For loop faster than vectorized?!

Hi,

Can anybody tell me why for loop is faster than vectorized form in this code? What is the fastest alternative to these?

Thanks

clear;clc;
s = rand(2,9);
DIM = 2;
iter = 100;
TIME = zeros(2,iter);
for i = iter;
    x = rand(12,1);
    for n =1:9
        for m = 1:3
            INDEX = 2*(m-1)*DIM;
            
            %For loop
            tic
            for j = 1:DIM
                TEMP1 = (x(INDEX+j)-s(j,n));
            end
            toc
            
            %Vectorized
            tic
            TEMP2 = (x(INDEX+1:INDEX+DIM)-s(:,n));
            toc
        end
    end
    
end

[KALYAN ACHARJYA]

Read here

Answer 1

[Arjun]

Hi Arash,

I see that you are trying to compare the performance in terms of elapsed time between “for” loop and vectorized version of the same code and found out that “for” loop version is faster than the vectorized version.

The “for” loop which is used for comparison is running only for 2 iteration and this is not a very accurate performance comparison as in this case the overhead of logical addressing in vectorized approach may be far greater to see any real performance gain. The effect of vectorization is more pronounced when there are many parallel computations to be performed. Apart from this, since the loop is very small, optimizations like loop unrolling makes it very fast at run time.

For the case above, using “for” loop seems to be the right choice.

Kindly refer to this MATLAB Answer as well: https://www.mathworks.com/matlabcentral/answers/421617-why-is-vectorizing-way-slower-than-the-for-next-loop-it-replaced

I hope this will help!.