Several nested for-loops within parfor-loop variable indexing

I am trying to run a Simulink model for a sensitivity analysis. Since I need to run the system with 5 different variables all with 6 different values, there are 7776 total trials I need to run. So I'm trying to use a parfor-loop to speed up the process.

I have a set of 4 nested for-loops within a parfor-loop. The problem is that MATLAB says, "The PARFOR loop cannot run due to the way variable 'savedVar' is used." How do I go about redefining or reindexing the variables such that parfor can run?

The loop structure looks something like this:

jn = 1:length(var2);
kn = 1:length(var3);
ln = 1:length(var4);
nn = 1:length(var5); % This is to avoid calling a function within the for-loop index variables
parfor i = 1:length(var1)
    savedVar_temp_var2 = zeros(length(var2),length(var3),length(var4),length(var5))
    savedVar_temp_var3 = zeros(length(var3),length(var4),length(var5))
    savedVar_temp_var4 = zeros(length(var4),length(var5))
    savedVar_temp_var5 = zeros(length(var5)) % Preallocating for my sliced variables
    for j = 1:jn
        for k = 1:kn
            for l = 1:ln
                for n = 1:nn
                    parsim('mysim.slx')
                    savedVar_temp_var5(n) = Vars; % Vars is the stand-in here for the 
                    % various variables I want to save during the simulation
                end
                savedVar_temp_var4(l,:) = savedVar_temp_var5; % Assigning the variables
                % entire rows/columns at a time (recommended in "Sliced
                % Variables" documentation
            end
            savedVar_temp_var3(k,:,:) = savedVar_temp_var4;
        end
        savedVar_temp_var2(j,:,:,:) = savedVar_temp_var3;
    end
    savedVar(i,:,:,:,:) = savedVar_temp_var2; % Allocating all temp variables to actual array

    saveVariablesFunc(savedVar) % MATLAB's "save" function does not work in parfor -- adding
    % a function to do it works 
end


% locally defined saving function
function saveVariablesFunc(savedVar)

    save('myfile.mat','savedVar','-mat')

end

[Paul]

Just want to point out that that use of parsim is undocumented.

I suspect there's probably a better way to accomplish the goal than using several loops. I suggest starting from this doc page for options and methods for running multiple simulations, if not read already.

Answer 1

[Matt J]

If you just remove the line,

 saveVariablesFunc(savedVar) 

form the loop it will work, after you also fix the defintion of jn...nn (they're supposed to be scalars).

I don't know what you're trying to accomplish with saveVariablesFunc. In a parallel loop, the order of the iterations is not predictable, so saving intermediate iterations to a file would not allow you to resume the loop from any well-defined starting point later on if your computer were to crash or something like that.

[Matt J]

This might work, though

parfor i = 1:length(var1)

...
   
    savedVar(i,:,:,:,:) = savedVar_temp_var2; % Allocating all temp variables to actual array

    saveVariablesFunc(savedVar(i,:,:,:,:),i) % MATLAB's "save" function does not work in parfor -- adding
    % a function to do it works 
end



% locally defined saving function
function saveVariablesFunc(ithSlice,i)

    save("myfile"+i,'ithSlice','i')

end