partition a 2D array without looping c++

Question

Im fairly new to c++ and am trying to program strassen's algorithm to multiply matrices. Part of the algorithm requires me to partition a matrix into four parts e.g

partitioned:

(each part is then used again recursively and partitioned). I want to partition the matrices without looping and copying the data from the original matrix (as this would take more time). The book i am reading says the matrices are partitioned using 'index calculations, identifying a submatrix by a range of row indices and a range of column indices of the original matrix.i am not sure what is meant by this.

Also, im not sure whether i should be using 2D arrays or vectors? Ive seen alot of people recommending vectors but ive already written everything so far in 2D arrays so im hoping what i want is possible with 2D arrays.

p.s it can be assumed the dimensions of the matrices will always be a power of 2 and be nxn (square). Also, i have seen alot of questions similar to this but none of them actually have the solution i am looking for.

Thanks

Looping is a fundamental algorithm. Trying to code anything of moderate complexity without a loop is somewhere on the order of trying to code something without using a semicolon. The only alternative to looping is to implement this process for a matrix of a known fixed size by writing a ton of repetitive spaghetti code that manually moves each value from the source matrix to the appropriate destination matrix's spot. If your book is telling you not to use looping, find another book to learn from. — Sam Varshavchik, Nov 15 '15 at 18:18

score 1 · Answer 1 · answered Nov 15 '15 at 18:21

You can create a matrix class that supports directly sub-matrices as views:

template<typename T>
struct Matrix {
    int rows, cols, stride;
    std::vector<T> data; // Possibly empty for a view
    T *ptr;

    // A fresh matrix (owning its data)
    Matrix(int rows, int cols)
        : rows(rows), cols(cols), stride(cols),
          data(rows*cols),
          ptr(&data[0])
    {
    }

    // A view of a sub-matrix (pointing to the original data!)
    Matrix(Matrix& m, int row0, int col0, int rows, int cols)
        : rows(rows), cols(cols), stride(m.stride),
          ptr[&m(row0, col0)]
    {
    }

    T& operator()(int row, int col) {
        return ptr[row*stride + col];
    }

    ...
};

Of course you need to ensure that views don't outlive the owning matrix and you need to pay attention to what you want to mean to copy a view object if that operation is not forbidden.

Adding explicit operations like transforming a view into an owning matrix can be useful (if a copy constructor is not going to do that).

partition a 2D array without looping c++

1 Answers1