Parallel version of upwind scheme

Dear all, I came up to the naive question which is contradictive, i.e.

How to parallelize upwind scheme?

Suppose, we have the following implementation of upwind

for k = 1:kFin % time-loop
    % space-loop
    un(2:nx) = u(2:nx) - cfl*(u(2:nx) - u(1:nx-1)); 
    % BC
    un(1) = un(nx);
    % update
    u = un; 
end % for

So, how to parallelize it via parfor or spmd or else?

Is it possible for such schemes?

[Walter Roberson]

Is cfl a scalar, or is it a row vector with (nx-1) columns, or is it a square 2D array with (nx-1) rows and columns ?

[Oleg Kravchenko]

Dear Walter,

Thank you for reply.

I thought that number is a scalar, but it could be a vector of matrix if it's necessary for sure. Suppose is a constant value for simplicity.