sqrm

static void sqrm(double *src, double *buffer, int size)
{
    memset(buffer, 0, size * size * sizeof(double));

#if defined(SEPARATED)
    #pragma omp parallel for
    for (int i = 0; i < size; i++)
    for (int k = 0; k < size; k++)
    for (int j = 0; j < size; j++)
        buffer[i * size + j] += src[i * size + k] * src[k * size + j];

    #pragma omp parallel for
    for (int i = 0; i < size; i++)
    for (int j = 0; j < size; j++)
        src[j * size + i] = buffer[i * size + j];
#elif defined(BLOCK_ROW)
    #pragma omp parallel for
    for (int ib = 0; ib < size; ib += BLOCK_SIZE)
    {
        int ie = ib + BLOCK_SIZE;
        if (size < ie)
            ie = size;

        for (int jb = 0; jb < size; jb += BLOCK_SIZE)
        {
            int je = jb + BLOCK_SIZE;
            if (size < je)
                je = size;

            for (int ii = ib; ii < ie;   ii++)
            for (int kk = jb; kk < je;   kk++)
            for (int jj =  0; jj < size; jj++)
                buffer[ii * size + jj] += src[ii * size + kk] * src[kk * size + jj];
        }
    }

    #pragma omp parallel for
    for (int i = 0; i < size; i++)
    for (int j = 0; j < size; j++)
        src[j * size + i] = buffer[i * size + j];
#elif defined(BLOCK_BLOCK)
    #pragma omp parallel for
    for (int ib = 0; ib < size; ib += BLOCK_SIZE)
    {
        int ie = ib + BLOCK_SIZE;
        if (size < ie)
            ie = size;

        for (int kb = 0; kb < size; kb += BLOCK_SIZE)
        {
            int ke = kb + BLOCK_SIZE;
            if (size < ke)
                ke = size;

            for (int jb = 0; jb < size; jb += BLOCK_SIZE)
            {
                int je = jb + BLOCK_SIZE;
                if (size < je)
                    je = size;

                for (int ii = ib; ii < ie; ii++)
                for (int kk = kb; kk < ke; kk++)
                for (int jj = jb; je < je; jj++)
                    buffer[ii * size + jj] += src[ii * size + kk] * src[kk * size + jj];
            }
        }
    }

    #pragma omp parallel for
    for (int i = 0; i < size; i++)
    for (int j = 0; j < size; j++)
        src[j * size + i] = buffer[i * size + j];
#endif
}