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
165
166
|
/* Copyright 2015 Google Inc. 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/common_runtime/kernel_benchmark_testlib.h"
#include <vector>
#include "tensorflow/core/common_runtime/device.h"
#include "tensorflow/core/common_runtime/device_factory.h"
#include "tensorflow/core/framework/op.h"
#include "tensorflow/core/framework/op_kernel.h"
#include "tensorflow/core/framework/op_segment.h"
#include "tensorflow/core/graph/graph.h"
#include "tensorflow/core/kernels/ops_util.h"
#include "tensorflow/core/lib/core/notification.h"
#include "tensorflow/core/lib/core/threadpool.h"
#include "tensorflow/core/lib/strings/str_util.h"
#include "tensorflow/core/platform/host_info.h"
#include "tensorflow/core/platform/logging.h"
#include "tensorflow/core/platform/test_benchmark.h"
#include "tensorflow/core/platform/types.h"
#include "tensorflow/core/public/session_options.h"
#include "tensorflow/core/public/version.h"
#include "tensorflow/core/util/device_name_utils.h"
namespace tensorflow {
namespace test {
Benchmark::Benchmark(const string& device, Graph* g,
const SessionOptions* options, Graph* init) {
RequireDefaultOps();
SessionOptions default_options;
if (!options) {
options = &default_options;
}
testing::StopTiming();
string t = str_util::Uppercase(device);
device_ =
DeviceFactory::NewDevice(t, *options, "/job:localhost/replica:0/task:0");
CHECK(device_) << "Could not create a " << device << " device";
pool_ = new thread::ThreadPool(options->env, "blocking",
port::NumSchedulableCPUs());
auto runner = [this](std::function<void()> closure) {
pool_->Schedule(closure);
};
rendez_ = NewLocalRendezvous();
const int graph_def_version = g->version();
LocalExecutorParams params;
params.device = device_;
params.function_library = nullptr;
params.create_kernel = [this, graph_def_version](const NodeDef& ndef,
OpKernel** kernel) {
return CreateNonCachedKernel(device_, nullptr, ndef, graph_def_version,
kernel);
};
params.delete_kernel = [](OpKernel* kernel) {
DeleteNonCachedKernel(kernel);
};
if (init) {
Executor* init_exec;
TF_CHECK_OK(NewLocalExecutor(params, init, &init_exec));
Executor::Args args;
args.rendezvous = rendez_;
args.runner = runner;
TF_CHECK_OK(init_exec->Run(args));
delete init_exec;
}
TF_CHECK_OK(NewLocalExecutor(params, g, &exec_));
}
Benchmark::~Benchmark() {
if (device_) {
rendez_->Unref();
delete exec_;
delete device_;
delete pool_;
}
}
void Benchmark::Run(int iters) { RunWithArgs({}, {}, iters); }
string GetRendezvousKey(const Node* node) {
string send_device;
TF_CHECK_OK(GetNodeAttr(node->def(), "send_device", &send_device));
string recv_device;
TF_CHECK_OK(GetNodeAttr(node->def(), "recv_device", &recv_device));
string tensor_name;
TF_CHECK_OK(GetNodeAttr(node->def(), "tensor_name", &tensor_name));
uint64 send_device_incarnation;
TF_CHECK_OK(GetNodeAttr(node->def(), "send_device_incarnation",
reinterpret_cast<int64*>(&send_device_incarnation)));
return Rendezvous::CreateKey(send_device, send_device_incarnation,
recv_device, tensor_name, FrameAndIter(0, 0));
}
void Benchmark::RunWithArgs(
const std::vector<std::pair<const Node*, Tensor>>& inputs,
const std::vector<const Node*>& outputs, int iters) {
if (device_) {
// Gets inputs' and outputs' rendezvous keys.
std::vector<std::pair<string, Tensor>> in;
for (const auto& p : inputs) {
in.push_back({GetRendezvousKey(p.first), p.second});
}
std::vector<string> out;
for (const auto& n : outputs) {
out.push_back(GetRendezvousKey(n));
}
Tensor unused; // In benchmark, we don't care the return value.
bool is_dead;
// Warm up
Executor::Args args;
args.rendezvous = rendez_;
args.runner = [this](std::function<void()> closure) {
pool_->Schedule(closure);
};
for (int i = 0; i < 3; ++i) {
for (const auto& p : in) {
rendez_->Send(p.first, Rendezvous::Args(), p.second, false);
}
TF_CHECK_OK(exec_->Run(args));
for (const string& key : out) {
rendez_->Recv(key, Rendezvous::Args(), &unused, &is_dead);
}
}
TF_CHECK_OK(device_->Sync());
testing::StartTiming();
while (iters-- > 0) {
for (const auto& p : in) {
rendez_->Send(p.first, Rendezvous::Args(), p.second, false);
}
TF_CHECK_OK(exec_->Run(args));
for (const string& key : out) {
rendez_->Recv(key, Rendezvous::Args(), &unused, &is_dead);
}
}
TF_CHECK_OK(device_->Sync());
testing::StopTiming();
}
}
} // end namespace test
} // end namespace tensorflow
|
