speed up 3 nested for loops

Hi,

I have a matrix, called old_matrix, that is of size width = 500, length = 400, height = 50 and at each point in old_matrix (width,length,height) there is an intensity from 0-255.

I have to remap each point in the old matrix to a new matrix of size new_width = 500, new_length = 400, new_height = 400 * cosd(50) +10.

the function for mapping specific point of old_matrix(width,length,height) => new_matrix(width, length*cosd(height), length*sind(height))

This is the code I have so far

new_z = round((400*cosd(50)+10));
new_matrix = zeros(500,400,new_z);

for height = 50:-1:1
  for width = 1:500
      for length = 1:400

            y_new = round(length*cosd(height));
            if (y_new == 0)
                y_new = 1;
            end

            z_new = round(length*sind(height));

            if (z_new == 0)
                z_new = 1;
            end

            current_val = old_matrix(width,length,height);
            new_matrix(width,y_new,z_new) = current_val;  

        end
    end
end

It works in terms of mapping the values however, it is very slow. Is it possible to vectorize this code to improve the speed for mapping? The intensities do not change, only the position of the intensity changes.

Thank you

Answer 1

[Roger Stafford]

I think you have the nested for-loops in the wrong order for maximum efficiency. You are having to repeat the same computation on 'y_new' and 'z_new' 500 times for each combination of 'height' and 'length' indices. Do it this way instead:

 for height = 50:-1:1
   for length = 1:400
     %Compute y_new & z_new here only once per 'height', 'length' comb.
     new_matrix(:,y_new,z_new) = old_matrix(:,length,height);
   end
 end

Note: The word 'length' is a reserved word in matlab and shouldn't be used for a variable name. It is the name of one of its functions.

[varun]

Thank you for those suggestion. It did improve the speed quite a bit and have replaced length with another variable.