Parallel for loop: Using broadcasting array variable values as indexes

Good day.

I have recently aquired the parallel computing toolbox to speed up some of our applications. I have run into the following challenge and hope that someone can give me advice on this.

topNodes    = [2 3 4 5 6 7];
bottomNodes = [1 2 3 4 5 6];
polarity    = [1 1 -1 1 1 1];
nodes       = max([topNodes bottomNodes]);
T           = zeros(numel(polarity),nodes);

for i = 1:numel(polarity);
  if polarity(i) < 0
      T(i,bottomNodes(i)) = -1;
      T(i,topNodes(i)) = 1;
  else
      T(i,bottomNodes(i)) = 1;
      T(i,topNodes(i)) = -1;
  end
end

Now, if I use a parfor loop, it complains. Some assistance will be much appreciated.

Kind regards,

Barend.

Answer 1

[Sarah Wait Zaranek]

For parfor loops, when you index into a sliced variables, restrictions are placed on the first-level variable indices. This allows parfor to easily distribute the right part of the variable to the right workers. One of these first-level indices must be the loop counter variable or the counter variable plus or minus a constant. Every other first-level index must be a constant, a non-loop counter variable, a colon, or an end.

I explain this more in the blog post where you also left this question.

http://blogs.mathworks.com/loren/2009/10/02/using-parfor-loops-getting-up-and-running/

See a possible fix below.

topNodes    = [2 3 4 5 6 7];
bottomNodes = [1 2 3 4 5 6];
polarity    = [1 1 -1 1 1 1];
nodes       = max([topNodes bottomNodes]);
T           = zeros(numel(polarity),nodes);

parfor i = 1:numel(polarity);
  myData = T(i,:);  
  if polarity(i) < 0
      myData(i,bottomNodes(i)) = -1;
      myData(i,topNodes(i)) = 1;
  else
      myData(i,bottomNodes(i)) = 1;
      myData(i,topNodes(i)) = -1;
  end
  T(i,:) = myData;
end

[Barend]

Thank you Sarah, this works fine. I notice that this script runs slower with parallel processing than with the normal vectorized approach. I suspect that it is due to the fact that the operations done in the loop is very simple?

Thank you for your help - much appreciated.

[Sarah Wait Zaranek]

Yes, there is an overhead to pass the data to and from the workers - so for code that takes seconds to run - there isn't often a speedup. Also, you are correct that vectorization often is a better choice then de-vectorizing and using parfor.

[Konrad Malkowski]

You should also try running the script as a function. In general MATLAB performance is better with functions that with scripts, as scripts are interpreted one line at a time.