sudo sysctl -w vm.nr_hugepages=256; gcc -O2 cache.c -o cache

cache.c

#include <fcntl.h>
#include <stdint.h>
#include <stdio.h>
#include <stdlib.h>
#include <sys/mman.h>
#include <time.h>
#include <unistd.h>

#define BASE_SIZE 128 // array base size = 1KB
#define MAX_SIZE 17   // array size ranges from 1KB to 64MB
#define LG_STRIDE 0   // stride = 1
#define MAX_STRIDE 7  // stide ranges from 1 to 64
#define TOTAL_RUNS 10
#define ITERS_PER_RUN (1<<25)

static int RunOnce(size_t n, size_t s) {
    struct timespec begin, end;
    uint64_t *arr;
    if ((n * sizeof(uint64_t)) < (1 << 21)) {
        arr = (uint64_t *) calloc(n, sizeof(uint64_t));
    } else {
        arr = aligned_alloc(1 << 21, n * sizeof(uint64_t));
        madvise(arr, n * sizeof(uint64_t), MADV_HUGEPAGE);
    }
    int fd = open("/proc/sys/vm/drop_caches", O_WRONLY);
    write(fd, "3", 1);
    clock_gettime(CLOCK_MONOTONIC, &begin);
    for (size_t i = 0; i < ITERS_PER_RUN; ++i) {
        arr[(i << s) & (n - 1)] += 0xdeadbeaf;
    }
    clock_gettime(CLOCK_MONOTONIC, &end);
    free(arr);
    close(fd);
    return (1000000000ULL * (end.tv_sec - begin.tv_sec) + end.tv_nsec - begin.tv_nsec) / 1000ULL;
}

static int Run(size_t n, size_t s) {
    int ret = 0;
    for (size_t i = 0; i < TOTAL_RUNS; ++i) {
        ret += RunOnce(n, s) / TOTAL_RUNS;
    }
    return ret;
}

int main() {
    for(size_t s = 0; s < MAX_STRIDE; ++s) {
        for (size_t i = 0; i < MAX_SIZE; ++i) {
            int res = Run(BASE_SIZE << i, LG_STRIDE+s);
            fprintf(stderr, "%d\t", res);
        }
        fprintf(stderr, "\n");
    }
}