#include <cstdlib>
#include <cstring>

#define LOC(m, i, j, row_size) m[(i) * (row_size) + (j)]

void matrix_mul1(int* a, int* b, int* c, int size)
{
    int res = 0, i, j, k;
    for (i = 0; i < size; ++i)
    {
        for (j = 0; j < size; ++j)
        {
            res = 0;
            for (k = 0; k < size; ++k)
            {
                res += LOC(a, i, k, size) * LOC(b, k, j, size);
            }
            LOC(c, i, j, size) = res;
        }
    }
}

void matrix_mul2(int* a, int* b, int* c, int size)
{
    int tmp = 0, i, j, k;
    for (j = 0; j < size; ++j)
    {
        for (k = 0; k < size; ++k)
        {
            tmp = LOC(b, k, j, size);
            for (i = 0; i < size; ++i)
            {
                LOC(c, i, j, size) += LOC(a, i, k, size) * tmp;
            }
        }
    }
}

// This is the one with least cache miss
void matrix_mul3(int* a, int* b, int* c, int size)
{
    int tmp = 0, i, j, k;
    for (k = 0; k < size; ++k)
    {
        for (i = 0; i < size; ++i)
        {
            tmp = LOC(a, i, k, size);
            for (j = 0; j < size; ++j)
            {
                LOC(c, i, j, size) += tmp * LOC(b, k, j, size);
            }
        }
    }
}


int main(int argc, char** argv)
{
    int size = 2;
    int times = 100000;
    int which = 1;
        if (argc > 1)
    {
        size = atoi(argv[1]);
        if (size < 2)
        {
            size = 2;
        }
    }
        if (argc > 2)
    {
        times = atoi(argv[2]);
        if (times <= 0)
        {
            times = 100000;
        }
    }
        if (argc > 3)
    {
        which = atoi(argv[3]);
        if (which < 1 || which > 3)
        {
            which = 1;
        }
    }

    int* a = new int[size * size];
    int* b = new int[size * size];
    int* c = new int[size * size];
    bzero(a, size * size * sizeof(int));
    bzero(b, size * size * sizeof(int));
    bzero(c, size * size * sizeof(int));

    switch (which)
    {
    case 1:
        for (int i = 0; i < times; ++i)
        {
            matrix_mul1(a, b, c, size);
        }
        break;
    case 2:
        for (int i = 0; i < times; ++i)
        {
            matrix_mul2(a, b, c, size);
        }
        break;
    case 3:
        for (int i = 0; i < times; ++i)
        {
            matrix_mul3(a, b, c, size);
        }
        break;
    default:
        break;
    }

    return 0;
}