How to efficiently generate a new array by indexing an array with another array

Is there a faster way to implement this loop, maybe vectorizing it?

for j = 1:m
    C(:,j) = B(A(:,j,k),j);
end

where:

A is a l x m x n 3-D array of int from 1 to 4

B is a p x m 2-D array of doubles

C is a l x m 2-D array of doubles

l = order of 1000 ; m = order of 10 ; n = order of 100 ; p = 4

[Jan]

What is "k"?

[Paolo Binetti]

Sorry, I forgot: k is an index from 2 to n Indeed, the piece of code above is inside a "k" for-loop

Answer 1

[Walter Roberson]

L = l;     %make it easier to distinguish lower-case L
C = B(sub2ind(size(B), A(:,:,k), repmat( 1:m, L, 1) ));

[Paolo Binetti]

Thank you: your solution is about 5-10 faster, which is good enough for me, as long as I compute

repmat( 1:m, L, 1)

only once and not inside the k-for-loop that contains this piece of code.

[Walter Roberson]

Yes, that is a good idea, to move that computation outside the loop.

If you have R2016b or later it can be improved even more:

C = B( (A(:,:,k) - 1) * p + (1:m) )

and the 1:m could be assigned to a variable before the loop