1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
|
/* Copyright 2017 The TensorFlow 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.
==============================================================================*/
#include "tensorflow/core/grappler/clusters/single_machine.h"
#include "tensorflow/cc/framework/scope.h"
#include "tensorflow/cc/ops/standard_ops.h"
#include "tensorflow/core/framework/cost_graph.pb.h"
#include "tensorflow/core/framework/step_stats.pb.h"
#include "tensorflow/core/grappler/grappler_item.h"
#include "tensorflow/core/grappler/inputs/trivial_test_graph_input_yielder.h"
#include "tensorflow/core/grappler/utils.h"
#include "tensorflow/core/platform/protobuf.h"
#include "tensorflow/core/platform/test.h"
namespace tensorflow {
namespace grappler {
namespace {
class SingleMachineTest : public ::testing::Test {
public:
void SetUp() override {
// Provision a single machine with 3 cpu cores
cluster_.reset(new SingleMachine(5 * 60, 3, 0));
TF_CHECK_OK(cluster_->Provision());
}
void TearDown() override {
cluster_.reset();
}
protected:
std::unique_ptr<SingleMachine> cluster_;
};
TEST_F(SingleMachineTest, CostModel) {
TrivialTestGraphInputYielder fake_input(4, 1, 10, false,
cluster_->GetDeviceNames());
GrapplerItem item;
CHECK(fake_input.NextItem(&item));
TF_CHECK_OK(cluster_->Initialize(item));
RunMetadata metadata;
const int64 start_micros = Env::Default()->NowMicros();
TF_CHECK_OK(cluster_->Run(item.graph, item.feed, item.fetch, &metadata));
const int64 run_duration_micros = Env::Default()->NowMicros() - start_micros;
// There should be at least 4 nodes corresponding to the 4 stages we created
// in the fake input.
EXPECT_LE(4, metadata.cost_graph().node_size());
for (const auto& node : metadata.cost_graph().node()) {
// Skip the special nodes inserted by TF: these are prefixed with an
// underscore.
if (node.name()[0] == '_' || node.name().find("/_") != string::npos) {
continue;
}
EXPECT_EQ(1, node.output_info_size());
EXPECT_LE(8, node.output_info(0).size());
const TensorShapeProto& shape = node.output_info(0).shape();
EXPECT_EQ(2, shape.dim_size());
EXPECT_EQ(10, shape.dim(0).size());
EXPECT_EQ(1, shape.dim(1).size());
EXPECT_LE(0, node.compute_cost());
EXPECT_GE(run_duration_micros, node.compute_cost());
}
}
TEST_F(SingleMachineTest, Queue) {
TrivialTestGraphInputYielder fake_input(4, 1, 10, true,
cluster_->GetDeviceNames());
GrapplerItem item;
CHECK(fake_input.NextItem(&item));
TF_CHECK_OK(cluster_->Initialize(item));
RunMetadata metadata;
TF_CHECK_OK(cluster_->Run(item.graph, item.feed, item.fetch, &metadata));
}
TEST_F(SingleMachineTest, MultipleItems) {
TrivialTestGraphInputYielder fake_input(4, 1, 10, false,
cluster_->GetDeviceNames());
for (int i = 0; i < 3; ++i) {
GrapplerItem item;
CHECK(fake_input.NextItem(&item));
TF_CHECK_OK(cluster_->Initialize(item));
RunMetadata metadata1;
TF_CHECK_OK(cluster_->Run(item.graph, item.feed, item.fetch, &metadata1));
RunMetadata metadata2;
TF_CHECK_OK(cluster_->Run(item.graph, item.feed, item.fetch, &metadata2));
// There should be at least 4 nodes corresponding to the 4 stages we created
// in the fake input, plus 1 enqueue and 1 dequeue node.
EXPECT_LE(6, metadata1.cost_graph().node_size());
for (const auto& node : metadata1.cost_graph().node()) {
if (node.name()[0] == '_' || node.name().find("/_") != string::npos ||
node.name() == "queue") {
continue;
}
EXPECT_EQ(1, node.output_info_size());
const TensorShapeProto& shape = node.output_info(0).shape();
EXPECT_EQ(2, shape.dim_size());
EXPECT_EQ(10, shape.dim(0).size());
EXPECT_EQ(1, shape.dim(1).size());
}
for (int i = 0; i < metadata1.cost_graph().node_size(); ++i) {
metadata1.mutable_cost_graph()->mutable_node(i)->set_compute_cost(0);
metadata1.clear_step_stats();
}
for (int i = 0; i < metadata2.cost_graph().node_size(); ++i) {
metadata2.mutable_cost_graph()->mutable_node(i)->set_compute_cost(0);
metadata2.clear_step_stats();
}
string s1;
::tensorflow::protobuf::TextFormat::PrintToString(metadata1, &s1);
string s2;
::tensorflow::protobuf::TextFormat::PrintToString(metadata2, &s2);
EXPECT_EQ(s1, s2);
}
}
TEST_F(SingleMachineTest, InitializationMemory) {
// Build a variable and its initialization graph.
tensorflow::Scope s = tensorflow::Scope::NewRootScope();
int batch_size = 10;
Output x =
ops::RandomNormal(s.WithOpName("x"), {batch_size, 1}, DataType::DT_FLOAT);
Output v = ops::Variable(s.WithOpName("v"), TensorShape({batch_size, 1}),
DataType::DT_FLOAT);
Output init = ops::Assign(s.WithOpName("init"), v, x);
GrapplerItem item;
TF_CHECK_OK(s.ToGraphDef(&item.graph));
item.init_ops.push_back(init.name());
item.fetch.push_back(v.name());
TF_CHECK_OK(cluster_->Initialize(item));
RunMetadata metadata;
TF_CHECK_OK(cluster_->Run(item.graph, item.feed, item.fetch, &metadata));
// Check that the initialization op is present in the cost model.
bool found = false;
for (const auto& node : metadata.cost_graph().node()) {
found |= (node.name() == NodeName(init.name()));
}
EXPECT_TRUE(found);
}
} // namespace
} // namespace grappler
} // namespace tensorflow
|
