C++ - [Concurrency 8] `std::this_thread::yield()`

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
// cpu_yield_vs_sleep_bench.cpp
#include <chrono>
#include <ctime>
#include <iomanip>
#include <iostream>
#include <string>
#include <thread>

using clock_tp = std::chrono::steady_clock::time_point;

struct Result {
    double wall_seconds{};
    double cpu_seconds{};
    double cpu_util_percent{};
    double iters_per_sec{};
    unsigned long long iters{};
};

Result run_yield_test(double seconds) {
    const auto t0 = std::chrono::steady_clock::now();
    const std::clock_t c0 = std::clock();

    unsigned long long iters = 0;
    while (true) {
        std::this_thread::yield();
        ++iters;

        if ((std::chrono::steady_clock::now() - t0) >= std::chrono::duration<double>(seconds))
            break;
    }

    const auto t1 = std::chrono::steady_clock::now();
    const std::clock_t c1 = std::clock();

    const double wall = std::chrono::duration<double>(t1 - t0).count();
    const double cpu  = double(c1 - c0) / CLOCKS_PER_SEC;

    Result r;
    r.wall_seconds = wall;
    r.cpu_seconds = cpu;
    r.cpu_util_percent = (wall > 0) ? (cpu / wall) * 100.0 : 0.0;
    r.iters = iters;
    r.iters_per_sec = (wall > 0) ? iters / wall : 0.0;
    return r;
}

Result run_sleep_test(double seconds, std::chrono::nanoseconds sleep_ns) {
    const auto t0 = std::chrono::steady_clock::now();
    const std::clock_t c0 = std::clock();

    unsigned long long iters = 0;
    while (true) {
        std::this_thread::sleep_for(sleep_ns);
        ++iters;

        if ((std::chrono::steady_clock::now() - t0) >= std::chrono::duration<double>(seconds))
            break;
    }

    const auto t1 = std::chrono::steady_clock::now();
    const std::clock_t c1 = std::clock();

    const double wall = std::chrono::duration<double>(t1 - t0).count();
    const double cpu  = double(c1 - c0) / CLOCKS_PER_SEC;

    Result r;
    r.wall_seconds = wall;
    r.cpu_seconds = cpu;
    r.cpu_util_percent = (wall > 0) ? (cpu / wall) * 100.0 : 0.0;
    r.iters = iters;
    r.iters_per_sec = (wall > 0) ? iters / wall : 0.0;
    return r;
}

void print_result(const std::string& name, const Result& r) {
    std::cout << std::fixed << std::setprecision(3);
    std::cout << "=== " << name << " ===\n";
    std::cout << "Wall time (s):        " << r.wall_seconds << "\n";
    std::cout << "CPU time (s):         " << r.cpu_seconds << "\n";
    std::cout << "CPU utilization (%):  " << r.cpu_util_percent << "\n";
    std::cout << "Iterations:           " << r.iters << "\n";
    std::cout << "Iters/sec:            " << r.iters_per_sec << "\n\n";
}

int main(int argc, char** argv) {
    // Defaults
    double duration_sec = 3.0;
    long long sleep_ns = 1'000'000; // 1 ms

    // Simple CLI:
    //   cpu_yield_vs_sleep_bench [duration_sec] [sleep_ns]
    if (argc >= 2) duration_sec = std::stod(argv[1]);
    if (argc >= 3) sleep_ns = std::stoll(argv[2]);

    std::cout << "Benchmark duration: " << duration_sec << " s\n";
    std::cout << "Sleep interval:     " << sleep_ns << " ns\n\n";

    // Warm-up to stabilize scheduling
    std::this_thread::sleep_for(std::chrono::milliseconds(200));

    auto ry = run_yield_test(duration_sec);
    print_result("this_thread::yield()", ry);

    auto rs = run_sleep_test(duration_sec, std::chrono::nanoseconds(sleep_ns));
    print_result("sleep_for(" + std::to_string(sleep_ns) + " ns)", rs);

    std::cout << "Note:\n"
              << " - On lightly loaded systems, yield() often shows very high CPU% (busy-wait style),\n"
              << "   whereas sleep_for() should be near 0% CPU during the sleep intervals.\n"
              << " - Actual numbers depend on OS scheduler, timer resolution, and system load.\n";
    return 0;
}