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/cpu_feature_guard.h"
17
18 #include <mutex>
19 #include <string>
20
21 #include "absl/base/call_once.h"
22 #include "tensorflow/core/platform/byte_order.h"
23 #include "tensorflow/core/platform/cpu_info.h"
24 #include "tensorflow/core/platform/logging.h"
25
26 namespace tensorflow {
27 namespace port {
28 namespace {
29
30 // If the CPU feature isn't present, log a fatal error.
CheckFeatureOrDie(CPUFeature feature,const string & feature_name)31 void CheckFeatureOrDie(CPUFeature feature, const string& feature_name) {
32 if (!TestCPUFeature(feature)) {
33 #ifdef __ANDROID__
34 // Some Android emulators seem to indicate they don't support SSE, so to
35 // avoid crashes when testing, switch this to a warning.
36 LOG(WARNING)
37 #else
38 LOG(FATAL)
39 #endif
40 << "The TensorFlow library was compiled to use " << feature_name
41 << " instructions, but these aren't available on your machine.";
42 }
43 }
44
45 // Check if CPU feature is included in the TensorFlow binary.
CheckIfFeatureUnused(CPUFeature feature,const string & feature_name,string & missing_instructions)46 void CheckIfFeatureUnused(CPUFeature feature, const string& feature_name,
47 string& missing_instructions) {
48 if (TestCPUFeature(feature)) {
49 missing_instructions.append(" ");
50 missing_instructions.append(feature_name);
51 }
52 }
53
54 // Raises an error if the binary has been compiled for a CPU feature (like AVX)
55 // that isn't available on the current machine. It also warns of performance
56 // loss if there's a feature available that's not being used.
57 // Depending on the compiler and initialization order, a SIGILL exception may
58 // occur before this code is reached, but this at least offers a chance to give
59 // a more meaningful error message.
60 class CPUFeatureGuard {
61 public:
CPUFeatureGuard()62 CPUFeatureGuard() {
63 #ifdef __SSE__
64 CheckFeatureOrDie(CPUFeature::SSE, "SSE");
65 #endif // __SSE__
66 #ifdef __SSE2__
67 CheckFeatureOrDie(CPUFeature::SSE2, "SSE2");
68 #endif // __SSE2__
69 #ifdef __SSE3__
70 CheckFeatureOrDie(CPUFeature::SSE3, "SSE3");
71 #endif // __SSE3__
72 #ifdef __SSE4_1__
73 CheckFeatureOrDie(CPUFeature::SSE4_1, "SSE4.1");
74 #endif // __SSE4_1__
75 #ifdef __SSE4_2__
76 CheckFeatureOrDie(CPUFeature::SSE4_2, "SSE4.2");
77 #endif // __SSE4_2__
78 #ifdef __AVX__
79 CheckFeatureOrDie(CPUFeature::AVX, "AVX");
80 #endif // __AVX__
81 #ifdef __AVX2__
82 CheckFeatureOrDie(CPUFeature::AVX2, "AVX2");
83 #endif // __AVX2__
84 #ifdef __AVX512F__
85 CheckFeatureOrDie(CPUFeature::AVX512F, "AVX512F");
86 #endif // __AVX512F__
87 #ifdef __FMA__
88 CheckFeatureOrDie(CPUFeature::FMA, "FMA");
89 #endif // __FMA__
90 }
91 };
92
93 CPUFeatureGuard g_cpu_feature_guard_singleton;
94
95 absl::once_flag g_cpu_feature_guard_warn_once_flag;
96
97 } // namespace
98
InfoAboutUnusedCPUFeatures()99 void InfoAboutUnusedCPUFeatures() {
100 absl::call_once(g_cpu_feature_guard_warn_once_flag, [] {
101 string missing_instructions;
102 #if defined(_MSC_VER) && !defined(__clang__)
103
104 #ifndef __AVX__
105 CheckIfFeatureUnused(CPUFeature::AVX, "AVX", missing_instructions);
106 #endif // __AVX__
107 #ifndef __AVX2__
108 CheckIfFeatureUnused(CPUFeature::AVX2, "AVX2", missing_instructions);
109 #endif // __AVX2__
110
111 #else // if defined(_MSC_VER) && !defined(__clang__)
112
113 #ifndef __SSE__
114 CheckIfFeatureUnused(CPUFeature::SSE, "SSE", missing_instructions);
115 #endif // __SSE__
116 #ifndef __SSE2__
117 CheckIfFeatureUnused(CPUFeature::SSE2, "SSE2", missing_instructions);
118 #endif // __SSE2__
119 #ifndef __SSE3__
120 CheckIfFeatureUnused(CPUFeature::SSE3, "SSE3", missing_instructions);
121 #endif // __SSE3__
122 #ifndef __SSE4_1__
123 CheckIfFeatureUnused(CPUFeature::SSE4_1, "SSE4.1", missing_instructions);
124 #endif // __SSE4_1__
125 #ifndef __SSE4_2__
126 CheckIfFeatureUnused(CPUFeature::SSE4_2, "SSE4.2", missing_instructions);
127 #endif // __SSE4_2__
128 #ifndef __AVX__
129 CheckIfFeatureUnused(CPUFeature::AVX, "AVX", missing_instructions);
130 #endif // __AVX__
131 #ifndef __AVX2__
132 CheckIfFeatureUnused(CPUFeature::AVX2, "AVX2", missing_instructions);
133 #endif // __AVX2__
134 #ifndef __AVX512F__
135 CheckIfFeatureUnused(CPUFeature::AVX512F, "AVX512F", missing_instructions);
136 #endif // __AVX512F__
137 #ifndef __FMA__
138 CheckIfFeatureUnused(CPUFeature::FMA, "FMA", missing_instructions);
139 #endif // __FMA__
140 #endif // else of if defined(_MSC_VER) && !defined(__clang__)
141 if (!missing_instructions.empty()) {
142 LOG(INFO) << "This TensorFlow binary is optimized with "
143 << "oneAPI Deep Neural Network Library (oneDNN) "
144 << "to use the following CPU instructions in performance-"
145 << "critical operations: " << missing_instructions << std::endl
146 << "To enable them in other operations, rebuild TensorFlow "
147 << "with the appropriate compiler flags.";
148 }
149 });
150 }
151
152 } // namespace port
153 } // namespace tensorflow
154