[ThinLTO] Add code comment. NFC
[llvm-core.git] / utils / benchmark / src / benchmark_register.cc
blobdc6f93568539b7fc0a7fe02de38be13a103b9c9f
1 // Copyright 2015 Google Inc. 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.
15 #include "benchmark_register.h"
17 #ifndef BENCHMARK_OS_WINDOWS
18 #ifndef BENCHMARK_OS_FUCHSIA
19 #include <sys/resource.h>
20 #endif
21 #include <sys/time.h>
22 #include <unistd.h>
23 #endif
25 #include <algorithm>
26 #include <atomic>
27 #include <condition_variable>
28 #include <cstdio>
29 #include <cstdlib>
30 #include <cstring>
31 #include <fstream>
32 #include <iostream>
33 #include <memory>
34 #include <sstream>
35 #include <thread>
37 #include "benchmark/benchmark.h"
38 #include "benchmark_api_internal.h"
39 #include "check.h"
40 #include "commandlineflags.h"
41 #include "complexity.h"
42 #include "internal_macros.h"
43 #include "log.h"
44 #include "mutex.h"
45 #include "re.h"
46 #include "statistics.h"
47 #include "string_util.h"
48 #include "timers.h"
50 namespace benchmark {
52 namespace {
53 // For non-dense Range, intermediate values are powers of kRangeMultiplier.
54 static const int kRangeMultiplier = 8;
55 // The size of a benchmark family determines is the number of inputs to repeat
56 // the benchmark on. If this is "large" then warn the user during configuration.
57 static const size_t kMaxFamilySize = 100;
58 } // end namespace
60 namespace internal {
62 //=============================================================================//
63 // BenchmarkFamilies
64 //=============================================================================//
66 // Class for managing registered benchmarks. Note that each registered
67 // benchmark identifies a family of related benchmarks to run.
68 class BenchmarkFamilies {
69 public:
70 static BenchmarkFamilies* GetInstance();
72 // Registers a benchmark family and returns the index assigned to it.
73 size_t AddBenchmark(std::unique_ptr<Benchmark> family);
75 // Clear all registered benchmark families.
76 void ClearBenchmarks();
78 // Extract the list of benchmark instances that match the specified
79 // regular expression.
80 bool FindBenchmarks(std::string re,
81 std::vector<Benchmark::Instance>* benchmarks,
82 std::ostream* Err);
84 private:
85 BenchmarkFamilies() {}
87 std::vector<std::unique_ptr<Benchmark>> families_;
88 Mutex mutex_;
91 BenchmarkFamilies* BenchmarkFamilies::GetInstance() {
92 static BenchmarkFamilies instance;
93 return &instance;
96 size_t BenchmarkFamilies::AddBenchmark(std::unique_ptr<Benchmark> family) {
97 MutexLock l(mutex_);
98 size_t index = families_.size();
99 families_.push_back(std::move(family));
100 return index;
103 void BenchmarkFamilies::ClearBenchmarks() {
104 MutexLock l(mutex_);
105 families_.clear();
106 families_.shrink_to_fit();
109 bool BenchmarkFamilies::FindBenchmarks(
110 std::string spec, std::vector<Benchmark::Instance>* benchmarks,
111 std::ostream* ErrStream) {
112 CHECK(ErrStream);
113 auto& Err = *ErrStream;
114 // Make regular expression out of command-line flag
115 std::string error_msg;
116 Regex re;
117 bool isNegativeFilter = false;
118 if(spec[0] == '-') {
119 spec.replace(0, 1, "");
120 isNegativeFilter = true;
122 if (!re.Init(spec, &error_msg)) {
123 Err << "Could not compile benchmark re: " << error_msg << std::endl;
124 return false;
127 // Special list of thread counts to use when none are specified
128 const std::vector<int> one_thread = {1};
130 MutexLock l(mutex_);
131 for (std::unique_ptr<Benchmark>& family : families_) {
132 // Family was deleted or benchmark doesn't match
133 if (!family) continue;
135 if (family->ArgsCnt() == -1) {
136 family->Args({});
138 const std::vector<int>* thread_counts =
139 (family->thread_counts_.empty()
140 ? &one_thread
141 : &static_cast<const std::vector<int>&>(family->thread_counts_));
142 const size_t family_size = family->args_.size() * thread_counts->size();
143 // The benchmark will be run at least 'family_size' different inputs.
144 // If 'family_size' is very large warn the user.
145 if (family_size > kMaxFamilySize) {
146 Err << "The number of inputs is very large. " << family->name_
147 << " will be repeated at least " << family_size << " times.\n";
149 // reserve in the special case the regex ".", since we know the final
150 // family size.
151 if (spec == ".") benchmarks->reserve(family_size);
153 for (auto const& args : family->args_) {
154 for (int num_threads : *thread_counts) {
155 Benchmark::Instance instance;
156 instance.name = family->name_;
157 instance.benchmark = family.get();
158 instance.report_mode = family->report_mode_;
159 instance.arg = args;
160 instance.time_unit = family->time_unit_;
161 instance.range_multiplier = family->range_multiplier_;
162 instance.min_time = family->min_time_;
163 instance.iterations = family->iterations_;
164 instance.repetitions = family->repetitions_;
165 instance.use_real_time = family->use_real_time_;
166 instance.use_manual_time = family->use_manual_time_;
167 instance.complexity = family->complexity_;
168 instance.complexity_lambda = family->complexity_lambda_;
169 instance.statistics = &family->statistics_;
170 instance.threads = num_threads;
172 // Add arguments to instance name
173 size_t arg_i = 0;
174 for (auto const& arg : args) {
175 instance.name += "/";
177 if (arg_i < family->arg_names_.size()) {
178 const auto& arg_name = family->arg_names_[arg_i];
179 if (!arg_name.empty()) {
180 instance.name +=
181 StrFormat("%s:", family->arg_names_[arg_i].c_str());
185 instance.name += StrFormat("%d", arg);
186 ++arg_i;
189 if (!IsZero(family->min_time_))
190 instance.name += StrFormat("/min_time:%0.3f", family->min_time_);
191 if (family->iterations_ != 0)
192 instance.name += StrFormat("/iterations:%d", family->iterations_);
193 if (family->repetitions_ != 0)
194 instance.name += StrFormat("/repeats:%d", family->repetitions_);
196 if (family->use_manual_time_) {
197 instance.name += "/manual_time";
198 } else if (family->use_real_time_) {
199 instance.name += "/real_time";
202 // Add the number of threads used to the name
203 if (!family->thread_counts_.empty()) {
204 instance.name += StrFormat("/threads:%d", instance.threads);
207 if ((re.Match(instance.name) && !isNegativeFilter) ||
208 (!re.Match(instance.name) && isNegativeFilter)) {
209 instance.last_benchmark_instance = (&args == &family->args_.back());
210 benchmarks->push_back(std::move(instance));
215 return true;
218 Benchmark* RegisterBenchmarkInternal(Benchmark* bench) {
219 std::unique_ptr<Benchmark> bench_ptr(bench);
220 BenchmarkFamilies* families = BenchmarkFamilies::GetInstance();
221 families->AddBenchmark(std::move(bench_ptr));
222 return bench;
225 // FIXME: This function is a hack so that benchmark.cc can access
226 // `BenchmarkFamilies`
227 bool FindBenchmarksInternal(const std::string& re,
228 std::vector<Benchmark::Instance>* benchmarks,
229 std::ostream* Err) {
230 return BenchmarkFamilies::GetInstance()->FindBenchmarks(re, benchmarks, Err);
233 //=============================================================================//
234 // Benchmark
235 //=============================================================================//
237 Benchmark::Benchmark(const char* name)
238 : name_(name),
239 report_mode_(RM_Unspecified),
240 time_unit_(kNanosecond),
241 range_multiplier_(kRangeMultiplier),
242 min_time_(0),
243 iterations_(0),
244 repetitions_(0),
245 use_real_time_(false),
246 use_manual_time_(false),
247 complexity_(oNone),
248 complexity_lambda_(nullptr) {
249 ComputeStatistics("mean", StatisticsMean);
250 ComputeStatistics("median", StatisticsMedian);
251 ComputeStatistics("stddev", StatisticsStdDev);
254 Benchmark::~Benchmark() {}
256 Benchmark* Benchmark::Arg(int64_t x) {
257 CHECK(ArgsCnt() == -1 || ArgsCnt() == 1);
258 args_.push_back({x});
259 return this;
262 Benchmark* Benchmark::Unit(TimeUnit unit) {
263 time_unit_ = unit;
264 return this;
267 Benchmark* Benchmark::Range(int64_t start, int64_t limit) {
268 CHECK(ArgsCnt() == -1 || ArgsCnt() == 1);
269 std::vector<int64_t> arglist;
270 AddRange(&arglist, start, limit, range_multiplier_);
272 for (int64_t i : arglist) {
273 args_.push_back({i});
275 return this;
278 Benchmark* Benchmark::Ranges(
279 const std::vector<std::pair<int64_t, int64_t>>& ranges) {
280 CHECK(ArgsCnt() == -1 || ArgsCnt() == static_cast<int>(ranges.size()));
281 std::vector<std::vector<int64_t>> arglists(ranges.size());
282 std::size_t total = 1;
283 for (std::size_t i = 0; i < ranges.size(); i++) {
284 AddRange(&arglists[i], ranges[i].first, ranges[i].second,
285 range_multiplier_);
286 total *= arglists[i].size();
289 std::vector<std::size_t> ctr(arglists.size(), 0);
291 for (std::size_t i = 0; i < total; i++) {
292 std::vector<int64_t> tmp;
293 tmp.reserve(arglists.size());
295 for (std::size_t j = 0; j < arglists.size(); j++) {
296 tmp.push_back(arglists[j].at(ctr[j]));
299 args_.push_back(std::move(tmp));
301 for (std::size_t j = 0; j < arglists.size(); j++) {
302 if (ctr[j] + 1 < arglists[j].size()) {
303 ++ctr[j];
304 break;
306 ctr[j] = 0;
309 return this;
312 Benchmark* Benchmark::ArgName(const std::string& name) {
313 CHECK(ArgsCnt() == -1 || ArgsCnt() == 1);
314 arg_names_ = {name};
315 return this;
318 Benchmark* Benchmark::ArgNames(const std::vector<std::string>& names) {
319 CHECK(ArgsCnt() == -1 || ArgsCnt() == static_cast<int>(names.size()));
320 arg_names_ = names;
321 return this;
324 Benchmark* Benchmark::DenseRange(int64_t start, int64_t limit, int step) {
325 CHECK(ArgsCnt() == -1 || ArgsCnt() == 1);
326 CHECK_GE(start, 0);
327 CHECK_LE(start, limit);
328 for (int64_t arg = start; arg <= limit; arg += step) {
329 args_.push_back({arg});
331 return this;
334 Benchmark* Benchmark::Args(const std::vector<int64_t>& args) {
335 CHECK(ArgsCnt() == -1 || ArgsCnt() == static_cast<int>(args.size()));
336 args_.push_back(args);
337 return this;
340 Benchmark* Benchmark::Apply(void (*custom_arguments)(Benchmark* benchmark)) {
341 custom_arguments(this);
342 return this;
345 Benchmark* Benchmark::RangeMultiplier(int multiplier) {
346 CHECK(multiplier > 1);
347 range_multiplier_ = multiplier;
348 return this;
351 Benchmark* Benchmark::MinTime(double t) {
352 CHECK(t > 0.0);
353 CHECK(iterations_ == 0);
354 min_time_ = t;
355 return this;
358 Benchmark* Benchmark::Iterations(size_t n) {
359 CHECK(n > 0);
360 CHECK(IsZero(min_time_));
361 iterations_ = n;
362 return this;
365 Benchmark* Benchmark::Repetitions(int n) {
366 CHECK(n > 0);
367 repetitions_ = n;
368 return this;
371 Benchmark* Benchmark::ReportAggregatesOnly(bool value) {
372 report_mode_ = value ? RM_ReportAggregatesOnly : RM_Default;
373 return this;
376 Benchmark* Benchmark::UseRealTime() {
377 CHECK(!use_manual_time_)
378 << "Cannot set UseRealTime and UseManualTime simultaneously.";
379 use_real_time_ = true;
380 return this;
383 Benchmark* Benchmark::UseManualTime() {
384 CHECK(!use_real_time_)
385 << "Cannot set UseRealTime and UseManualTime simultaneously.";
386 use_manual_time_ = true;
387 return this;
390 Benchmark* Benchmark::Complexity(BigO complexity) {
391 complexity_ = complexity;
392 return this;
395 Benchmark* Benchmark::Complexity(BigOFunc* complexity) {
396 complexity_lambda_ = complexity;
397 complexity_ = oLambda;
398 return this;
401 Benchmark* Benchmark::ComputeStatistics(std::string name,
402 StatisticsFunc* statistics) {
403 statistics_.emplace_back(name, statistics);
404 return this;
407 Benchmark* Benchmark::Threads(int t) {
408 CHECK_GT(t, 0);
409 thread_counts_.push_back(t);
410 return this;
413 Benchmark* Benchmark::ThreadRange(int min_threads, int max_threads) {
414 CHECK_GT(min_threads, 0);
415 CHECK_GE(max_threads, min_threads);
417 AddRange(&thread_counts_, min_threads, max_threads, 2);
418 return this;
421 Benchmark* Benchmark::DenseThreadRange(int min_threads, int max_threads,
422 int stride) {
423 CHECK_GT(min_threads, 0);
424 CHECK_GE(max_threads, min_threads);
425 CHECK_GE(stride, 1);
427 for (auto i = min_threads; i < max_threads; i += stride) {
428 thread_counts_.push_back(i);
430 thread_counts_.push_back(max_threads);
431 return this;
434 Benchmark* Benchmark::ThreadPerCpu() {
435 thread_counts_.push_back(CPUInfo::Get().num_cpus);
436 return this;
439 void Benchmark::SetName(const char* name) { name_ = name; }
441 int Benchmark::ArgsCnt() const {
442 if (args_.empty()) {
443 if (arg_names_.empty()) return -1;
444 return static_cast<int>(arg_names_.size());
446 return static_cast<int>(args_.front().size());
449 //=============================================================================//
450 // FunctionBenchmark
451 //=============================================================================//
453 void FunctionBenchmark::Run(State& st) { func_(st); }
455 } // end namespace internal
457 void ClearRegisteredBenchmarks() {
458 internal::BenchmarkFamilies::GetInstance()->ClearBenchmarks();
461 } // end namespace benchmark