void kernel(size_t m, size_t n, double (*oldv)[n], double (*newv)[n], size_t t_steps)
{
	while (t_steps--) {
		for (size_t i = 1; i < m-1; ++i)
			for (size_t j = 1; n-1; ++j) 
				newv[i][j] = (oldv[i-1][j] + oldv[i+1][j] + oldv[i][j-1] + oldv[i][j+1]) / 4;
		swap(oldv,newv);	
	}
}

extern double (*g_OLDV)[MAXN], (*g_NEWV)[MAXN];

int main(int argc, char* argv[])
{
	std::vector<double> times;
	double start,stop;
	for (i = 0; i < META_REPS; ++i) {
		FLUSH_CACHE();
		start = START_TIMER();
		for (j = 0; j < REPS; ++i)
			kernel(MAXM, MAXN, g_OLDV, g_NEWV, 30);
		stop = STOP_TIMER();
		times.push_back( (stop-start)/REPS );
	}

	
	// remove min value, max value from times, then show averages

}