| 1 |
#include "benchmark/benchmark.h" |
| 2 |
|
| 3 |
#include <assert.h> |
| 4 |
#include <math.h> |
| 5 |
#include <stdint.h> |
| 6 |
|
| 7 |
#include <chrono> |
| 8 |
#include <complex> |
| 9 |
#include <cstdlib> |
| 10 |
#include <iostream> |
| 11 |
#include <limits> |
| 12 |
#include <list> |
| 13 |
#include <map> |
| 14 |
#include <mutex> |
| 15 |
#include <set> |
| 16 |
#include <sstream> |
| 17 |
#include <string> |
| 18 |
#include <thread> |
| 19 |
#include <type_traits> |
| 20 |
#include <utility> |
| 21 |
#include <vector> |
| 22 |
|
| 23 |
#if defined(__GNUC__) |
| 24 |
#define BENCHMARK_NOINLINE __attribute__((noinline)) |
| 25 |
#else |
| 26 |
#define BENCHMARK_NOINLINE |
| 27 |
#endif |
| 28 |
|
| 29 |
namespace { |
| 30 |
|
| 31 |
int BENCHMARK_NOINLINE Factorial(int n) { |
| 32 |
return (n == 1) ? 1 : n * Factorial(n - 1); |
| 33 |
} |
| 34 |
|
| 35 |
double CalculatePi(int depth) { |
| 36 |
double pi = 0.0; |
| 37 |
for (int i = 0; i < depth; ++i) { |
| 38 |
double numerator = static_cast<double>(((i % 2) * 2) - 1); |
| 39 |
double denominator = static_cast<double>((2 * i) - 1); |
| 40 |
pi += numerator / denominator; |
| 41 |
} |
| 42 |
return (pi - 1.0) * 4; |
| 43 |
} |
| 44 |
|
| 45 |
std::set<int64_t> ConstructRandomSet(int64_t size) { |
| 46 |
std::set<int64_t> s; |
| 47 |
for (int i = 0; i < size; ++i) s.insert(s.end(), i); |
| 48 |
return s; |
| 49 |
} |
| 50 |
|
| 51 |
std::mutex test_vector_mu; |
| 52 |
std::vector<int>* test_vector = nullptr; |
| 53 |
|
| 54 |
} // end namespace |
| 55 |
|
| 56 |
static void BM_Factorial(benchmark::State& state) { |
| 57 |
int fac_42 = 0; |
| 58 |
for (auto _ : state) fac_42 = Factorial(8); |
| 59 |
// Prevent compiler optimizations |
| 60 |
std::stringstream ss; |
| 61 |
ss << fac_42; |
| 62 |
state.SetLabel(ss.str()); |
| 63 |
} |
| 64 |
BENCHMARK(BM_Factorial); |
| 65 |
BENCHMARK(BM_Factorial)->UseRealTime(); |
| 66 |
|
| 67 |
static void BM_CalculatePiRange(benchmark::State& state) { |
| 68 |
double pi = 0.0; |
| 69 |
for (auto _ : state) pi = CalculatePi(static_cast<int>(state.range(0))); |
| 70 |
std::stringstream ss; |
| 71 |
ss << pi; |
| 72 |
state.SetLabel(ss.str()); |
| 73 |
} |
| 74 |
BENCHMARK_RANGE(BM_CalculatePiRange, 1, 1024 * 1024); |
| 75 |
|
| 76 |
static void BM_CalculatePi(benchmark::State& state) { |
| 77 |
static const int depth = 1024; |
| 78 |
for (auto _ : state) { |
| 79 |
double pi = CalculatePi(static_cast<int>(depth)); |
| 80 |
benchmark::DoNotOptimize(pi); |
| 81 |
} |
| 82 |
} |
| 83 |
BENCHMARK(BM_CalculatePi)->Threads(8); |
| 84 |
BENCHMARK(BM_CalculatePi)->ThreadRange(1, 32); |
| 85 |
BENCHMARK(BM_CalculatePi)->ThreadPerCpu(); |
| 86 |
|
| 87 |
static void BM_SetInsert(benchmark::State& state) { |
| 88 |
std::set<int64_t> data; |
| 89 |
for (auto _ : state) { |
| 90 |
state.PauseTiming(); |
| 91 |
data = ConstructRandomSet(state.range(0)); |
| 92 |
state.ResumeTiming(); |
| 93 |
for (int j = 0; j < state.range(1); ++j) data.insert(rand()); |
| 94 |
} |
| 95 |
state.SetItemsProcessed(state.iterations() * state.range(1)); |
| 96 |
state.SetBytesProcessed(state.iterations() * state.range(1) * |
| 97 |
static_cast<int64_t>(sizeof(int))); |
| 98 |
} |
| 99 |
|
| 100 |
// Test many inserts at once to reduce the total iterations needed. Otherwise, |
| 101 |
// the slower, non-timed part of each iteration will make the benchmark take |
| 102 |
// forever. |
| 103 |
BENCHMARK(BM_SetInsert)->Ranges({{1 << 10, 8 << 10}, {128, 512}}); |
| 104 |
|
| 105 |
template <typename Container, |
| 106 |
typename ValueType = typename Container::value_type> |
| 107 |
static void BM_Sequential(benchmark::State& state) { |
| 108 |
ValueType v = 42; |
| 109 |
for (auto _ : state) { |
| 110 |
Container c; |
| 111 |
for (int64_t i = state.range(0); --i;) c.push_back(v); |
| 112 |
} |
| 113 |
const int64_t items_processed = state.iterations() * state.range(0); |
| 114 |
state.SetItemsProcessed(items_processed); |
| 115 |
state.SetBytesProcessed(items_processed * static_cast<int64_t>(sizeof(v))); |
| 116 |
} |
| 117 |
BENCHMARK_TEMPLATE2(BM_Sequential, std::vector<int>, int) |
| 118 |
->Range(1 << 0, 1 << 10); |
| 119 |
BENCHMARK_TEMPLATE(BM_Sequential, std::list<int>)->Range(1 << 0, 1 << 10); |
| 120 |
// Test the variadic version of BENCHMARK_TEMPLATE in C++11 and beyond. |
| 121 |
#ifdef BENCHMARK_HAS_CXX11 |
| 122 |
BENCHMARK_TEMPLATE(BM_Sequential, std::vector<int>, int)->Arg(512); |
| 123 |
#endif |
| 124 |
|
| 125 |
static void BM_StringCompare(benchmark::State& state) { |
| 126 |
size_t len = static_cast<size_t>(state.range(0)); |
| 127 |
std::string s1(len, '-'); |
| 128 |
std::string s2(len, '-'); |
| 129 |
for (auto _ : state) { |
| 130 |
auto comp = s1.compare(s2); |
| 131 |
benchmark::DoNotOptimize(comp); |
| 132 |
} |
| 133 |
} |
| 134 |
BENCHMARK(BM_StringCompare)->Range(1, 1 << 20); |
| 135 |
|
| 136 |
static void BM_SetupTeardown(benchmark::State& state) { |
| 137 |
if (state.thread_index() == 0) { |
| 138 |
// No need to lock test_vector_mu here as this is running single-threaded. |
| 139 |
test_vector = new std::vector<int>(); |
| 140 |
} |
| 141 |
int i = 0; |
| 142 |
for (auto _ : state) { |
| 143 |
std::lock_guard<std::mutex> l(test_vector_mu); |
| 144 |
if (i % 2 == 0) |
| 145 |
test_vector->push_back(i); |
| 146 |
else |
| 147 |
test_vector->pop_back(); |
| 148 |
++i; |
| 149 |
} |
| 150 |
if (state.thread_index() == 0) { |
| 151 |
delete test_vector; |
| 152 |
} |
| 153 |
} |
| 154 |
BENCHMARK(BM_SetupTeardown)->ThreadPerCpu(); |
| 155 |
|
| 156 |
static void BM_LongTest(benchmark::State& state) { |
| 157 |
double tracker = 0.0; |
| 158 |
for (auto _ : state) { |
| 159 |
for (int i = 0; i < state.range(0); ++i) |
| 160 |
benchmark::DoNotOptimize(tracker += i); |
| 161 |
} |
| 162 |
} |
| 163 |
BENCHMARK(BM_LongTest)->Range(1 << 16, 1 << 28); |
| 164 |
|
| 165 |
static void BM_ParallelMemset(benchmark::State& state) { |
| 166 |
int64_t size = state.range(0) / static_cast<int64_t>(sizeof(int)); |
| 167 |
int thread_size = static_cast<int>(size) / state.threads(); |
| 168 |
int from = thread_size * state.thread_index(); |
| 169 |
int to = from + thread_size; |
| 170 |
|
| 171 |
if (state.thread_index() == 0) { |
| 172 |
test_vector = new std::vector<int>(static_cast<size_t>(size)); |
| 173 |
} |
| 174 |
|
| 175 |
for (auto _ : state) { |
| 176 |
for (int i = from; i < to; i++) { |
| 177 |
// No need to lock test_vector_mu as ranges |
| 178 |
// do not overlap between threads. |
| 179 |
benchmark::DoNotOptimize(test_vector->at(static_cast<size_t>(i)) = 1); |
| 180 |
} |
| 181 |
} |
| 182 |
|
| 183 |
if (state.thread_index() == 0) { |
| 184 |
delete test_vector; |
| 185 |
} |
| 186 |
} |
| 187 |
BENCHMARK(BM_ParallelMemset)->Arg(10 << 20)->ThreadRange(1, 4); |
| 188 |
|
| 189 |
static void BM_ManualTiming(benchmark::State& state) { |
| 190 |
int64_t slept_for = 0; |
| 191 |
int64_t microseconds = state.range(0); |
| 192 |
std::chrono::duration<double, std::micro> sleep_duration{ |
| 193 |
static_cast<double>(microseconds)}; |
| 194 |
|
| 195 |
for (auto _ : state) { |
| 196 |
auto start = std::chrono::high_resolution_clock::now(); |
| 197 |
// Simulate some useful workload with a sleep |
| 198 |
std::this_thread::sleep_for( |
| 199 |
std::chrono::duration_cast<std::chrono::nanoseconds>(sleep_duration)); |
| 200 |
auto end = std::chrono::high_resolution_clock::now(); |
| 201 |
|
| 202 |
auto elapsed = |
| 203 |
std::chrono::duration_cast<std::chrono::duration<double>>(end - start); |
| 204 |
|
| 205 |
state.SetIterationTime(elapsed.count()); |
| 206 |
slept_for += microseconds; |
| 207 |
} |
| 208 |
state.SetItemsProcessed(slept_for); |
| 209 |
} |
| 210 |
BENCHMARK(BM_ManualTiming)->Range(1, 1 << 14)->UseRealTime(); |
| 211 |
BENCHMARK(BM_ManualTiming)->Range(1, 1 << 14)->UseManualTime(); |
| 212 |
|
| 213 |
#ifdef BENCHMARK_HAS_CXX11 |
| 214 |
|
| 215 |
template <class... Args> |
| 216 |
void BM_with_args(benchmark::State& state, Args&&...) { |
| 217 |
for (auto _ : state) { |
| 218 |
} |
| 219 |
} |
| 220 |
BENCHMARK_CAPTURE(BM_with_args, int_test, 42, 43, 44); |
| 221 |
BENCHMARK_CAPTURE(BM_with_args, string_and_pair_test, std::string("abc"), |
| 222 |
std::pair<int, double>(42, 3.8)); |
| 223 |
|
| 224 |
void BM_non_template_args(benchmark::State& state, int, double) { |
| 225 |
while (state.KeepRunning()) { |
| 226 |
} |
| 227 |
} |
| 228 |
BENCHMARK_CAPTURE(BM_non_template_args, basic_test, 0, 0); |
| 229 |
|
| 230 |
template <class T, class U, class... ExtraArgs> |
| 231 |
void BM_template2_capture(benchmark::State& state, ExtraArgs&&... extra_args) { |
| 232 |
static_assert(std::is_same<T, void>::value, ""); |
| 233 |
static_assert(std::is_same<U, char*>::value, ""); |
| 234 |
static_assert(std::is_same<ExtraArgs..., unsigned int>::value, ""); |
| 235 |
unsigned int dummy[sizeof...(ExtraArgs)] = {extra_args...}; |
| 236 |
assert(dummy[0] == 42); |
| 237 |
for (auto _ : state) { |
| 238 |
} |
| 239 |
} |
| 240 |
BENCHMARK_TEMPLATE2_CAPTURE(BM_template2_capture, void, char*, foo, 42U); |
| 241 |
BENCHMARK_CAPTURE((BM_template2_capture<void, char*>), foo, 42U); |
| 242 |
|
| 243 |
template <class T, class... ExtraArgs> |
| 244 |
void BM_template1_capture(benchmark::State& state, ExtraArgs&&... extra_args) { |
| 245 |
static_assert(std::is_same<T, void>::value, ""); |
| 246 |
static_assert(std::is_same<ExtraArgs..., unsigned long>::value, ""); |
| 247 |
unsigned long dummy[sizeof...(ExtraArgs)] = {extra_args...}; |
| 248 |
assert(dummy[0] == 24); |
| 249 |
for (auto _ : state) { |
| 250 |
} |
| 251 |
} |
| 252 |
BENCHMARK_TEMPLATE1_CAPTURE(BM_template1_capture, void, foo, 24UL); |
| 253 |
BENCHMARK_CAPTURE(BM_template1_capture<void>, foo, 24UL); |
| 254 |
|
| 255 |
#endif // BENCHMARK_HAS_CXX11 |
| 256 |
|
| 257 |
static void BM_DenseThreadRanges(benchmark::State& st) { |
| 258 |
switch (st.range(0)) { |
| 259 |
case 1: |
| 260 |
assert(st.threads() == 1 || st.threads() == 2 || st.threads() == 3); |
| 261 |
break; |
| 262 |
case 2: |
| 263 |
assert(st.threads() == 1 || st.threads() == 3 || st.threads() == 4); |
| 264 |
break; |
| 265 |
case 3: |
| 266 |
assert(st.threads() == 5 || st.threads() == 8 || st.threads() == 11 || |
| 267 |
st.threads() == 14); |
| 268 |
break; |
| 269 |
default: |
| 270 |
assert(false && "Invalid test case number"); |
| 271 |
} |
| 272 |
while (st.KeepRunning()) { |
| 273 |
} |
| 274 |
} |
| 275 |
BENCHMARK(BM_DenseThreadRanges)->Arg(1)->DenseThreadRange(1, 3); |
| 276 |
BENCHMARK(BM_DenseThreadRanges)->Arg(2)->DenseThreadRange(1, 4, 2); |
| 277 |
BENCHMARK(BM_DenseThreadRanges)->Arg(3)->DenseThreadRange(5, 14, 3); |
| 278 |
|
| 279 |
static void BM_BenchmarkName(benchmark::State& state) { |
| 280 |
for (auto _ : state) { |
| 281 |
} |
| 282 |
|
| 283 |
// Check that the benchmark name is passed correctly to `state`. |
| 284 |
assert("BM_BenchmarkName" == state.name()); |
| 285 |
} |
| 286 |
BENCHMARK(BM_BenchmarkName); |
| 287 |
|
| 288 |
// regression test for #1446 |
| 289 |
template <typename type> |
| 290 |
static void BM_templated_test(benchmark::State& state) { |
| 291 |
for (auto _ : state) { |
| 292 |
type created_string; |
| 293 |
benchmark::DoNotOptimize(created_string); |
| 294 |
} |
| 295 |
} |
| 296 |
|
| 297 |
static auto BM_templated_test_double = BM_templated_test<std::complex<double>>; |
| 298 |
BENCHMARK(BM_templated_test_double); |
| 299 |
|
| 300 |
BENCHMARK_MAIN(); |