1/* Copyright 2017 The TensorFlow Authors. All Rights Reserved.
2
3Licensed under the Apache License, Version 2.0 (the "License");
4you may not use this file except in compliance with the License.
5You may obtain a copy of the License at
6
7    http://www.apache.org/licenses/LICENSE-2.0
8
9Unless required by applicable law or agreed to in writing, software
10distributed under the License is distributed on an "AS IS" BASIS,
11WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
12See the License for the specific language governing permissions and
13limitations under the License.
14==============================================================================*/
15
16
17%include "tensorflow/python/platform/base.i"
18%include "cluster.i"
19
20%typemap(in) const tensorflow::MetaGraphDef& (tensorflow::MetaGraphDef temp) {
21  char* c_string;
22  Py_ssize_t py_size;
23  if (PyBytes_AsStringAndSize($input, &c_string, &py_size) == -1) {
24    // Python has raised an error (likely TypeError or UnicodeEncodeError).
25    SWIG_fail;
26  }
27
28  if (!temp.ParseFromString(string(c_string, py_size))) {
29    PyErr_SetString(
30        PyExc_TypeError,
31        "The MetaGraphDef could not be parsed as a valid protocol buffer");
32    SWIG_fail;
33  }
34  $1 = &temp;
35}
36
37%typemap(in) const tensorflow::RewriterConfig& (
38    tensorflow::RewriterConfig temp) {
39  char* c_string;
40  Py_ssize_t py_size;
41  if (PyBytes_AsStringAndSize($input, &c_string, &py_size) == -1) {
42    // Python has raised an error (likely TypeError or UnicodeEncodeError).
43    SWIG_fail;
44  }
45
46  if (!temp.ParseFromString(string(c_string, py_size))) {
47    PyErr_SetString(
48        PyExc_TypeError,
49        "The RewriterConfig could not be parsed as a valid protocol buffer");
50    SWIG_fail;
51  }
52  $1 = &temp;
53}
54
55%{
56  #include <memory>
57  #include "tensorflow/c/tf_status_helper.h"
58  #include "tensorflow/core/lib/core/status.h"
59  #include "tensorflow/core/common_runtime/device.h"
60  #include "tensorflow/core/framework/device_base.h"
61  #include "tensorflow/core/common_runtime/device_factory.h"
62  #include "tensorflow/core/framework/device_attributes.pb.h"
63  #include "tensorflow/core/framework/graph.pb.h"
64  #include "tensorflow/core/grappler/grappler_item.h"
65  #include "tensorflow/core/grappler/grappler_item_builder.h"
66  #include "tensorflow/core/grappler/clusters/cluster.h"
67  #include "tensorflow/core/grappler/clusters/utils.h"
68  #include "tensorflow/core/grappler/clusters/virtual_cluster.h"
69  #include "tensorflow/core/grappler/optimizers/meta_optimizer.h"
70  #include "tensorflow/core/protobuf/meta_graph.pb.h"
71  #include "tensorflow/core/protobuf/rewriter_config.pb.h"
72  #include "tensorflow/core/public/session_options.h"
73
74
75void DetectDevices(std::unordered_map<string, tensorflow::DeviceProperties>* device_map) {
76  tensorflow::SessionOptions options;
77  std::vector<tensorflow::Device*> devices;
78  tensorflow::Status status = tensorflow::DeviceFactory::AddDevices(options, "", &devices);
79  if (!status.ok()) {
80    return;
81  }
82
83  for (const tensorflow::Device* device : devices) {
84    tensorflow::DeviceProperties& prop = (*device_map)[device->name()];
85    prop = tensorflow::grappler::GetDeviceInfo(device->parsed_name());
86
87    // Overwrite the memory limit since users might have requested to use only a fraction of the
88    // available device memory.
89    const tensorflow::DeviceAttributes& attr = device->attributes();
90    prop.set_memory_size(attr.memory_limit());
91    delete device;
92  }
93}
94
95PyObject* TF_OptimizeGraph(
96      GCluster cluster,
97      const tensorflow::RewriterConfig& rewriter_config,
98      const tensorflow::MetaGraphDef& metagraph,
99      bool verbose, const string& graph_id, TF_Status* out_status) {
100    tensorflow::grappler::ItemConfig item_config;
101    item_config.inline_functions = false;
102    item_config.apply_optimizations = false;
103    std::unique_ptr<tensorflow::grappler::GrapplerItem> grappler_item =
104        tensorflow::grappler::GrapplerItemFromMetaGraphDef(graph_id, metagraph, item_config);
105
106    if (!grappler_item) {
107      TF_SetStatus(out_status, TF_INVALID_ARGUMENT, "Failed to import metagraph, check error log for more info.");
108      return nullptr;
109    }
110
111    tensorflow::DeviceBase* cpu_device = nullptr;
112    tensorflow::GraphDef out_graph;
113    tensorflow::grappler::MetaOptimizer optimizer(cpu_device, rewriter_config);
114    tensorflow::Status status = optimizer.Optimize(cluster.get(), *grappler_item, &out_graph);
115    if (verbose) {
116      optimizer.PrintResult();
117    }
118    tensorflow::Set_TF_Status_from_Status(out_status, status);
119    string out_graph_str = out_graph.SerializeAsString();
120    PyObject* ret = PyBytes_FromStringAndSize(out_graph_str.data(),
121                                              out_graph_str.size());
122    return ret;
123  }
124%}
125
126
127// Wrap this function
128PyObject* TF_OptimizeGraph(
129    GCluster cluster,
130    const tensorflow::RewriterConfig& rewriter_config,
131    const tensorflow::MetaGraphDef& metagraph, bool verbose,
132    const string& graph_id, TF_Status* out_status);
133
134
135
136