How to modify code to use parfor rather than more than two nested for loops?

Hi All

I am having a problem wrapping my head around using parfor command with more than one nested for loop.

The article: http://blogs.mathworks.com/seth/2010/10/17/parallel-computing-with-simulink-running-thousands-of-simulations/ is very informative and uses meshgrid to give the correct index values. However this is only valid if replacing two for loops with parfor.

I am currently running a parameter sweep with a simulation in Simulink, which I call from matlab. There are 6 variables I am iterating, with 8 values for each. Each simulation is independent of one another. The dependency only comes in because I am using the for loops to iterate through the parameters. I have attached an example of what I mean below.

Any help with speeding up the process, or help with understanding what I need to be doing would be greatly appreciated.

Thanks

   % code
for c1=1:itrsize
    for c2=1:itrsize

        [Css]=DSpec2(c1,c2,MRr);
        Csrs=[Css(:,1),Css(:,2)];
        for c3=1:itrsize
            for c4=1:itrsize

                [Css]=DSpec2(c3,c4,MRr);
                Csfs=[Css(:,1),Css(:,2)];
                for c5=1:2

                    f=Springs(c5);
                    for c6=1:2

                        r=Springs(c6);
                        sim('Example') %Run Simulation
                    end

                end
            end
        end
    end
end

[K E]

If I could give this question 5 votes I would! This is the main barrier to using parfor on legacy code which might have 5-6 nests of parameters.

Answer 1

[Edric Ellis]

I think you should be able to solve this problem by using linear indexing - i.e. taking advantage of MATLAB's ability to index arrays using fewer than the 'natural' number of subscripts. For example,

    inputData = rand(4, 5, 6, 7); % 4-D input data
    outputData = zeros(size(inputData)); % output same size and shape

    % use a single loop over all elements of 'inputData'
    parfor ii = 1:numel( inputData )
        outputData(ii) = myCalculation( inputData(ii) );
    end

Note that by pre-allocating 'outputData', we ensure it ends up the correct shape.

[K E]

In case it is helpful: you can generate all combinations of the parameter sets using allcomb on the File Exchange.

Answer 2

[Jonathan]

Hi guys

Thanks for the comments.

Edric, I dont really understand what you were preposing. It does sound abit like the method proposed by K E.

K E what you're proposing makes a lot of sense. In a way, I just used my existing for loops to iterate the variables and then save them in a much larger matrix. I then saved this as just call it prior to running the program. Thanks for the file.

I can now use the parfor loop for the time consuming simulation.