xref: /aosp_15_r20/external/google-benchmark/src/complexity.cc (revision dbb99499c3810fa1611fa2242a2fc446be01a57c)
1 // Copyright 2016 Ismael Jimenez Martinez. 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 // Source project : https://github.com/ismaelJimenez/cpp.leastsq
16 // Adapted to be used with google benchmark
17 
18 #include "complexity.h"
19 
20 #include <algorithm>
21 #include <cmath>
22 
23 #include "benchmark/benchmark.h"
24 #include "check.h"
25 
26 namespace benchmark {
27 
28 // Internal function to calculate the different scalability forms
FittingCurve(BigO complexity)29 BigOFunc* FittingCurve(BigO complexity) {
30   switch (complexity) {
31     case oN:
32       return [](IterationCount n) -> double { return static_cast<double>(n); };
33     case oNSquared:
34       return [](IterationCount n) -> double { return std::pow(n, 2); };
35     case oNCubed:
36       return [](IterationCount n) -> double { return std::pow(n, 3); };
37     case oLogN:
38       return [](IterationCount n) -> double {
39         return std::log2(static_cast<double>(n));
40       };
41     case oNLogN:
42       return [](IterationCount n) -> double {
43         return static_cast<double>(n) * std::log2(static_cast<double>(n));
44       };
45     case o1:
46     default:
47       return [](IterationCount) { return 1.0; };
48   }
49 }
50 
51 // Function to return an string for the calculated complexity
GetBigOString(BigO complexity)52 std::string GetBigOString(BigO complexity) {
53   switch (complexity) {
54     case oN:
55       return "N";
56     case oNSquared:
57       return "N^2";
58     case oNCubed:
59       return "N^3";
60     case oLogN:
61       return "lgN";
62     case oNLogN:
63       return "NlgN";
64     case o1:
65       return "(1)";
66     default:
67       return "f(N)";
68   }
69 }
70 
71 // Find the coefficient for the high-order term in the running time, by
72 // minimizing the sum of squares of relative error, for the fitting curve
73 // given by the lambda expression.
74 //   - n             : Vector containing the size of the benchmark tests.
75 //   - time          : Vector containing the times for the benchmark tests.
76 //   - fitting_curve : lambda expression (e.g. [](ComplexityN n) {return n; };).
77 
78 // For a deeper explanation on the algorithm logic, please refer to
79 // https://en.wikipedia.org/wiki/Least_squares#Least_squares,_regression_analysis_and_statistics
80 
MinimalLeastSq(const std::vector<ComplexityN> & n,const std::vector<double> & time,BigOFunc * fitting_curve)81 LeastSq MinimalLeastSq(const std::vector<ComplexityN>& n,
82                        const std::vector<double>& time,
83                        BigOFunc* fitting_curve) {
84   double sigma_gn_squared = 0.0;
85   double sigma_time = 0.0;
86   double sigma_time_gn = 0.0;
87 
88   // Calculate least square fitting parameter
89   for (size_t i = 0; i < n.size(); ++i) {
90     double gn_i = fitting_curve(n[i]);
91     sigma_gn_squared += gn_i * gn_i;
92     sigma_time += time[i];
93     sigma_time_gn += time[i] * gn_i;
94   }
95 
96   LeastSq result;
97   result.complexity = oLambda;
98 
99   // Calculate complexity.
100   result.coef = sigma_time_gn / sigma_gn_squared;
101 
102   // Calculate RMS
103   double rms = 0.0;
104   for (size_t i = 0; i < n.size(); ++i) {
105     double fit = result.coef * fitting_curve(n[i]);
106     rms += std::pow((time[i] - fit), 2);
107   }
108 
109   // Normalized RMS by the mean of the observed values
110   double mean = sigma_time / static_cast<double>(n.size());
111   result.rms = std::sqrt(rms / static_cast<double>(n.size())) / mean;
112 
113   return result;
114 }
115 
116 // Find the coefficient for the high-order term in the running time, by
117 // minimizing the sum of squares of relative error.
118 //   - n          : Vector containing the size of the benchmark tests.
119 //   - time       : Vector containing the times for the benchmark tests.
120 //   - complexity : If different than oAuto, the fitting curve will stick to
121 //                  this one. If it is oAuto, it will be calculated the best
122 //                  fitting curve.
MinimalLeastSq(const std::vector<ComplexityN> & n,const std::vector<double> & time,const BigO complexity)123 LeastSq MinimalLeastSq(const std::vector<ComplexityN>& n,
124                        const std::vector<double>& time, const BigO complexity) {
125   BM_CHECK_EQ(n.size(), time.size());
126   BM_CHECK_GE(n.size(), 2);  // Do not compute fitting curve is less than two
127                              // benchmark runs are given
128   BM_CHECK_NE(complexity, oNone);
129 
130   LeastSq best_fit;
131 
132   if (complexity == oAuto) {
133     std::vector<BigO> fit_curves = {oLogN, oN, oNLogN, oNSquared, oNCubed};
134 
135     // Take o1 as default best fitting curve
136     best_fit = MinimalLeastSq(n, time, FittingCurve(o1));
137     best_fit.complexity = o1;
138 
139     // Compute all possible fitting curves and stick to the best one
140     for (const auto& fit : fit_curves) {
141       LeastSq current_fit = MinimalLeastSq(n, time, FittingCurve(fit));
142       if (current_fit.rms < best_fit.rms) {
143         best_fit = current_fit;
144         best_fit.complexity = fit;
145       }
146     }
147   } else {
148     best_fit = MinimalLeastSq(n, time, FittingCurve(complexity));
149     best_fit.complexity = complexity;
150   }
151 
152   return best_fit;
153 }
154 
ComputeBigO(const std::vector<BenchmarkReporter::Run> & reports)155 std::vector<BenchmarkReporter::Run> ComputeBigO(
156     const std::vector<BenchmarkReporter::Run>& reports) {
157   typedef BenchmarkReporter::Run Run;
158   std::vector<Run> results;
159 
160   if (reports.size() < 2) return results;
161 
162   // Accumulators.
163   std::vector<ComplexityN> n;
164   std::vector<double> real_time;
165   std::vector<double> cpu_time;
166 
167   // Populate the accumulators.
168   for (const Run& run : reports) {
169     BM_CHECK_GT(run.complexity_n, 0)
170         << "Did you forget to call SetComplexityN?";
171     n.push_back(run.complexity_n);
172     real_time.push_back(run.real_accumulated_time /
173                         static_cast<double>(run.iterations));
174     cpu_time.push_back(run.cpu_accumulated_time /
175                        static_cast<double>(run.iterations));
176   }
177 
178   LeastSq result_cpu;
179   LeastSq result_real;
180 
181   if (reports[0].complexity == oLambda) {
182     result_cpu = MinimalLeastSq(n, cpu_time, reports[0].complexity_lambda);
183     result_real = MinimalLeastSq(n, real_time, reports[0].complexity_lambda);
184   } else {
185     const BigO* InitialBigO = &reports[0].complexity;
186     const bool use_real_time_for_initial_big_o =
187         reports[0].use_real_time_for_initial_big_o;
188     if (use_real_time_for_initial_big_o) {
189       result_real = MinimalLeastSq(n, real_time, *InitialBigO);
190       InitialBigO = &result_real.complexity;
191       // The Big-O complexity for CPU time must have the same Big-O function!
192     }
193     result_cpu = MinimalLeastSq(n, cpu_time, *InitialBigO);
194     InitialBigO = &result_cpu.complexity;
195     if (!use_real_time_for_initial_big_o) {
196       result_real = MinimalLeastSq(n, real_time, *InitialBigO);
197     }
198   }
199 
200   // Drop the 'args' when reporting complexity.
201   auto run_name = reports[0].run_name;
202   run_name.args.clear();
203 
204   // Get the data from the accumulator to BenchmarkReporter::Run's.
205   Run big_o;
206   big_o.run_name = run_name;
207   big_o.family_index = reports[0].family_index;
208   big_o.per_family_instance_index = reports[0].per_family_instance_index;
209   big_o.run_type = BenchmarkReporter::Run::RT_Aggregate;
210   big_o.repetitions = reports[0].repetitions;
211   big_o.repetition_index = Run::no_repetition_index;
212   big_o.threads = reports[0].threads;
213   big_o.aggregate_name = "BigO";
214   big_o.aggregate_unit = StatisticUnit::kTime;
215   big_o.report_label = reports[0].report_label;
216   big_o.iterations = 0;
217   big_o.real_accumulated_time = result_real.coef;
218   big_o.cpu_accumulated_time = result_cpu.coef;
219   big_o.report_big_o = true;
220   big_o.complexity = result_cpu.complexity;
221 
222   // All the time results are reported after being multiplied by the
223   // time unit multiplier. But since RMS is a relative quantity it
224   // should not be multiplied at all. So, here, we _divide_ it by the
225   // multiplier so that when it is multiplied later the result is the
226   // correct one.
227   double multiplier = GetTimeUnitMultiplier(reports[0].time_unit);
228 
229   // Only add label to mean/stddev if it is same for all runs
230   Run rms;
231   rms.run_name = run_name;
232   rms.family_index = reports[0].family_index;
233   rms.per_family_instance_index = reports[0].per_family_instance_index;
234   rms.run_type = BenchmarkReporter::Run::RT_Aggregate;
235   rms.aggregate_name = "RMS";
236   rms.aggregate_unit = StatisticUnit::kPercentage;
237   rms.report_label = big_o.report_label;
238   rms.iterations = 0;
239   rms.repetition_index = Run::no_repetition_index;
240   rms.repetitions = reports[0].repetitions;
241   rms.threads = reports[0].threads;
242   rms.real_accumulated_time = result_real.rms / multiplier;
243   rms.cpu_accumulated_time = result_cpu.rms / multiplier;
244   rms.report_rms = true;
245   rms.complexity = result_cpu.complexity;
246   // don't forget to keep the time unit, or we won't be able to
247   // recover the correct value.
248   rms.time_unit = reports[0].time_unit;
249 
250   results.push_back(big_o);
251   results.push_back(rms);
252   return results;
253 }
254 
255 }  // end namespace benchmark
256