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
|
/* 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/c/eager/runtime.h"
#include <memory>
#include <vector>
#include "tensorflow/cc/client/client_session.h"
#include "tensorflow/cc/framework/ops.h"
#include "tensorflow/cc/framework/scope.h"
#include "tensorflow/cc/ops/standard_ops.h"
#include "tensorflow/core/common_runtime/device_factory.h"
#include "tensorflow/core/platform/test.h"
#include "tensorflow/core/platform/test_benchmark.h"
namespace tensorflow {
namespace {
Device* CPUDevice() {
return DeviceFactory::NewDevice("CPU", {}, "/job:a/replica:0/task:0");
}
TEST(AttrTypeMap, Lookup) {
const AttrTypeMap* m = nullptr;
Status s = AttrTypeMapForOp("ThisOpCannotPossiblyExist", &m);
EXPECT_FALSE(s.ok());
s = AttrTypeMapForOp("MatMul", &m);
ASSERT_TRUE(s.ok()) << s;
TF_AttrType t;
unsigned char is_list = 1;
s = AttrTypeByName(m, "ThisAttribyteCannotPossiblyExist", &t, &is_list);
EXPECT_FALSE(s.ok());
EXPECT_NE(is_list, 0);
s = AttrTypeByName(m, "transpose_a", &t, &is_list);
ASSERT_TRUE(s.ok()) << s;
EXPECT_EQ(TF_ATTR_BOOL, t);
EXPECT_EQ(is_list, 0);
s = AttrTypeMapForOp("Squeeze", &m);
ASSERT_TRUE(s.ok()) << s;
s = AttrTypeByName(m, "squeeze_dims", &t, &is_list);
ASSERT_TRUE(s.ok()) << s;
EXPECT_EQ(TF_ATTR_INT, t);
EXPECT_NE(is_list, 0);
}
TEST(KernelAndDevice, Run) {
Tensor t(Input({{1.0f, 2.0f}, {3.0f, 4.0f}}).tensor());
std::vector<Tensor> inputs;
inputs.push_back(t);
inputs.push_back(t);
NodeDef ndef(AttrBuilder("MatMul")
.Set("T", DT_FLOAT)
.Set("transpose_a", false)
.Set("transpose_b", false)
.NumInputs(inputs.size())
.BuildNodeDef());
std::unique_ptr<Device> device(CPUDevice());
KernelAndDevice kernel;
Status s = KernelAndDevice::InitOp(device.get(), ndef, &kernel);
ASSERT_TRUE(s.ok()) << s;
std::vector<Tensor> outputs;
s = kernel.Run(&inputs, &outputs);
ASSERT_TRUE(s.ok()) << s;
ASSERT_EQ(1, outputs.size());
const Tensor& out = outputs[0];
EXPECT_EQ(7, out.matrix<float>()(0, 0));
EXPECT_EQ(10, out.matrix<float>()(0, 1));
EXPECT_EQ(15, out.matrix<float>()(1, 0));
EXPECT_EQ(22, out.matrix<float>()(1, 1));
}
// TODO(apassos) uncomment after rewriting to use the right benchmark API
// void BM_CreateGraph(benchmark::State& state) {
// for (auto _ : state) {
// Scope root = Scope::NewRootScope();
// auto C = ops::Const(root, {{1.0, 2.0}, {3.0, 4.0}});
// auto M = ops::MatMul(root, C, C);
// TF_CHECK_OK(root.status());
// }
// }
// BENCHMARK(BM_CreateGraph);
// void BM_RunGraph(benchmark::State& state) {
// Scope root = Scope::NewRootScope();
// auto C = ops::Const(root, {{1.0, 2.0}, {3.0, 4.0}});
// auto M = ops::MatMul(root, C, C);
// SessionOptions opts;
// opts.config.set_inter_op_parallelism_threads(1);
// opts.config.set_intra_op_parallelism_threads(1);
// ClientSession sess(root, opts);
// std::vector<Tensor> outputs;
// for (auto _ : state) {
// outputs.clear();
// TF_CHECK_OK(sess.Run({M}, &outputs));
// }
// }
// BENCHMARK(BM_RunGraph);
// void BM_CreateAndDestroySession(benchmark::State& state) {
// Scope root = Scope::NewRootScope();
// auto C = ops::Const(root, {{1.0, 2.0}, {3.0, 4.0}});
// auto M = ops::MatMul(root, C, C);
// for (auto _ : state) {
// ClientSession sess(root);
// }
// }
// BENCHMARK(BM_CreateAndDestroySession);
// void BM_KernelAndDeviceInit(benchmark::State& state) {
// NodeDef ndef(AttrBuilder("MatMul")
// .Set("T", DT_FLOAT)
// .Set("transpose_a", false)
// .Set("transpose_b", false)
// .NumInputs(2)
// .BuildNodeDef());
// std::unique_ptr<Device> device(CPUDevice());
// KernelAndDevice k;
// for (auto _ : state) {
// TF_CHECK_OK(KernelAndDevice::InitOp(device.get(), ndef, &k));
// }
// }
// BENCHMARK(BM_KernelAndDeviceInit);
// void BM_KernelAndDeviceRun(benchmark::State& state) {
// Tensor t(Input({{1.0f, 2.0f}, {3.0f, 4.0f}}).tensor());
// std::vector<Tensor> inputs;
// inputs.push_back(t);
// inputs.push_back(t);
// std::vector<Tensor> outputs;
// NodeDef ndef(AttrBuilder("MatMul")
// .Set("T", DT_FLOAT)
// .Set("transpose_a", false)
// .Set("transpose_b", false)
// .NumInputs(inputs.size())
// .BuildNodeDef());
// std::unique_ptr<Device> device(CPUDevice());
// KernelAndDevice kernel;
// TF_CHECK_OK(KernelAndDevice::InitOp(device.get(), ndef, &kernel));
// for (auto _ : state) {
// TF_CHECK_OK(kernel.Run(&inputs, &outputs));
// }
// }
// BENCHMARK(BM_KernelAndDeviceRun);
} // namespace
} // namespace tensorflow
|
