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
203
204
205
206
207
208
209
|
// Copyright 2016 Ismael Jimenez Martinez. All rights reserved.
// Copyright 2017 Roman Lebedev. 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 "statistics.h"
#include <algorithm>
#include <cmath>
#include <numeric>
#include <string>
#include <vector>
#include "benchmark/benchmark.h"
#include "check.h"
namespace benchmark {
auto StatisticsSum = [](const std::vector<double>& v) {
return std::accumulate(v.begin(), v.end(), 0.0);
};
double StatisticsMean(const std::vector<double>& v) {
if (v.empty()) return 0.0;
return StatisticsSum(v) * (1.0 / v.size());
}
double StatisticsMedian(const std::vector<double>& v) {
if (v.size() < 3) return StatisticsMean(v);
std::vector<double> copy(v);
auto center = copy.begin() + v.size() / 2;
std::nth_element(copy.begin(), center, copy.end());
// Did we have an odd number of samples? If yes, then center is the median.
// If not, then we are looking for the average between center and the value
// before. Instead of resorting, we just look for the max value before it,
// which is not necessarily the element immediately preceding `center` Since
// `copy` is only partially sorted by `nth_element`.
if (v.size() % 2 == 1) return *center;
auto center2 = std::max_element(copy.begin(), center);
return (*center + *center2) / 2.0;
}
// Return the sum of the squares of this sample set
auto SumSquares = [](const std::vector<double>& v) {
return std::inner_product(v.begin(), v.end(), v.begin(), 0.0);
};
auto Sqr = [](const double dat) { return dat * dat; };
auto Sqrt = [](const double dat) {
// Avoid NaN due to imprecision in the calculations
if (dat < 0.0) return 0.0;
return std::sqrt(dat);
};
double StatisticsStdDev(const std::vector<double>& v) {
const auto mean = StatisticsMean(v);
if (v.empty()) return mean;
// Sample standard deviation is undefined for n = 1
if (v.size() == 1) return 0.0;
const double avg_squares = SumSquares(v) * (1.0 / v.size());
return Sqrt(v.size() / (v.size() - 1.0) * (avg_squares - Sqr(mean)));
}
double StatisticsCV(const std::vector<double>& v) {
if (v.size() < 2) return 0.0;
const auto stddev = StatisticsStdDev(v);
const auto mean = StatisticsMean(v);
return stddev / mean;
}
std::vector<BenchmarkReporter::Run> ComputeStats(
const std::vector<BenchmarkReporter::Run>& reports) {
typedef BenchmarkReporter::Run Run;
std::vector<Run> results;
auto error_count = std::count_if(reports.begin(), reports.end(),
[](Run const& run) { return run.skipped; });
if (reports.size() - error_count < 2) {
// We don't report aggregated data if there was a single run.
return results;
}
// Accumulators.
std::vector<double> real_accumulated_time_stat;
std::vector<double> cpu_accumulated_time_stat;
real_accumulated_time_stat.reserve(reports.size());
cpu_accumulated_time_stat.reserve(reports.size());
// All repetitions should be run with the same number of iterations so we
// can take this information from the first benchmark.
const IterationCount run_iterations = reports.front().iterations;
// create stats for user counters
struct CounterStat {
Counter c;
std::vector<double> s;
};
std::map<std::string, CounterStat> counter_stats;
for (Run const& r : reports) {
for (auto const& cnt : r.counters) {
auto it = counter_stats.find(cnt.first);
if (it == counter_stats.end()) {
it = counter_stats
.emplace(cnt.first,
CounterStat{cnt.second, std::vector<double>{}})
.first;
it->second.s.reserve(reports.size());
} else {
BM_CHECK_EQ(it->second.c.flags, cnt.second.flags);
}
}
}
// Populate the accumulators.
for (Run const& run : reports) {
BM_CHECK_EQ(reports[0].benchmark_name(), run.benchmark_name());
BM_CHECK_EQ(run_iterations, run.iterations);
if (run.skipped) continue;
real_accumulated_time_stat.emplace_back(run.real_accumulated_time);
cpu_accumulated_time_stat.emplace_back(run.cpu_accumulated_time);
// user counters
for (auto const& cnt : run.counters) {
auto it = counter_stats.find(cnt.first);
BM_CHECK_NE(it, counter_stats.end());
it->second.s.emplace_back(cnt.second);
}
}
// Only add label if it is same for all runs
std::string report_label = reports[0].report_label;
for (std::size_t i = 1; i < reports.size(); i++) {
if (reports[i].report_label != report_label) {
report_label = "";
break;
}
}
const double iteration_rescale_factor =
double(reports.size()) / double(run_iterations);
for (const auto& Stat : *reports[0].statistics) {
// Get the data from the accumulator to BenchmarkReporter::Run's.
Run data;
data.run_name = reports[0].run_name;
data.family_index = reports[0].family_index;
data.per_family_instance_index = reports[0].per_family_instance_index;
data.run_type = BenchmarkReporter::Run::RT_Aggregate;
data.threads = reports[0].threads;
data.repetitions = reports[0].repetitions;
data.repetition_index = Run::no_repetition_index;
data.aggregate_name = Stat.name_;
data.aggregate_unit = Stat.unit_;
data.report_label = report_label;
// It is incorrect to say that an aggregate is computed over
// run's iterations, because those iterations already got averaged.
// Similarly, if there are N repetitions with 1 iterations each,
// an aggregate will be computed over N measurements, not 1.
// Thus it is best to simply use the count of separate reports.
data.iterations = reports.size();
data.real_accumulated_time = Stat.compute_(real_accumulated_time_stat);
data.cpu_accumulated_time = Stat.compute_(cpu_accumulated_time_stat);
if (data.aggregate_unit == StatisticUnit::kTime) {
// We will divide these times by data.iterations when reporting, but the
// data.iterations is not necessarily the scale of these measurements,
// because in each repetition, these timers are sum over all the iters.
// And if we want to say that the stats are over N repetitions and not
// M iterations, we need to multiply these by (N/M).
data.real_accumulated_time *= iteration_rescale_factor;
data.cpu_accumulated_time *= iteration_rescale_factor;
}
data.time_unit = reports[0].time_unit;
// user counters
for (auto const& kv : counter_stats) {
// Do *NOT* rescale the custom counters. They are already properly scaled.
const auto uc_stat = Stat.compute_(kv.second.s);
auto c = Counter(uc_stat, counter_stats[kv.first].c.flags,
counter_stats[kv.first].c.oneK);
data.counters[kv.first] = c;
}
results.push_back(data);
}
return results;
}
} // end namespace benchmark
|
