1 | /* Copyright 2017 The TensorFlow Authors. All Rights Reserved. |
2 | |
3 | Licensed under the Apache License, Version 2.0 (the "License"); |
4 | you may not use this file except in compliance with the License. |
5 | You may obtain a copy of the License at |
6 | |
7 | http://www.apache.org/licenses/LICENSE-2.0 |
8 | |
9 | Unless required by applicable law or agreed to in writing, software |
10 | distributed under the License is distributed on an "AS IS" BASIS, |
11 | WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. |
12 | See the License for the specific language governing permissions and |
13 | limitations under the License. |
14 | ==============================================================================*/ |
15 | |
16 | #ifndef TENSORFLOW_C_PYTHON_API_H_ |
17 | #define TENSORFLOW_C_PYTHON_API_H_ |
18 | |
19 | #include <string> |
20 | |
21 | #include "tensorflow/c/c_api.h" |
22 | #include "tensorflow/core/framework/full_type.pb.h" |
23 | |
24 | // These functions can be removed without notice. They exist to facilitate some |
25 | // refactoring of graph construction code in the Python API. |
26 | |
27 | namespace tensorflow { |
28 | |
29 | void AddControlInput(TF_Graph* graph, TF_Operation* op, TF_Operation* input); |
30 | |
31 | // Changes an attr value in the node_def Protocol Buffer and sets a status upon |
32 | // completion. |
33 | void SetAttr(TF_Graph* graph, TF_Operation* op, const char* attr_name, |
34 | TF_Buffer* attr_value_proto, TF_Status* status); |
35 | |
36 | // Clears the attr in the node_def Protocol Buffer and sets a status upon |
37 | // completion. |
38 | void ClearAttr(TF_Graph* graph, TF_Operation* op, const char* attr_name, |
39 | TF_Status* status); |
40 | |
41 | // Sets the experimental_type` field in the node_def Protocol Buffer. |
42 | void SetFullType(TF_Graph* graph, TF_Operation* op, |
43 | const FullTypeDef& full_type); |
44 | |
45 | void SetRequestedDevice(TF_Graph* graph, TF_Operation* op, const char* device); |
46 | |
47 | // Updates 'dst' to consume 'new_src'. |
48 | void UpdateEdge(TF_Graph* graph, TF_Output new_src, TF_Input dst, |
49 | TF_Status* status); |
50 | |
51 | void RemoveAllControlInputs(TF_Graph* graph, TF_Operation* op); |
52 | |
53 | // Sets whether ops missing a shape inference function should trigger an |
54 | // error. The default is true. |
55 | void SetRequireShapeInferenceFns(TF_Graph* graph, bool require); |
56 | |
57 | // Extends `session` with any new operations added to its associated graph. |
58 | // Usually this happens automatically in TF_SessionRun. After this is called, |
59 | // TF_SessionRun will no longer extend the session on every call. |
60 | // |
61 | // We expose this here to allow fine-grained synchronization in multi-threaded |
62 | // workloads, which is required since the Python implementation depends on the |
63 | // above mutation methods. This allows us to prevent modifications to nodes in |
64 | // the graph after the session has been made aware of them. |
65 | void ExtendSession(TF_Session* session, TF_Status* status); |
66 | |
67 | // Returns the serialized CppShapeInferenceResult::HandleData proto for |
68 | // `output` if its a resource or variant tensor, or otherwise returns the empty |
69 | // string. |
70 | std::string GetHandleShapeAndType(TF_Graph* graph, TF_Output output); |
71 | |
72 | // Sets `output` based on `proto`, which should be a serialized |
73 | // CppShapeInferenceResult::HandleData proto. `output` should be a resource |
74 | // or variant tensor. |
75 | // NOTE(skyewm): `proto` is passed a void*/size_t pair instead of a std::string |
76 | // because I couldn't get SWIG to work otherwise. |
77 | void SetHandleShapeAndType(TF_Graph* graph, TF_Output output, const void* proto, |
78 | size_t proto_len, TF_Status* status); |
79 | |
80 | // This method is used to add a new input edge to 'dst', which must be a While |
81 | // op. The While op's "T" attribute must have already been updated to include |
82 | // the new edge. This is used to construct tf.while_loop gradients. |
83 | void AddWhileInputHack(TF_Graph* graph, TF_Output new_src, TF_Operation* dst, |
84 | TF_Status* status); |
85 | |
86 | } // namespace tensorflow |
87 | |
88 | #endif // TENSORFLOW_C_PYTHON_API_H_ |
89 | |