| 1 | //===----------------------------------------------------------------------===// |
| 2 | // |
| 3 | // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. |
| 4 | // See https://llvm.org/LICENSE.txt for license information. |
| 5 | // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception |
| 6 | // |
| 7 | //===----------------------------------------------------------------------===// |
| 8 | |
| 9 | // UNSUPPORTED: c++03, c++11, c++14, c++17 |
| 10 | |
| 11 | #include "atomic_wait_helper.h" |
| 12 | |
| 13 | #include <atomic> |
| 14 | #include <cstdint> |
| 15 | #include <numeric> |
| 16 | #include <stop_token> |
| 17 | #include <pthread.h> |
| 18 | #include <sched.h> |
| 19 | #include <thread> |
| 20 | #include <chrono> |
| 21 | #include <array> |
| 22 | |
| 23 | #include "benchmark/benchmark.h" |
| 24 | #include "make_test_thread.h" |
| 25 | |
| 26 | using namespace std::chrono_literals; |
| 27 | |
| 28 | template <class NotifyPolicy, class NumberOfAtomics, class NumPrioTasks> |
| 29 | void BM_N_atomics_N_waiter_N_notifier(benchmark::State& state) { |
| 30 | [[maybe_unused]] std::array<HighPrioTask, NumPrioTasks::value> tasks{}; |
| 31 | const std::uint64_t total_loop_test_param = state.range(0); |
| 32 | constexpr std::uint64_t num_atomics = NumberOfAtomics::value; |
| 33 | std::vector<std::atomic<std::uint64_t>> atomics(num_atomics); |
| 34 | |
| 35 | auto notify_func = [&](std::stop_token st, size_t idx) { |
| 36 | while (!st.stop_requested()) { |
| 37 | NotifyPolicy::notify(atomics[idx], st); |
| 38 | } |
| 39 | }; |
| 40 | |
| 41 | std::atomic<std::uint64_t> start_flag = 0; |
| 42 | std::atomic<std::uint64_t> done_count = 0; |
| 43 | |
| 44 | auto wait_func = [&, total_loop_test_param](std::stop_token st, size_t idx) { |
| 45 | auto old_start = 0; |
| 46 | while (!st.stop_requested()) { |
| 47 | start_flag.wait(old_start); |
| 48 | old_start = start_flag.load(); |
| 49 | for (std::uint64_t i = 0; i < total_loop_test_param; ++i) { |
| 50 | auto old = atomics[idx].load(std::memory_order_relaxed); |
| 51 | atomics[idx].wait(old); |
| 52 | } |
| 53 | done_count.fetch_add(i: 1); |
| 54 | } |
| 55 | }; |
| 56 | |
| 57 | std::vector<std::jthread> notify_threads; |
| 58 | notify_threads.reserve(num_atomics); |
| 59 | |
| 60 | std::vector<std::jthread> wait_threads; |
| 61 | wait_threads.reserve(num_atomics); |
| 62 | |
| 63 | for (size_t i = 0; i < num_atomics; ++i) { |
| 64 | notify_threads.emplace_back(support::make_test_jthread(notify_func, i)); |
| 65 | } |
| 66 | |
| 67 | for (size_t i = 0; i < num_atomics; ++i) { |
| 68 | wait_threads.emplace_back(support::make_test_jthread(wait_func, i)); |
| 69 | } |
| 70 | |
| 71 | for (auto _ : state) { |
| 72 | done_count = 0; |
| 73 | start_flag.fetch_add(1); |
| 74 | start_flag.notify_all(); |
| 75 | while (done_count < num_atomics) { |
| 76 | std::this_thread::yield(); |
| 77 | } |
| 78 | } |
| 79 | for (auto& t : wait_threads) { |
| 80 | t.request_stop(); |
| 81 | } |
| 82 | start_flag.fetch_add(i: 1); |
| 83 | start_flag.notify_all(); |
| 84 | for (auto& t : wait_threads) { |
| 85 | t.join(); |
| 86 | } |
| 87 | } |
| 88 | |
| 89 | BENCHMARK(BM_N_atomics_N_waiter_N_notifier<KeepNotifying, NumberOfAtomics<2>, NumHighPrioTasks<0>>) |
| 90 | ->RangeMultiplier(2) |
| 91 | ->Range(1 << 12, 1 << 14); |
| 92 | BENCHMARK(BM_N_atomics_N_waiter_N_notifier<KeepNotifying, NumberOfAtomics<3>, NumHighPrioTasks<0>>) |
| 93 | ->RangeMultiplier(2) |
| 94 | ->Range(1 << 10, 1 << 12); |
| 95 | BENCHMARK(BM_N_atomics_N_waiter_N_notifier<KeepNotifying, NumberOfAtomics<5>, NumHighPrioTasks<0>>) |
| 96 | ->RangeMultiplier(2) |
| 97 | ->Range(1 << 10, 1 << 12); |
| 98 | BENCHMARK(BM_N_atomics_N_waiter_N_notifier<KeepNotifying, NumberOfAtomics<7>, NumHighPrioTasks<0>>) |
| 99 | ->RangeMultiplier(2) |
| 100 | ->Range(1 << 8, 1 << 10); |
| 101 | |
| 102 | BENCHMARK(BM_N_atomics_N_waiter_N_notifier<NotifyEveryNus<50>, NumberOfAtomics<2>, NumHighPrioTasks<0>>) |
| 103 | ->RangeMultiplier(2) |
| 104 | ->Range(1 << 10, 1 << 12); |
| 105 | BENCHMARK(BM_N_atomics_N_waiter_N_notifier<NotifyEveryNus<50>, NumberOfAtomics<3>, NumHighPrioTasks<0>>) |
| 106 | ->RangeMultiplier(2) |
| 107 | ->Range(1 << 8, 1 << 10); |
| 108 | BENCHMARK(BM_N_atomics_N_waiter_N_notifier<NotifyEveryNus<50>, NumberOfAtomics<5>, NumHighPrioTasks<0>>) |
| 109 | ->RangeMultiplier(2) |
| 110 | ->Range(1 << 8, 1 << 10); |
| 111 | BENCHMARK(BM_N_atomics_N_waiter_N_notifier<NotifyEveryNus<50>, NumberOfAtomics<7>, NumHighPrioTasks<0>>) |
| 112 | ->RangeMultiplier(2) |
| 113 | ->Range(1 << 6, 1 << 8); |
| 114 | |
| 115 | BENCHMARK(BM_N_atomics_N_waiter_N_notifier<NotifyEveryNus<100>, NumberOfAtomics<2>, NumHighPrioTasks<0>>) |
| 116 | ->RangeMultiplier(2) |
| 117 | ->Range(1 << 8, 1 << 10); |
| 118 | BENCHMARK(BM_N_atomics_N_waiter_N_notifier<NotifyEveryNus<100>, NumberOfAtomics<3>, NumHighPrioTasks<0>>) |
| 119 | ->RangeMultiplier(2) |
| 120 | ->Range(1 << 8, 1 << 10); |
| 121 | BENCHMARK(BM_N_atomics_N_waiter_N_notifier<NotifyEveryNus<100>, NumberOfAtomics<5>, NumHighPrioTasks<0>>) |
| 122 | ->RangeMultiplier(2) |
| 123 | ->Range(1 << 7, 1 << 9); |
| 124 | BENCHMARK(BM_N_atomics_N_waiter_N_notifier<NotifyEveryNus<100>, NumberOfAtomics<7>, NumHighPrioTasks<0>>) |
| 125 | ->RangeMultiplier(2) |
| 126 | ->Range(1 << 6, 1 << 8); |
| 127 | |
| 128 | BENCHMARK(BM_N_atomics_N_waiter_N_notifier<KeepNotifying, NumberOfAtomics<2>, NumHighPrioTasks<4>>) |
| 129 | ->RangeMultiplier(2) |
| 130 | ->Range(1 << 7, 1 << 9); |
| 131 | BENCHMARK(BM_N_atomics_N_waiter_N_notifier<KeepNotifying, NumberOfAtomics<3>, NumHighPrioTasks<4>>) |
| 132 | ->RangeMultiplier(2) |
| 133 | ->Range(1 << 7, 1 << 9); |
| 134 | BENCHMARK(BM_N_atomics_N_waiter_N_notifier<KeepNotifying, NumberOfAtomics<5>, NumHighPrioTasks<4>>) |
| 135 | ->RangeMultiplier(2) |
| 136 | ->Range(1 << 6, 1 << 8); |
| 137 | BENCHMARK(BM_N_atomics_N_waiter_N_notifier<KeepNotifying, NumberOfAtomics<7>, NumHighPrioTasks<4>>) |
| 138 | ->RangeMultiplier(2) |
| 139 | ->Range(1 << 4, 1 << 6); |
| 140 | |
| 141 | BENCHMARK(BM_N_atomics_N_waiter_N_notifier<NotifyEveryNus<50>, NumberOfAtomics<2>, NumHighPrioTasks<4>>) |
| 142 | ->RangeMultiplier(2) |
| 143 | ->Range(1 << 7, 1 << 9); |
| 144 | BENCHMARK(BM_N_atomics_N_waiter_N_notifier<NotifyEveryNus<50>, NumberOfAtomics<3>, NumHighPrioTasks<4>>) |
| 145 | ->RangeMultiplier(2) |
| 146 | ->Range(1 << 7, 1 << 9); |
| 147 | BENCHMARK(BM_N_atomics_N_waiter_N_notifier<NotifyEveryNus<50>, NumberOfAtomics<5>, NumHighPrioTasks<4>>) |
| 148 | ->RangeMultiplier(2) |
| 149 | ->Range(1 << 5, 1 << 7); |
| 150 | BENCHMARK(BM_N_atomics_N_waiter_N_notifier<NotifyEveryNus<50>, NumberOfAtomics<7>, NumHighPrioTasks<4>>) |
| 151 | ->RangeMultiplier(2) |
| 152 | ->Range(1 << 3, 1 << 5); |
| 153 | |
| 154 | BENCHMARK(BM_N_atomics_N_waiter_N_notifier<NotifyEveryNus<100>, NumberOfAtomics<2>, NumHighPrioTasks<4>>) |
| 155 | ->RangeMultiplier(2) |
| 156 | ->Range(1 << 6, 1 << 8); |
| 157 | BENCHMARK(BM_N_atomics_N_waiter_N_notifier<NotifyEveryNus<100>, NumberOfAtomics<3>, NumHighPrioTasks<4>>) |
| 158 | ->RangeMultiplier(2) |
| 159 | ->Range(1 << 6, 1 << 8); |
| 160 | BENCHMARK(BM_N_atomics_N_waiter_N_notifier<NotifyEveryNus<100>, NumberOfAtomics<5>, NumHighPrioTasks<4>>) |
| 161 | ->RangeMultiplier(2) |
| 162 | ->Range(1 << 5, 1 << 7); |
| 163 | BENCHMARK(BM_N_atomics_N_waiter_N_notifier<NotifyEveryNus<100>, NumberOfAtomics<7>, NumHighPrioTasks<4>>) |
| 164 | ->RangeMultiplier(2) |
| 165 | ->Range(1 << 3, 1 << 5); |
| 166 | |
| 167 | BENCHMARK_MAIN(); |
| 168 | |