execution of code very long

Hi, I need help to improve the following code. The execution is too long 79.014s

function [Dmin]=dist_am(Q,R,Obs,N,d,n_obs,contact,comp)
% Dp is matrix of distances 
% attention D(c) corresponds to c ieme contact possible
% Dmin= min(D)
Dmin=0;
if comp>0   %  table  of contact is not empty   
    Dp=zeros(comp,1);    
    % on parcourt le tableau de contact
    for c=1:comp
        if(contact(c,3)==0) % contact particule-particule
            i=contact(c,1);
            j=contact(c,2);
            p=norm(Q(:,j)-Q(:,i));
            Dp(c)=p-(R(i)+R(j));
        else
            % contact obstacle-particule
            i=contact(c,1);
            l=contact(c,3);
            % zone  ou se trouve la part i par rapport a l'obst l
            if (Q(:,i)-Obs(1:2,l))'*(Obs(1:2,l)-Obs(3:4,l))>0
                % la part est dans la zone 1
                p=norm(Q(:,i)-Obs(1:2,l));
                Dp(c)=p-R(i);
            elseif (Q(:,i)-Obs(3:4,l))'*(Obs(3:4,l)-Obs(1:2,l))>0
                % la part est dans la zone 2
                p=norm(Q(:,i)-Obs(3:4,l));
                Dp(c)=p-R(i); 
            else
                pro_scal=(Q(:,i)-Obs(1:2,l))'*(N(:,l));
                if pro_scal>0
                    % la part est dans la zone 3
                    Dp(c)=pro_scal-R(i);
                else
                    % la part est dans la zone 4
                    Dp(c)=-pro_scal-R(i);
                end
            end
        end
    end
    Dmin=min(Dp);
end

Thank you.

[Friedrich]

Please format you code to make it readable. Thanks :)

Answer 1

[Daniel Shub]

Have you tried to profile your code to figure out where the bottleneck is? If comp is really big, you might be able to get a little speed up by not constructing the entire Dp matrix. You could do the comparison on the fly and only need the current Dp and Dmin.

Depending on what contact looks like, you might be recomputing components. The only dependency on c in your loop is in the construction of i and j (and i and l). If you do the calculations for individual pairs multiple times, that will slow you down.

Have you thought about a parfor or using the PCT?