key"], test_list[0],
+                          test_list_dict[0]["test_key"]
+                      ])
         self.assertTrue(
             np.allclose(
                 np.asarray(res[0]),
                 np.full(
-                    shape=(1), fill_value=9, dtype=np.int64)))
+                    shape=(1), fill_value=10, dtype=np.int64)))
+        self.assertTrue(
+            np.allclose(
+                np.asarray(res[1]),
+                np.full(
+                    shape=(2, 1), fill_value=10, dtype=np.int64)))
+        self.assertTrue(
+            np.allclose(
+                np.asarray(res[2]),
+                np.full(
+                    shape=(1), fill_value=10, dtype=np.float32)))
 
 
 class TestApiWhileLoop_Nested(unittest.TestCase):