#include <stdio.h>
#include <stdlib.h>
#include <time.h>
// 256 bytes = 4 * 64 = 4 * cachelines
// with 4-way associative cache one row
// traverse each set.
// With 8-way associative cache the col number
// should be doubled.
#define COLS 256
#define ROWS 1024
#define VFL 6291456
#ifndef OFF
#define OFF 0
#endif

char arr[ROWS][COLS+OFF*4] __attribute__((aligned(4096)));
char arr2[VFL];

int main() {
	int i, j, k;
	time_t t;
	volatile int xor;
	srand(0x42069);
	// init array with random values
	for (i = 0; i < ROWS; i++) {
		for (j = 0; j < COLS; j++) {
			arr[i][j] = rand()%255;
		}
	}

	// "flush" 6MiB cache
	// for(i = 0; i < VFL; i++)arr2[i] = i*i;
	// for(i = 0; i < VFL; i++)arr2[i] += 1;

	// mem barrier
	__sync_synchronize();
	t = clock();
	for (i = 1; i < ROWS; i++) {
		for (j = 0; j < COLS; j++) {
			arr[i][j] *= arr[i-1][j];
			xor ^= arr[i][j];
		}
	}
	t = clock() - t;
	printf("%ld",t);
}