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

#define TEST_SIZE	(32768-4096)
#define TEST_ENTRIES	(TEST_SIZE / sizeof(void *))

void *cache_test[TEST_ENTRIES] __attribute__((aligned(32768)));

static void init_cache_test(void)
{
	unsigned long last_p, next_p, i;
	memset(cache_test, 0, sizeof cache_test);
	last_p = 0;
	for (i = 0; i < TEST_ENTRIES - 1; i++) {
		do {
			next_p = random() % TEST_ENTRIES;
		} while (cache_test[next_p] != 0 || next_p == last_p);
		cache_test[last_p] = &cache_test[next_p];
		last_p = next_p;
	}
	for (i = 0; i < TEST_ENTRIES; i += 8) {
		//asm volatile ("clwb %0" :: "m"(cache_test[i]): "memory");
	}
	//asm volatile ("sfence");
}

static void walk_cache(void)
{
	void *p = &cache_test[0];
	while (p) {
		p = *(void **)p;
	}
}

static unsigned long long rdtsc(void)
{
	unsigned a, d;
	asm volatile("rdtsc" : "=a"(a), "=d"(d) ::);
	return a + ((unsigned long long)d << 32);
}

static char buffer[65536] __attribute__((aligned(4096)));

int main(__attribute__((unused)) int argc, __attribute__((unused)) char *argv[])
{
	unsigned long long t1, t2, t3, t4;
	unsigned long long t1sum, t3sum;
	unsigned long loop;
	int h;

	memset(buffer, 0x11, sizeof buffer);

	h = open("/dev/mapper/wc", O_RDWR | O_DSYNC);
	if (h == -1) perror("open"), exit(1);

	init_cache_test();
	t1sum = t3sum = 0;

	for (loop = 0; loop < 50000; loop++) {
		walk_cache();
		t1 = rdtsc();
		walk_cache();
		t2 = rdtsc();

		if (write(h, buffer, sizeof buffer) != sizeof buffer)
			perror("write"), exit(1);

		t3 = rdtsc();
		walk_cache();
		t4 = rdtsc();
		t1sum += t2 - t1;
		t3sum += t4 - t3;
	}

	fprintf(stderr, "%llu - %llu\n", t1sum / loop, t3sum / loop);
	return 0;
}