File size: 7,938 Bytes
8b7c501
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
// Copyright 2019 Google LLC
//
// This source code is licensed under the BSD-style license found in the
// LICENSE file in the root directory of this source tree.

#include <algorithm>
#include <cmath>
#include <functional>
#include <memory>
#include <random>
#include <vector>

#include <benchmark/benchmark.h>

#include "bench/utils.h"

#include <xnnpack.h>
#include <xnnpack/models.h>


static void End2EndBenchmark(
  benchmark::State& state,
  models::ExecutionPlanFactory model_factory)
{
  if (xnn_initialize(nullptr /* allocator */) != xnn_status_success) {
    state.SkipWithError("failed to initialize XNNPACK");
    return;
  }

  const size_t num_threads = state.range(0);
  std::unique_ptr<pthreadpool, decltype(&pthreadpool_destroy)> threadpool(
    pthreadpool_create(num_threads), pthreadpool_destroy);

  auto execution_plan = model_factory(threadpool.get());
  if (execution_plan.empty()) {
    state.SkipWithError("failed to create a model");
    return;
  }

  for (auto _ : state) {
    for (const std::unique_ptr<xnn_operator, decltype(&xnn_delete_operator)>& op : execution_plan) {
      xnn_status status = xnn_run_operator(op.get(), threadpool.get());
      if (status != xnn_status_success) {
        state.SkipWithError("failed to run a model");
        return;
      }
    }
  }

  const uint64_t cpu_frequency = benchmark::utils::GetCurrentCpuFrequency();
  if (cpu_frequency != 0) {
    state.counters["cpufreq"] = cpu_frequency;
  }
}

static void FP32MobileNetV1(benchmark::State& state) {
  End2EndBenchmark(state, models::FP32MobileNetV1);
}

static void FP32MobileNetV2(benchmark::State& state) {
  End2EndBenchmark(state, models::FP32MobileNetV2);
}

static void FP32MobileNetV3Large(benchmark::State& state) {
  End2EndBenchmark(state, models::FP32MobileNetV3Large);
}

static void FP32MobileNetV3Small(benchmark::State& state) {
  End2EndBenchmark(state, models::FP32MobileNetV3Small);
}

#if XNN_PLATFORM_JIT && XNN_ENABLE_JIT
static void FP32MobileNetV3SmallFused(benchmark::State& state) {
  End2EndBenchmark(state, models::FP32MobileNetV3SmallFused);
}
#endif  // XNN_PLATFORM_JIT && XNN_ENABLE_JIT

static void FP32Sparse80MobileNetV1(benchmark::State& state) {
  End2EndBenchmark(state, [](pthreadpool_t threadpool) {
    return models::FP32SparseMobileNetV1(0.8f, threadpool);
  });
}

static void FP32Sparse80MobileNetV2(benchmark::State& state) {
  End2EndBenchmark(state, [](pthreadpool_t threadpool) {
    return models::FP32SparseMobileNetV2(0.8f, threadpool);
  });
}

static void FP32Sparse80MobileNetV3Large(benchmark::State& state) {
  End2EndBenchmark(state, [](pthreadpool_t threadpool) {
    return models::FP32SparseMobileNetV3Large(0.8f, threadpool);
  });
}

static void FP32Sparse80MobileNetV3Small(benchmark::State& state) {
  End2EndBenchmark(state, [](pthreadpool_t threadpool) {
    return models::FP32SparseMobileNetV3Small(0.8f, threadpool);
  });
}

static void FP16MobileNetV1(benchmark::State& state) {
  End2EndBenchmark(state, models::FP16MobileNetV1);
}

static void FP16MobileNetV2(benchmark::State& state) {
  End2EndBenchmark(state, models::FP16MobileNetV2);
}

static void FP16MobileNetV3Large(benchmark::State& state) {
  End2EndBenchmark(state, models::FP16MobileNetV3Large);
}

static void FP16MobileNetV3Small(benchmark::State& state) {
  End2EndBenchmark(state, models::FP16MobileNetV3Small);
}

static void FP16Sparse80MobileNetV1(benchmark::State& state) {
  End2EndBenchmark(state, [](pthreadpool_t threadpool) {
    return models::FP16SparseMobileNetV1(0.8f, threadpool);
  });
}

static void FP16Sparse80MobileNetV2(benchmark::State& state) {
  End2EndBenchmark(state, [](pthreadpool_t threadpool) {
    return models::FP16SparseMobileNetV2(0.8f, threadpool);
  });
}

static void FP16Sparse80MobileNetV3Large(benchmark::State& state) {
  End2EndBenchmark(state, [](pthreadpool_t threadpool) {
    return models::FP16SparseMobileNetV3Large(0.8f, threadpool);
  });
}

static void FP16Sparse80MobileNetV3Small(benchmark::State& state) {
  End2EndBenchmark(state, [](pthreadpool_t threadpool) {
    return models::FP16SparseMobileNetV3Small(0.8f, threadpool);
  });
}

static void QC8MobileNetV1(benchmark::State& state) {
  End2EndBenchmark(state, models::QC8MobileNetV1);
}

static void QC8MobileNetV2(benchmark::State& state) {
  End2EndBenchmark(state, models::QC8MobileNetV2);
}

static void QS8MobileNetV1(benchmark::State& state) {
  End2EndBenchmark(state, models::QS8MobileNetV1);
}

static void QS8MobileNetV2(benchmark::State& state) {
  End2EndBenchmark(state, models::QS8MobileNetV2);
}

static void QU8MobileNetV1(benchmark::State& state) {
  End2EndBenchmark(state, models::QU8MobileNetV1);
}

static void QU8MobileNetV2(benchmark::State& state) {
  End2EndBenchmark(state, models::QU8MobileNetV2);
}

BENCHMARK(FP32MobileNetV1)->Apply(benchmark::utils::MultiThreadingParameters)->Unit(benchmark::kMicrosecond)->UseRealTime();
BENCHMARK(FP32MobileNetV2)->Apply(benchmark::utils::MultiThreadingParameters)->Unit(benchmark::kMicrosecond)->UseRealTime();
BENCHMARK(FP32MobileNetV3Large)->Apply(benchmark::utils::MultiThreadingParameters)->Unit(benchmark::kMicrosecond)->UseRealTime();
BENCHMARK(FP32MobileNetV3Small)->Apply(benchmark::utils::MultiThreadingParameters)->Unit(benchmark::kMicrosecond)->UseRealTime();

BENCHMARK(FP32Sparse80MobileNetV1)->Apply(benchmark::utils::MultiThreadingParameters)->Unit(benchmark::kMicrosecond)->UseRealTime();
BENCHMARK(FP32Sparse80MobileNetV2)->Apply(benchmark::utils::MultiThreadingParameters)->Unit(benchmark::kMicrosecond)->UseRealTime();
BENCHMARK(FP32Sparse80MobileNetV3Large)->Apply(benchmark::utils::MultiThreadingParameters)->Unit(benchmark::kMicrosecond)->UseRealTime();
BENCHMARK(FP32Sparse80MobileNetV3Small)->Apply(benchmark::utils::MultiThreadingParameters)->Unit(benchmark::kMicrosecond)->UseRealTime();

BENCHMARK(FP16MobileNetV1)->Apply(benchmark::utils::MultiThreadingParameters)->Unit(benchmark::kMicrosecond)->UseRealTime();
BENCHMARK(FP16MobileNetV2)->Apply(benchmark::utils::MultiThreadingParameters)->Unit(benchmark::kMicrosecond)->UseRealTime();
BENCHMARK(FP16MobileNetV3Large)->Apply(benchmark::utils::MultiThreadingParameters)->Unit(benchmark::kMicrosecond)->UseRealTime();
BENCHMARK(FP16MobileNetV3Small)->Apply(benchmark::utils::MultiThreadingParameters)->Unit(benchmark::kMicrosecond)->UseRealTime();

BENCHMARK(FP16Sparse80MobileNetV1)->Apply(benchmark::utils::MultiThreadingParameters)->Unit(benchmark::kMicrosecond)->UseRealTime();
BENCHMARK(FP16Sparse80MobileNetV2)->Apply(benchmark::utils::MultiThreadingParameters)->Unit(benchmark::kMicrosecond)->UseRealTime();
BENCHMARK(FP16Sparse80MobileNetV3Large)->Apply(benchmark::utils::MultiThreadingParameters)->Unit(benchmark::kMicrosecond)->UseRealTime();
BENCHMARK(FP16Sparse80MobileNetV3Small)->Apply(benchmark::utils::MultiThreadingParameters)->Unit(benchmark::kMicrosecond)->UseRealTime();

BENCHMARK(QC8MobileNetV1)->Apply(benchmark::utils::MultiThreadingParameters)->Unit(benchmark::kMicrosecond)->UseRealTime();
BENCHMARK(QC8MobileNetV2)->Apply(benchmark::utils::MultiThreadingParameters)->Unit(benchmark::kMicrosecond)->UseRealTime();

BENCHMARK(QS8MobileNetV1)->Apply(benchmark::utils::MultiThreadingParameters)->Unit(benchmark::kMicrosecond)->UseRealTime();
BENCHMARK(QS8MobileNetV2)->Apply(benchmark::utils::MultiThreadingParameters)->Unit(benchmark::kMicrosecond)->UseRealTime();

BENCHMARK(QU8MobileNetV1)->Apply(benchmark::utils::MultiThreadingParameters)->Unit(benchmark::kMicrosecond)->UseRealTime();
BENCHMARK(QU8MobileNetV2)->Apply(benchmark::utils::MultiThreadingParameters)->Unit(benchmark::kMicrosecond)->UseRealTime();

#if XNN_PLATFORM_JIT && XNN_ENABLE_JIT
BENCHMARK(FP32MobileNetV3SmallFused)->Apply(benchmark::utils::MultiThreadingParameters)->Unit(benchmark::kMicrosecond)->UseRealTime();
#endif  // XNN_PLATFORM_JIT && XNN_ENABLE_JIT

#ifndef XNNPACK_BENCHMARK_NO_MAIN
BENCHMARK_MAIN();
#endif