Efficient Cartesian Product algorithm

Question

Can somebody please demonstrate for me a more efficient Cartesian product algorithm than the one I am using currently (assuming there is one). I\'ve looked around SO and go

Answer 1

If cache locality (or local memory required to maintain the j's) is a problem, you can make your algorithm more cache-friendly by bisecting the input arrays recursively. Something like:

cartprod(is,istart,ilen, js,jstart,jlen) {
  if(ilen <= IMIN && jlen <= JMIN) { // base case
    for(int i in is) {
      for(int j in js) {
        // pair i and j
      }
    }
    return;
  }
  if(ilen > IMIN && jlen > JMIN) { // divide in 4
    ilen2= ilen>>1;
    jlen2= jlen>>1;
    cartprod(is,istart,ilen2,            js,jstart,jlen2);
    cartprod(is,istart+ilen2,ilen-ilen2, js,jstart,jlen2);
    cartprod(is,istart+ilen2,ilen-ilen2, js,jstart+jlen2,jlen-jlen2);
    cartprod(is,istart,ilen2,            js,jstart+jlen2,jlen-jlen2);
    return;
  }
  // handle other cases...
}

Note that this access pattern will automatically take fairly good advantage of all levels of automatic cache; this kind of technique is called cache-oblivious algorithm design.