for-loop vectorization and structure

Hello all,

I've tried to vectorize the following for-loop:

for i=1:ne

       S(1,i).P(1,1).x    = [v_proj(fv(i,1),1) v_proj(fv(i,2),1)      
                             v_proj(fv(i,3),1)];
       S(1,i).P(1,1).y    = [v_proj(fv(i,1),2) v_proj(fv(i,2),2) 
                             v_proj(fv(i,3),2)];
       S(1,i).P(1,1).hole = 0;

end

But unfortunately I did not manage to do so. I've tried what follows:

S(1:ne).P.x    = [v_proj(fv(1:ne,1),1) v_proj(fv(1:ne,2),1) v_proj(fv(1:ne,3),1)];

Even though everything turns out to be correct for the right-hand side, it does not work for the left one as I would have expected. Can anyone help? Because I really need to feed this structure in a more efficient way and increase original poor speed of the algorithm.

Thank you very much for your help,

Regards,

% Romain

Answer 1

[Walter Roberson]

t = mat2cell( [v_proj(fv(1:ne,1),1) v_proj(fv(1:ne,2),1) v_proj(fv(1:ne,3),1)], ones(1,ne), 3);
[S(1:ne).P.x] = t{:};

This is not tested but it looks about right.

[Jan]

@Walter: Do you assume that v_proj and fv are arrays?

@Romain: I very curious about a timing comparison...

[Romain W]

@Walter: thank you for your reply, unfortunately I get the same error message as in what I've tried to do so far: "??? Scalar index required for this type of multi-level indexing."

@Jan: if the error mentioned above can be fixed,I will post a timing comparison using my data. And thank you for your reply again.

[Jan]

The [S(index).B.C] operation does not work for non-scalar "index". Matlab resolves on level of substructs only, e.g. [S(index).B].

Answer 2

[Jan]

The vectorized version will not be faster, as far as I can see.

Creating the potentially large array [v_proj(fv(1:ne,1),1) v_proj(fv(1:ne,2),1) v_proj(fv(1:ne,3),1)] will allocate a lot of memory. Spliiting it afterwards to vectors will require more memory. Therefore the overall procedure is most likely less efficient than your FOR-loop.

Please try this:

P.hole = 0;
for i = ne:-1:1   % Backwards for implicite pre-allocation
  P.x = [v_proj(fv(i,1),1), v_proj(fv(i,2),1), v_proj(fv(i,3),1)];
  P.y = [v_proj(fv(i,1),2), v_proj(fv(i,2),2), v_proj(fv(i,3),2)];
  S(i, 1).P = P; 
end

what is v_proj and fv? Arrays or functions? Would something like this work:

t   = fv(i, 1:3);
P.x = v_proj(t, 1);
P.y = v_proj(t, 2);

[Romain W]

Thank you for your reply. I am gonna try your solution. v_proj is a matrix size [ne x 2]