How to gather results from a parfor loop?

Hi all,

I'm new to parallel computing in MATLAB, please forgive my simple question.

I have a test code like this, while I'd like to extract the otpt matrix at 10000th iteration:

inpt = rand(5);

parfor i = 1:10000

      otpt = sin(inpt);

end

After running this paralleled code, no error appears, but if I call otpt:

>> otpt
Undefined function or variable 'otpt'.

What if I need the value of otpt?

I did something like this:

inpt = rand(5);

temp = zeros(5);
parfor i = 1:10000

      otpt = sin(inpt);
      if i == 10000
          temp = temp + otpt;
      end
  end

In this way I can extract the value of otpt at 10000th iteration, but intuitively I do not feel this is correct. So how can I get the value of otpt?

Thank you!

[Joss Knight]

If you need the result from a loop then either you need a result from every iteration, in which case you need to assign into an array that you've initialised before the loop; or your loop is not parallelizable because each iteration depends on the one before - in which case you can't use parfor.

Answer 1

[Brendan Hamm]

The issue here is that otpt is assigned separately on each of the workers, this makes its classification as a temporary variable. If you would like to access the values afterwards you would want to assign to a reduction or a slicing variable. What you did in the second example is perfectly acceptable, although the example is unrealistic as why use a parfor loop if you only care about the result at a specific iteration?