foolish.c

#include <stdlib.h>
#include <stdio.h>
#include <time.h>

typedef int T;

T foo_local_array(T a, T b, T c, int i) {
    const T abc[3] = { a, b, c};
    return abc[i];
}

T foo_switch_case(T a, T b, T c, int i) {
    switch (i) {
        case 0: return a;
        case 1: return b;
        case 2: return c;
        default: __builtin_unreachable();
    }
}

T foo_if_else_ret(T a, T b, T c, int i) {
    if (i == 0)
        return a;
    else if (i == 1)
        return b;
    return c;
}

T foo_select_nobr(T a, T b, T c, int i) {
    return a + (i & 1) * (b - a) + (i > 1) * (c - a);
}

__attribute__((noinline)) long bench(int N, T* buf, int* indices, char* name, T (*fn)(T, T, T, int)) {
    struct timespec ts;
    timespec_get(&ts, TIME_UTC);
    srand(ts.tv_nsec);

    T acc = 0, j = 0;
    for (int i = 0; i < N; i++) {
        j = buf[3 * (N - i) - 1] % N;
        //j = i;
        acc += fn(buf[3 * j + 0], buf[3 * j + 1], buf[3 * j + 2], indices[j]);
    }

    struct timespec te;
    timespec_get(&te, TIME_UTC);

    long end_us   = te.tv_sec * 1000000 + te.tv_nsec / 1000;
    long start_us = ts.tv_sec * 1000000 + ts.tv_nsec / 1000;
    long delta = end_us - start_us;
    printf("ran '%s' took %dus acc = %d \n", name, (int) (delta), acc);
    return delta;
}

int main(int argc, char** argv) {
    struct timespec tss;
    timespec_get(&tss, TIME_UTC);
    srand(tss.tv_nsec);

    int N = 65536 * 16;
    int R = 16;
    
    T* buf = malloc(N * 3 * sizeof(T));
    int* indices = malloc(N * sizeof(int));
    for (int i = 0; i < N; i++) {
        buf[i * 3 + 0] = rand();
        buf[i * 3 + 1] = rand();
        buf[i * 3 + 2] = rand();
        indices[i] = rand() % 3;
    }

    long acc[4] = { 0 };
    char* str[4] = { "foo_local_array", "foo_if_else_ret", "foo_switch_case", "foo_select_nobr" };
    T (*fns[4])(T, T, T, int) = { foo_local_array, foo_if_else_ret, foo_switch_case, foo_select_nobr };

    for (int r = 0; r < R; r++) {
        for (int t = 0; t < sizeof(acc) / sizeof(acc[0]); t++)
          acc[t] += bench(N, buf, indices, str[t], fns[t]);
    }

    printf("Final results (%d iterations):\n", R);
    for (int t = 0; t < sizeof(acc) / sizeof(acc[0]); t++) {
        printf("Test '%s', took %dus on average\n", str[t], acc[t] / R);
    }
}