cell_window.num_pis() vs. foreach_cell_fanin about mockturtle HOT 3 CLOSED

Hello,

Thanks for the question. The cell_window view is doing something different than you expect. Starting from some pivot node, which is a cell root, it creates a window around that node (bounded in size by some parameters) such that the window is covered itself by cells. So it does not cut through any cells. The motivation behind this is that you could throw away the current mapping of the window, remap it, and then insert it back into the original network without making the overall mapping inconsistent.

If you like to have a network structure around the cell then you can use cut_view instead.

I hope that this clarifies the problem.

from mockturtle.

Thanks for the quick response and the helpful comment! The cut_view does exactly what I want. Here is an updated version of the code above, just in case someone has the same question and comes across this issue:

template<class Ntk>
void print_cell_io( Ntk const& ntk ) // ntk is a mapped network
{
  ntk.foreach_node( [&]( auto const n ) {
    if ( ntk.is_constant( n ) || ntk.is_pi( n ) || !ntk.is_cell_root( n ) )
      return;

    std::vector<typename Ntk::node> leaves;
    ntk.foreach_cell_fanin( n, [&]( auto const f ) {
      leaves.push_back( f );
    } );

    std::cout << "cell has " << leaves.size() << " fanins\n";

    auto cut = mockturtle::cut_view( ntk, leaves, n );
    std::cout << "cut_view i/o = " << cut.num_pis() << "/" << cut.num_pos() << "\n";
  } );
}

Best regards!

from mockturtle.

Thanks for providing the updated example!

from mockturtle.