1 /* Copyright 2016 The TensorFlow Authors. All Rights Reserved.
2
3 Licensed under the Apache License, Version 2.0 (the "License");
4 you may not use this file except in compliance with the License.
5 You may obtain a copy of the License at
6
7 http://www.apache.org/licenses/LICENSE-2.0
8
9 Unless required by applicable law or agreed to in writing, software
10 distributed under the License is distributed on an "AS IS" BASIS,
11 WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
12 See the License for the specific language governing permissions and
13 limitations under the License.
14 ==============================================================================*/
15
16 #include "tensorflow/core/platform/profile_utils/cpu_utils.h"
17
18 #include <fstream>
19 #include <limits>
20 #include <mutex>
21
22 #if defined(_WIN32)
23 #include <windows.h>
24 #endif
25
26 #if defined(__APPLE__)
27 #include <sys/sysctl.h>
28 #endif
29
30 #include "absl/base/call_once.h"
31 #include "tensorflow/core/platform/logging.h"
32 #include "tensorflow/core/platform/profile_utils/android_armv7a_cpu_utils_helper.h"
33
34 namespace tensorflow {
35 namespace profile_utils {
36
37 /* static */ constexpr int64_t CpuUtils::INVALID_FREQUENCY;
38
39 static ICpuUtilsHelper* cpu_utils_helper_instance_ = nullptr;
40
41 #if (defined(__powerpc__) || \
42 defined(__ppc__) && (__BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__)) || \
43 (defined(__s390x__))
GetCycleCounterFrequency()44 /* static */ uint64 CpuUtils::GetCycleCounterFrequency() {
45 static const uint64 cpu_frequency = GetCycleCounterFrequencyImpl();
46 return cpu_frequency;
47 }
48 #else
GetCycleCounterFrequency()49 /* static */ int64_t CpuUtils::GetCycleCounterFrequency() {
50 static const int64_t cpu_frequency = GetCycleCounterFrequencyImpl();
51 return cpu_frequency;
52 }
53 #endif
54
GetMicroSecPerClock()55 /* static */ double CpuUtils::GetMicroSecPerClock() {
56 static const double micro_sec_per_clock =
57 (1000.0 * 1000.0) / static_cast<double>(GetCycleCounterFrequency());
58 return micro_sec_per_clock;
59 }
60
ResetClockCycle()61 /* static */ void CpuUtils::ResetClockCycle() {
62 GetCpuUtilsHelperSingletonInstance().ResetClockCycle();
63 }
64
EnableClockCycleProfiling()65 /* static */ void CpuUtils::EnableClockCycleProfiling() {
66 GetCpuUtilsHelperSingletonInstance().EnableClockCycleProfiling();
67 }
68
DisableClockCycleProfiling()69 /* static */ void CpuUtils::DisableClockCycleProfiling() {
70 GetCpuUtilsHelperSingletonInstance().DisableClockCycleProfiling();
71 }
72
ConvertClockCycleToTime(const int64_t clock_cycle)73 /* static */ std::chrono::duration<double> CpuUtils::ConvertClockCycleToTime(
74 const int64_t clock_cycle) {
75 return std::chrono::duration<double>(static_cast<double>(clock_cycle) /
76 GetCycleCounterFrequency());
77 }
78
GetCycleCounterFrequencyImpl()79 /* static */ int64_t CpuUtils::GetCycleCounterFrequencyImpl() {
80 // TODO(satok): do not switch by macro here
81 #if defined(__ANDROID__)
82 return GetCpuUtilsHelperSingletonInstance().CalculateCpuFrequency();
83 #elif defined(__linux__)
84 // Read the contents of /proc/cpuinfo.
85 std::ifstream cpuinfo("/proc/cpuinfo");
86 if (!cpuinfo) {
87 LOG(WARNING) << "Failed to open /proc/cpuinfo";
88 return INVALID_FREQUENCY;
89 }
90 string line;
91 while (std::getline(cpuinfo, line)) {
92 double cpu_freq = 0.0;
93 int retval = 0;
94 double freq_factor = 2.0;
95 #if (defined(__powerpc__) || \
96 defined(__ppc__) && (__BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__))
97 retval = sscanf(line.c_str(), "clock : %lfMHz", &cpu_freq);
98 freq_factor = 1.0;
99 #elif defined(__s390x__)
100 retval = sscanf(line.c_str(), "bogomips per cpu: %lf", &cpu_freq);
101 #elif defined(__aarch64__)
102 retval = sscanf(line.c_str(), "BogoMIPS : %lf", &cpu_freq);
103 #else
104 retval = sscanf(line.c_str(), "bogomips : %lf", &cpu_freq);
105 #endif
106 if (retval > 0) {
107 const double freq_ghz = cpu_freq / 1000.0 / freq_factor;
108 if (retval != 1 || freq_ghz < 0.01) {
109 LOG(WARNING) << "Failed to get CPU frequency: " << freq_ghz << " GHz";
110 return INVALID_FREQUENCY;
111 }
112 const int64_t freq_n =
113 static_cast<int64_t>(freq_ghz * 1000.0 * 1000.0 * 1000.0);
114 VLOG(1) << "CPU Frequency: " << freq_n << " Hz";
115 return freq_n;
116 }
117 }
118 LOG(WARNING)
119 << "Failed to find bogomips or clock in /proc/cpuinfo; cannot determine "
120 "CPU frequency";
121 return INVALID_FREQUENCY;
122 #elif defined(__APPLE__)
123 int64 freq_hz = 0;
124 size_t freq_size = sizeof(freq_hz);
125 int retval =
126 sysctlbyname("hw.cpufrequency_max", &freq_hz, &freq_size, NULL, 0);
127 if (retval != 0 || freq_hz < 1e6) {
128 LOG(WARNING) << "Failed to get CPU frequency: " << freq_hz << " Hz";
129 return INVALID_FREQUENCY;
130 }
131 return freq_hz;
132 #elif defined(_WIN32)
133 LARGE_INTEGER freq;
134 QueryPerformanceFrequency(&freq);
135 return freq.QuadPart;
136 #else
137 // TODO(satok): Support other OS if needed
138 // Return INVALID_FREQUENCY on unsupported OS
139 return INVALID_FREQUENCY;
140 #endif
141 }
142
GetCpuUtilsHelperSingletonInstance()143 /* static */ ICpuUtilsHelper& CpuUtils::GetCpuUtilsHelperSingletonInstance() {
144 static absl::once_flag flag;
145 absl::call_once(flag, []() {
146 if (cpu_utils_helper_instance_ != nullptr) {
147 LOG(FATAL) << "cpu_utils_helper_instance_ is already instantiated.";
148 }
149 #if defined(__ANDROID__) && (__ANDROID_API__ >= 21) && \
150 (defined(__ARM_ARCH_7A__) || defined(__aarch64__))
151 cpu_utils_helper_instance_ = new AndroidArmV7ACpuUtilsHelper();
152 #else
153 cpu_utils_helper_instance_ = new DefaultCpuUtilsHelper();
154 #endif
155 });
156 return *cpu_utils_helper_instance_;
157 }
158
159 } // namespace profile_utils
160 } // namespace tensorflow
161