/
util.cc
185 lines (164 loc) · 5.44 KB
/
util.cc
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
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
/* Copyright 2015 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/util/util.h"
#include <string>
#include <vector>
#include "absl/base/call_once.h"
#include "tensorflow/core/framework/device_factory.h"
#include "tensorflow/core/lib/gtl/inlined_vector.h"
#include "tensorflow/core/lib/strings/strcat.h"
#include "tensorflow/core/platform/cpu_info.h"
#include "tensorflow/core/platform/logging.h"
#include "tensorflow/core/util/env_var.h"
namespace tensorflow {
StringPiece NodeNamePrefix(const StringPiece& op_name) {
StringPiece sp(op_name);
auto p = sp.find('/');
if (p == StringPiece::npos || p == 0) {
return "";
} else {
return StringPiece(sp.data(), p);
}
}
StringPiece NodeNameFullPrefix(const StringPiece& op_name) {
StringPiece sp(op_name);
auto p = sp.rfind('/');
if (p == StringPiece::npos || p == 0) {
return "";
} else {
return StringPiece(sp.data(), p);
}
}
MovingAverage::MovingAverage(int window)
: window_(window),
sum_(0.0),
data_(new double[window_]),
head_(0),
count_(0) {
CHECK_GE(window, 1);
}
MovingAverage::~MovingAverage() { delete[] data_; }
void MovingAverage::Clear() {
count_ = 0;
head_ = 0;
sum_ = 0;
}
double MovingAverage::GetAverage() const {
if (count_ == 0) {
return 0;
} else {
return static_cast<double>(sum_) / count_;
}
}
void MovingAverage::AddValue(double v) {
if (count_ < window_) {
// This is the warmup phase. We don't have a full window's worth of data.
head_ = count_;
data_[count_++] = v;
} else {
if (window_ == ++head_) {
head_ = 0;
}
// Toss the oldest element
sum_ -= data_[head_];
// Add the newest element
data_[head_] = v;
}
sum_ += v;
}
static char hex_char[] = "0123456789abcdef";
string PrintMemory(const char* ptr, size_t n) {
string ret;
ret.resize(n * 3);
for (int i = 0; i < n; ++i) {
ret[i * 3] = ' ';
ret[i * 3 + 1] = hex_char[ptr[i] >> 4];
ret[i * 3 + 2] = hex_char[ptr[i] & 0xf];
}
return ret;
}
string SliceDebugString(const TensorShape& shape, const int64_t flat) {
// Special case rank 0 and 1
const int dims = shape.dims();
if (dims == 0) return "";
if (dims == 1) return strings::StrCat("[", flat, "]");
// Compute strides
gtl::InlinedVector<int64_t, 32> strides(dims);
strides.back() = 1;
for (int i = dims - 2; i >= 0; i--) {
strides[i] = strides[i + 1] * shape.dim_size(i + 1);
}
// Unflatten index
int64_t left = flat;
string result;
for (int i = 0; i < dims; i++) {
strings::StrAppend(&result, i ? "," : "[", left / strides[i]);
left %= strides[i];
}
strings::StrAppend(&result, "]");
return result;
}
bool IsMKLEnabled() {
#ifndef INTEL_MKL
return false;
#endif // !INTEL_MKL
static absl::once_flag once;
#ifdef ENABLE_MKL
// Keeping TF_DISABLE_MKL env variable for legacy reasons.
static bool oneDNN_disabled = false;
absl::call_once(once, [&] {
TF_CHECK_OK(ReadBoolFromEnvVar("TF_DISABLE_MKL", false, &oneDNN_disabled));
if (oneDNN_disabled) VLOG(2) << "TF-MKL: Disabling oneDNN";
});
return (!oneDNN_disabled);
#else
// Linux: Turn oneDNN on by default for CPUs with neural network features.
// Windows: oneDNN is off by default.
// No need to guard for other platforms here because INTEL_MKL is only defined
// for non-mobile Linux or Windows.
static bool oneDNN_enabled =
#ifdef __linux__
port::TestCPUFeature(port::CPUFeature::AVX512_VNNI) ||
port::TestCPUFeature(port::CPUFeature::AVX512_BF16) ||
port::TestCPUFeature(port::CPUFeature::AVX_VNNI) ||
port::TestCPUFeature(port::CPUFeature::AMX_TILE) ||
port::TestCPUFeature(port::CPUFeature::AMX_INT8) ||
port::TestCPUFeature(port::CPUFeature::AMX_BF16);
#else
false;
#endif // __linux__
absl::call_once(once, [&] {
auto status = ReadBoolFromEnvVar("TF_ENABLE_ONEDNN_OPTS", oneDNN_enabled,
&oneDNN_enabled);
if (!status.ok()) {
LOG(WARNING) << "TF_ENABLE_ONEDNN_OPTS is not set to either '0', 'false',"
<< " '1', or 'true'. Using the default setting: "
<< oneDNN_enabled;
}
if (oneDNN_enabled) {
#ifndef DNNL_AARCH64_USE_ACL
LOG(INFO) << "oneDNN custom operations are on. "
<< "You may see slightly different numerical results due to "
<< "floating-point round-off errors from different computation "
<< "orders. To turn them off, set the environment variable "
<< "`TF_ENABLE_ONEDNN_OPTS=0`.";
#else
LOG(INFO) << "Experimental oneDNN custom operations are on. "
<< "If you experience issues, please turn them off by setting "
<< "the environment variable `TF_ENABLE_ONEDNN_OPTS=0`.";
#endif // !DNNL_AARCH64_USE_ACL
}
});
return oneDNN_enabled;
#endif // ENABLE_MKL
}
} // namespace tensorflow