Making a mex function for a rather complex code

Hello friends,

I have a code which is very time consumming. Therefore, I thought to use mex function to do the heavy part of calculations using C or C++. I have difficulty to define input types properly as my code has some inputs which are rather complex, for instance one input is a function handle and the other one is a variable sized cell whose elements are function handles. The rest of inputs are easy.

To make things easier I thought to ask my question via a very simple code (attached as main.m) rather than mine (this code does nothing for me but if I understand how to tackle this code I am able to tackle my own code). In this code there is a subrutine called 'Cost' which is the 'heay part'. I would like to make a mex function for this subrutine to speed up the calculations. Please note that this code is not my real code and I just made it for the purpose of learning how to make a mex file out of this code (I know that this code does not solve any scientific problem).

I impatiently look forward to hearing from you soon,

Thanks in advance!

Babak

[Mohammad Shojaei Arani]

Friends,

Did I ask a stupid question?

I can imagine that my question might have felt wierd to you as I am quite naive to mex technology!

I was hoping that one of you can help me or give me any (=ANY) clue on this, which is of course greatly

appreciated!

[Jan]

As far as I understand, you post a simplified code which differs from the real code. This is a usual cause of not helpful answers, because the actual problm can be hidden in the concealed code.

Symbolic variables and function handles do not profit frem mexing, because the mex function have to call Matlab to evaluate the expressions.

Use the profiler to find the bottleneck of the real code. Post the results together with the real code. Then an efficient help is possible and the time spent fpr typing the answer will less likely be lost.

[Mohammad Shojaei Arani]

Hello Jan,

Thanks for your kind response and attention!

Well, if I wanted to post my actual problem that would be really big with lots of heavy math. So, I thought to make a very tiny and short problem which is not boring for you. I do not have any symbolic calculations inside my procedure but I do have function handles. You do not need to know the philosophy behind my code, my question should be extremely simple for highly experienced people that you really are. So, perhaps I can explain a bit further. In bellow, you see my code. My question is only related to the part being highlighted in bold (i.e., the function called Cost). I just want this part to be mexed and be done using C. To be more clear, I just need to know how to define type inputs in the matlab coder. For instance the input H is an anonymos function and I do not know how to define its type in matlab coder box. I If you teach me how to do this then I am done!

Thanks Jan, in advance!

clc;
data=rand(1,10);
J=5;
syms x 
H=sym('H',[J 1]);
for j=1:J
    H(j)=horner(simplify(exp(x^2/2)*diff(exp(-x^2/2),j)),x);
end
H=matlabFunction(H);

syms w
Coeff=cell(1,J);
for j=1:J
    A=zeros(1,j+1);
    for k=1:j+1
        A(k)=(-1)^(k-1)*nchoosek(j,k-1);                                
    end
    C=A.* w.^(0:j);C=C(2:end);
    Coeff{j}=matlabFunction(C);
end

cost=@(par)Cost(par,data,J,Coeff,H);
par=rand(1,J);
f=Cost(par,data,J,Coeff,H);

function cost=Cost(par,data,J,Coeff,H)
A=par*H(data);
B=0;
for j=1:J
    C=Coeff{j}(par.');
    B=B+norm(C);
end
cost=norm(A)+B;
end