Yes, but my code was just a contrieved example of the problem, not what I am actually doing.
I have a matrix of size (n_dim, n_points). I then want to pass points to different distance functions. I therefore pass point_matrix[:, current_point] to those distance functions. What you are saying is that I could manually inline these distance functions and do the explicit unrolling in each of them. That might gain me the speed but I lose the modularity and code legibility. As previously suggested, maybe having immutable types of points with a vector of coordinates in each of them is the way to go. On Monday, February 2, 2015 at 11:11:17 AM UTC+1, Yuuki Soho wrote: > > If you access directly the elements of your arrays it will not allocate > memory: > > cum += points[1, j] > > It's often worth adding a loop or reshaping your inputs so you can write > your inner code like that > and avoid memory allocation. >
