https://stackoverflow.com/q/52357616/6789498

so-52357616.c

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

#define NNN 2048
#define BENCH_COUNT 10
#define timespecsub(tsp, usp, vsp)                        \
    do                                                    \
    {                                                     \
        (vsp)->tv_sec = (tsp)->tv_sec - (usp)->tv_sec;    \
        (vsp)->tv_nsec = (tsp)->tv_nsec - (usp)->tv_nsec; \
        if ((vsp)->tv_nsec < 0)                           \
        {                                                 \
            (vsp)->tv_sec--;                              \
            (vsp)->tv_nsec += 1000000000L;                \
        }                                                 \
    } while (0)

static void benchmark(void (*f)(void), const char *name);
static void copyij();
static void copyji();
static void init_mat();

static volatile int src[NNN][NNN], dst[NNN][NNN];

int main(void)
{
    size_t i;
    printf("ij:\n");
    for (i = 0; i < BENCH_COUNT; i++)
    {
        benchmark(copyij, "copyij");
    }
    printf("ji:\n");
    for (i = 0; i < BENCH_COUNT; i++)
    {
        benchmark(copyji, "copyji");
    }
}

static void
benchmark(void (*f)(void), const char *name)
{
    struct timespec start, end, delta;

    init_mat();
    assert(clock_gettime(CLOCK_MONOTONIC, &start) != -1);
    f();
    assert(clock_gettime(CLOCK_MONOTONIC, &end) != -1);

    timespecsub(&end, &start, &delta);
    printf("%s: NNN=%d: elapsed=%lld.%.9ld secs\n", name, NNN, delta.tv_sec,
           delta.tv_nsec);
}

#pragma GCC push_options
#pragma GCC optimize("O0")

static void
copyij()
{
    size_t i, j;

    for (i = 0; i < NNN; i++)
    {
        for (j = 0; j < NNN; j++)
        {
            src[i][j] += 1;
        }
    }
}

static void
copyji()
{
    size_t i, j;

    for (i = 0; i < NNN; i++)
    {
        for (j = 0; j < NNN; j++)
        {
            dst[j][i] += 1;
        }
    }
}

#pragma GCC pop_options

static void
init_mat()
{
    size_t i, j;

    for (i = 0; i < NNN; i++)
    {
        for (j = 0; j < NNN; j++)
        {
            src[i][j] = dst[i][j] = 1;
        }
    }
}