[Dy2Stat]Fix losting pre/post hook from outermost layer while jit.save (#42273) (#42388)

* [Dy2Stat]Fix losting pre/post hook from outermost layer while jit.save

* fix kwargs

* fix unittest

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

@@ -196,10 +196,11 @@ class CacheKey(object):
 
     def __hash__(self):
         error_msg = "Arguments to a `@paddle.jit.to_static` must be a hashable Python objects (or nested structures of these types)."
+        with_hook = self.kwargs.get("with_hook", False)
         return hash((id(self.function_spec),
                      make_hashable(self.input_args_with_spec, error_msg),
                      make_hashable(self.input_kwargs_with_spec, error_msg),
-                     self._spec_names_id, self.class_instance))
+                     self._spec_names_id, self.class_instance, with_hook))
 
     def __eq__(self, other):
         return (type(self) is type(other)) and hash(self) == hash(other)
@@ -413,6 +414,8 @@ class StaticFunction(object):
             Traced ConcreteProgram and executable translated Layer.
         """
 
+        with_hook = kwargs.get("with_hook", False)
+        if "with_hook" in kwargs: kwargs.pop("with_hook")
         # 1. unify args/kwargs and replace Tensor with InputSpec
         if len(args) != len(self._function_spec.args_name):
             args, kwargs = self._function_spec.unified_args_and_kwargs(args,
@@ -421,9 +424,13 @@ class StaticFunction(object):
             args, kwargs)
 
         # 2. generate cache key
-        cache_key = CacheKey(self._function_spec, input_args_with_spec,
-                             input_kwargs_with_spec, self._class_instance,
-                             **self._kwargs)
+        cache_key = CacheKey(
+            self._function_spec,
+            input_args_with_spec,
+            input_kwargs_with_spec,
+            self._class_instance,
+            **self._kwargs,
+            with_hook=with_hook)
 
         # 3. check whether hit the cache or build a new program for the input arguments
         concrete_program, partial_program_layer = self._program_cache[cache_key]
@@ -480,11 +487,13 @@ class StaticFunction(object):
         """
         return self.concrete_program_specify_input_spec(input_spec=None)
 
-    def concrete_program_specify_input_spec(self, input_spec=None):
+    def concrete_program_specify_input_spec(self,
+                                            input_spec=None,
+                                            with_hook=False):
         """
         Returns recent ConcreteProgram instance of decorated function while
         specifying input_spec. If the self._function_spec already has
-        input_spce, it will check the compatibility of input input_spec and
+        input_spec, it will check the compatibility of input input_spec and
         the self._function_spec.input_spec. If input input_spec=None, then
         this method uses self._function_spec.input_spec
 
@@ -516,12 +525,18 @@ class StaticFunction(object):
             has_input_spec = (desired_input_spec is not None)
             if has_input_spec:
                 concrete_program, _ = self.get_concrete_program(
-                    *desired_input_spec)
+                    *desired_input_spec, with_hook=with_hook)
                 return concrete_program
             else:
                 raise ValueError(
                     "No valid transformed program for {}.\n\t    Please specific `input_spec` in `@paddle.jit.to_static` or feed input tensor to call the decorated function at once.\n".
                     format(self._function_spec))
+        elif with_hook:
+            cache_key = self._program_cache._recent_cache_key
+            cache_key.kwargs["with_hook"] = True
+            concrete_program, _ = self._program_cache[cache_key]
+            return concrete_program
+
         # If more than one programs have been cached, return the recent converted program by default.
         elif cached_program_len > 1:
             logging_utils.warn(
@@ -588,6 +603,54 @@ def _verify_init_in_dynamic_mode(class_instance):
                     class_instance))
 
 
+class HookHelper(object):
+    """
+    Only For converting pre/post hooks operation in outermost layer while jit.save.
+    Because hooks in sublayer have been processed automatically.
+    """
+
+    def __init__(self, func, class_instance, with_hook=False):
+        self.func = func
+        self.class_instance = class_instance
+        self.with_hook = with_hook
+        self.need_apply_hook = with_hook and isinstance(
+            self.class_instance,
+            layers.Layer) and getattr(func, "__name__") == "forward"
+
+    def apply_pre_hooks(self, inputs):
+        """
+        Apply _forward_pre_hooks from outermost layer
+        """
+        if not self.need_apply_hook: return inputs
+
+        inputs = inputs[1:]
+        for forward_pre_hook in self.class_instance._forward_pre_hooks.values():
+            hook_result = forward_pre_hook(self.class_instance, inputs)
+            if hook_result is not None:
+                if not isinstance(hook_result, tuple):
+                    hook_result = (hook_result, )
+                inputs = hook_result
+
+        return [self.class_instance] + list(inputs)
+
+    def apply_post_hooks(self, inputs, outputs):
+        """
+        Apply _forward_post_hooks from outermost layer
+        """
+        if not self.need_apply_hook: return outputs
+
+        inputs = inputs[1:]
+        for forward_post_hook in self.class_instance._forward_post_hooks.values(
+        ):
+            hook_result = forward_post_hook(self.class_instance, inputs,
+                                            outputs)
+            if hook_result is not None:
+                outputs = hook_result
+
+        inputs.insert(0, self.class_instance)
+        return outputs
+
+
 class ConcreteProgram(object):
 
     __slots__ = [
@@ -629,6 +692,9 @@ class ConcreteProgram(object):
         # Transforms dygraph function into static function and caches it.
         dygraph_function = func_spec.dygraph_function
         static_func = convert_to_static(dygraph_function)
+        # apply pre\post hook for outermost layer
+        hook_helper = HookHelper(dygraph_function, class_instance,
+                                 kwargs.get("with_hook", False))
 
         main_program, startup_program = framework.Program(), framework.Program()
         # Note: The random seed should be synchronized into cached program
@@ -642,12 +708,13 @@ class ConcreteProgram(object):
         with framework.program_guard(main_program, startup_program):
             with _switch_declarative_mode_guard_(is_declarative=True):
                 # 1. Adds `fluid.data` layers for input if needed
-                inputs = func_spec.to_static_inputs_with_spec(input_spec,
-                                                              main_program)
+                static_inputs = func_spec.to_static_inputs_with_spec(
+                    input_spec, main_program)
                 _kwargs = func_spec.to_static_inputs_with_spec(
                     input_kwargs_spec, main_program)
                 if class_instance:
-                    inputs = tuple([class_instance] + list(inputs))
+                    static_inputs = tuple([class_instance] + list(
+                        static_inputs))
 
                 # 2. Gets all ParamBases and buffered VarBases in the function
                 all_parameters_and_buffers = _extract_indeed_params_buffers(
@@ -658,10 +725,13 @@ class ConcreteProgram(object):
                         class_instance, False)), param_guard(
                             get_buffers(class_instance, False)):
                     try:
+                        # only for jit.save, do nothing while train and eval process
+                        inputs = hook_helper.apply_pre_hooks(static_inputs)
                         if _kwargs:
                             outputs = static_func(*inputs, **_kwargs)
                         else:
                             outputs = static_func(*inputs)
+                        outputs = hook_helper.apply_post_hooks(inputs, outputs)
                     except BaseException as e:
                         # NOTE: If e is raised in compile time, e should be attached to ERROR_DATA here.
                         error.attach_error_data(e)
@@ -679,7 +749,7 @@ class ConcreteProgram(object):
         main_program = update_op_callstack_with_origin_info(main_program)
 
         return ConcreteProgram(
-            inputs=inputs,
+            inputs=static_inputs,
             outputs=outputs,
             parameters=all_parameters_and_buffers,
             function=dygraph_function,
@@ -709,6 +779,7 @@ class ProgramCache(object):
         self._caches = collections.OrderedDict()
         # trace mostly recent used program 
         self._recent_key = None
+        self._recent_cache_key = None
 
     def _build_once(self, cache_key):
         concrete_program = ConcreteProgram.from_func_spec(
@@ -724,6 +795,7 @@ class ProgramCache(object):
             raise ValueError('type(item) should be CacheKey, but received %s' %
                              type_name(item))
         item_id = hash(item)
+        self._recent_cache_key = item
         self._recent_key = item_id
         if item_id not in self._caches:
             self._caches[item_id] = self._build_once(item)

python/paddle/fluid/dygraph/jit.py

@@ -302,6 +302,7 @@ class _SaveLoadConfig(object):
 
         # If True, It will save inference program only, and do not save params of Program
         self._program_only = False
+        self.with_hook = False
 
     @property
     def output_spec(self):
@@ -370,7 +371,7 @@ class _SaveLoadConfig(object):
 
 
 def _parse_save_configs(configs):
-    supported_configs = ['output_spec']
+    supported_configs = ['output_spec', "with_hook"]
 
     # input check
     for key in configs:
@@ -382,6 +383,7 @@ def _parse_save_configs(configs):
     # construct inner config
     inner_config = _SaveLoadConfig()
     inner_config.output_spec = configs.get('output_spec', None)
+    inner_config.with_hook = configs.get('with_hook', False)
 
     return inner_config
 
@@ -454,11 +456,15 @@ def _get_input_var_names(inputs, input_spec):
     return result_list
 
 
-def _get_output_vars(outputs, output_spec):
+def _get_output_vars(outputs, output_spec, with_hook=False):
     name_no_exists_error = "The tensor `%s` does not exists. " \
         "Please make sure the name of example Tensor " \
         "in configs.output_spec is the output tensor of " \
         "Layer.forward method."
+    if output_spec and with_hook:
+        raise RuntimeError(
+            "Currently not support specify output_spec while founding pre/post hooks in your outermost layer."
+        )
     result_list = []
     output_vars_dict = OrderedDict()
     for var in flatten(outputs):
@@ -830,10 +836,16 @@ def save(layer, path, input_spec=None, **configs):
 
     # parse configs
     configs = _parse_save_configs(configs)
+    # whether outermost layer has pre/post hook, if does, we need also save
+    # these operators in program. 
+    with_hook = configs.with_hook
+
     scope = core.Scope()
     extra_var_info = dict()
     if isinstance(layer, Layer):
         functions = dir(inner_layer)
+        if inner_layer._forward_pre_hooks or inner_layer._forward_post_hooks:
+            with_hook = True
     else:
         # layer is function
         functions = [layer, ]
@@ -842,7 +854,7 @@ def save(layer, path, input_spec=None, **configs):
             static_func = getattr(inner_layer, attr_func, None)
             if isinstance(static_func, StaticFunction):
                 concrete_program = static_func.concrete_program_specify_input_spec(
-                    inner_input_spec)
+                    inner_input_spec, with_hook=with_hook)
             elif 'forward' == attr_func:
                 # transform in jit.save, if input_spec is incomplete, declarative will throw error
                 # inner_input_spec is list[InputSpec], it should be packed with same structure
@@ -852,7 +864,8 @@ def save(layer, path, input_spec=None, **configs):
                                                         inner_input_spec)
                 static_forward = declarative(
                     inner_layer.forward, input_spec=inner_input_spec)
-                concrete_program = static_forward.concrete_program
+                concrete_program = static_forward.concrete_program_specify_input_spec(
+                    with_hook=with_hook)
                 # the input_spec has been used in declarative, which is equal to
                 # @declarative with input_spec and jit.save without input_spec,
                 # avoid needless warning
@@ -943,8 +956,10 @@ def save(layer, path, input_spec=None, **configs):
         # the rule is like [ Get input variables name ]. For output var,
         # we only support VarBase spec, and actually, we only need the
         # var name of output, and we don't recommended to use output_spec
+        # print(concrete_program.main_program)
+        # print(concrete_program.outputs, configs.output_spec)
         output_vars = _get_output_vars(concrete_program.outputs,
-                                       configs.output_spec)
+                                       configs.output_spec, with_hook)
 
         # 5. save inference model
         from paddle.fluid.io import save_inference_model

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

+# 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.
+
+import unittest
+import paddle
+
+import numpy as np
+
+
+def forward_post_hook1(layer, input, output):
+    return output + output
+
+
+def forward_pre_hook1(layer, input):
+    input_return = (input[0] * 2, )
+    return input_return
+
+
+class SimpleNet(paddle.nn.Layer):
+    def __init__(self, ):
+        super(SimpleNet, self).__init__()
+        self.fc1 = paddle.nn.Linear(10, 10)
+        # sublayer1 register post hook
+        self.fc1.register_forward_post_hook(forward_post_hook1)
+
+        self.fc2 = paddle.nn.Linear(10, 10)
+        # sublayer2 register pre hook
+        self.fc2.register_forward_pre_hook(forward_pre_hook1)
+
+        # register pre/post hook
+        self.register_forward_pre_hook(forward_pre_hook1)
+        self.register_forward_post_hook(forward_post_hook1)
+
+    def forward(self, x):
+        x = self.fc1(x)
+        x = self.fc2(x)
+        out = paddle.mean(x)
+
+        return out
+
+
+class TestNestLayerHook(unittest.TestCase):
+    def setUp(self):
+        paddle.seed(2022)
+        self.x = paddle.randn([4, 10])
+        self.path = "./net_hook"
+
+    def train_net(self, to_static=False):
+        paddle.seed(2022)
+        net = SimpleNet()
+        if to_static:
+            net = paddle.jit.to_static(net)
+        out = net(self.x)
+
+        if to_static:
+            paddle.jit.save(net, self.path)
+
+        return out.numpy()[0]
+
+    def load_train(self):
+        net = paddle.jit.load(self.path)
+        out = net(self.x)
+        return out.numpy()[0]
+
+    def test_hook(self):
+        dy_out = self.train_net(to_static=False)
+        st_out = self.train_net(to_static=True)
+        load_out = self.load_train()
+        print(st_out, dy_out, load_out)
+        self.assertTrue(
+            np.allclose(st_out, dy_out),
+            msg='dygraph_res is {}\nstatic_res is {}'.format(dy_out, st_out))
+        self.assertTrue(
+            np.allclose(st_out, load_out),
+            msg='load_out is {}\nstatic_res is {}'.format(load_out, st_out))
+
+
+if __name__ == "__main__":
+    unittest.main()