semaphore_benchmark.cpp

#include <benchmark/benchmark.h>

#include <atomic>
#include <thread>

#include <userver/engine/async.hpp>
#include <userver/engine/run_standalone.hpp>
#include <userver/engine/semaphore.hpp>
#include <userver/engine/sleep.hpp>
#include <utils/impl/parallelize_benchmark.hpp>

USERVER_NAMESPACE_BEGIN

/// [RunStandalone sample]
void semaphore_lock(benchmark::State& state) {
    engine::RunStandalone([&]() {
        std::size_t i = 0;
        engine::Semaphore sem{std::numeric_limits<std::size_t>::max()};

        for ([[maybe_unused]] auto _ : state) {
            sem.lock_shared();
            ++i;
        }

        for (std::size_t j = 0; j < i; ++j) {
            sem.unlock_shared();
        }
    });
}
BENCHMARK(semaphore_lock);
/// [RunStandalone sample]

void semaphore_unlock(benchmark::State& state) {
    engine::RunStandalone([&]() {
        unsigned i = 0;
        engine::Semaphore sem{std::numeric_limits<std::size_t>::max()};

        auto lock_multiple = [&i, &sem]() {
            for (; i < 1024; ++i) {
                sem.lock_shared();
            }
        };
        lock_multiple();

        for ([[maybe_unused]] auto _ : state) {
            sem.unlock_shared();

            if (--i == 0) {
                state.PauseTiming();
                lock_multiple();
                state.ResumeTiming();
            }
        }

        while (i) {
            sem.unlock_shared();
            --i;
        }
    });
}
BENCHMARK(semaphore_unlock);

void semaphore_lock_unlock_contention(benchmark::State& state) {
    engine::RunStandalone(state.range(0), [&] {
        engine::Semaphore sem{1};

        RunParallelBenchmark(state, [&](auto& range) {
            for ([[maybe_unused]] auto _ : range) {
                sem.lock_shared();
                sem.unlock_shared();
            }
        });
    });
}
BENCHMARK(semaphore_lock_unlock_contention)->RangeMultiplier(2)->Range(1, 32);

void semaphore_lock_unlock_payload_contention(benchmark::State& state) {
    engine::RunStandalone(state.range(0), [&] {
        engine::Semaphore sem{1};

        RunParallelBenchmark(state, [&](auto& range) {
            for ([[maybe_unused]] auto _ : range) {
                sem.lock_shared();
                {
                    std::vector<int> tmp(32, 32);
                    benchmark::DoNotOptimize(tmp);
                }
                sem.unlock_shared();
            }
        });
    });
}
BENCHMARK(semaphore_lock_unlock_payload_contention)->RangeMultiplier(2)->Range(1, 32);

void semaphore_lock_unlock_coro_contention(benchmark::State& state) {
    engine::RunStandalone(4, [&] {
        engine::Semaphore sem{1};

        RunParallelBenchmark(state, [&](auto& range) {
            for ([[maybe_unused]] auto _ : range) {
                sem.lock_shared();
                sem.unlock_shared();
            }
        });
    });
}
BENCHMARK(semaphore_lock_unlock_coro_contention)->RangeMultiplier(2)->Range(1, 1024);

void semaphore_lock_unlock_payload_coro_contention(benchmark::State& state) {
    engine::RunStandalone(4, [&] {
        engine::Semaphore sem{1};

        RunParallelBenchmark(state, [&](auto& range) {
            for ([[maybe_unused]] auto _ : range) {
                sem.lock_shared();
                {
                    std::vector<int> tmp(32, 32);
                    benchmark::DoNotOptimize(tmp);
                }
                sem.unlock_shared();
            }
        });
    });
}
BENCHMARK(semaphore_lock_unlock_payload_coro_contention)->RangeMultiplier(2)->Range(1, 1024);

void semaphore_lock_unlock_st_coro_contention(benchmark::State& state) {
    engine::RunStandalone([&]() {
        engine::Semaphore sem{1};

        RunParallelBenchmark(state, [&](auto& range) {
            for ([[maybe_unused]] auto _ : range) {
                sem.lock_shared();
                engine::Yield();
                sem.unlock_shared();
            }
        });
    });
}
BENCHMARK(semaphore_lock_unlock_st_coro_contention)->RangeMultiplier(2)->Range(1, 1024);

USERVER_NAMESPACE_END