#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <fcntl.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);
}

#if 1
#define LEN	(1UL*512*1024*1024)
#else
#define LEN	(1UL*4096)
#endif

static int h;
static char *m;

#define m_l	((unsigned long *)m)
#define LEN_L	(LEN / sizeof(unsigned long))

static void map(char *p)
{
	unsigned long len = LEN;
	if (len < 2 * 1024 * 1024)
		len = 2 * 1024 * 1024;
	h = open(p, O_RDWR);
	if (h < 0)
		perror(p), exit(1);
	m = mmap(NULL, len, PROT_READ | PROT_WRITE, MAP_SHARED_VALIDATE, h, 0);
	if (m == MAP_FAILED)
		perror("mmap"), exit(1);
	memset(m, 0, LEN);
}

int main(__attribute__((unused)) int argc, __attribute__((unused)) char *argv[])
{
	unsigned long long t1, t2, t3, t4;
	unsigned long i;
	char mode;

	if (argc != 3)
		fprintf(stderr, "args\n"), exit(1);

	map(argv[1]);

	mode = argv[2][0];

	init_cache_test();
	walk_cache();
	t1 = rdtsc();
	walk_cache();
	t2 = rdtsc();

	if (mode == 'f') {
		for (i = 0; i < LEN_L; i += 8) {
			m_l[i+0] = 0x12345;
			m_l[i+1] = 0x12345;
			m_l[i+2] = 0x12345;
			m_l[i+3] = 0x12345;
			m_l[i+4] = 0x12345;
			m_l[i+5] = 0x12345;
			m_l[i+6] = 0x12345;
			m_l[i+7] = 0x12345;
			asm volatile ("clflushopt %0" :: "m"(m_l[i]): "memory");
		}
	} else if (mode == 'w') {
		for (i = 0; i < LEN_L; i += 8) {
			m_l[i+0] = 0x12345;
			m_l[i+1] = 0x12345;
			m_l[i+2] = 0x12345;
			m_l[i+3] = 0x12345;
			m_l[i+4] = 0x12345;
			m_l[i+5] = 0x12345;
			m_l[i+6] = 0x12345;
			m_l[i+7] = 0x12345;
			asm volatile ("clwb %0" :: "m"(m_l[i]): "memory");
		}
	} else if (mode == 'n') {
		for (i = 0; i < LEN; i += 8) {
			asm volatile ("movntiq %%rax, %0" : "=m"(m[i]) :: "rax", "memory");
		}
	} else if (mode == 'm' ) {
		memset(m_l, 0, LEN);
	} else if (mode == 'c' ) {
		long l = atol(argv[2] + 1);
		while (l) {
			m_l[--l] = 0x12345;
		}
	}

	t3 = rdtsc();
	walk_cache();
	t4 = rdtsc();
	fprintf(stderr, "%llu - %llu\n", t2 - t1, t4 - t3);
}