Parallel processing for asynchronous plotting

Hello,

I have a script that plots two different 3D graphs continuously as follows:

while(condition)
    
    mex_function() %continuous communication with a C++ MEX function, which provides the variables a, b, c, x, y, z, etc.
    
    subplot(1,2,1)
    function1(a,b,c) %simple plotting function
    
    subplot(1,2,2)
    function2(x,y,z) %time-consuming plotting function
    
end

Because the second subplot takes so long, and the first one needs to be updated in real time, I tried to rewrite this using parfor. These plotting functions do not interact with each other.

parfor i = 1:2
    if i == 1
        % code containing variables required by mex_function
        
        while(condition)
            mex_function()
            f1 = figure;
            function1(a,b,c)
        end
        
    else
        % code containing variables required by mex_function
        
        while(condition)
            mex_function()
            f2 = figure;
            function2(x,y,z)
        end
        
    end
end

The MEX function saves and gets the variables using matlabPtr->setVariable(u"var_name",variable) and variable = matlabPtr->getVariable(u"variable_name"), respectively

I get the error message that my MEX function can't get the variables declared in the 'base' workspace. I am not familiar with how/where parallel workers get their variables, so does anyone have any suggestions on how to solve this issue?

As an alternative, is there a way of running two scripts completely separately in parallel?

Thank you in advance.

Answer 1

[Edric Ellis]

One of the restrictions of parfor is that all variable access in the body of the loop must be "transparent". This is to ensure that the parfor machinery knows which variables to transfer from the client to the workers and back again.

If possible, I would recommend restructuring your MEX function to work with explicit input and output arguments.

A second option would be to hide your calls to your MEX function inside a MATLAB function that sets up the base workspace as required. (This is required because matlabPtr->getVariable() access the base workspace). Maybe something like this:

function mexWrapper(aVal,bVal,cVal)
% Set up the base workspace
assignin('base', 'a', aVal);
assignin('base', 'b', bVal);
assignin('base', 'c', cVal);
mex_function();
... % more stuff
end
% Then, call in parfor like this
parfor i = 1:2
    if i == 1
        mexWrapper(a,b,c);
    else
        ...
    end
end

Note however when workers manipulate figures and other graphical objects, they are not visible at the client. (You can create graphics on workers and then save/print them to file from there).

[André C. D.]

Thanks for the suggestion, I used the second option and the issues with getVariable in the MEX function seem to be fixed — now the MEX function is able to access the base variables. However, I still have an issue with setVariable, since the workers can't access the variables that are saved using the following pattern:

Array str1 = factory.createCharArray(str);

matlabPtr -> setVariable(u"a",str1);

"a" being the string that is supposedly saved to the workspace for the matlab script to use.

I also don't understand how using "assignin('base', 'a', aVal)" is relevant in this case, since the variable "a" is actually provided by the MEX function, and doesn't exist before the MEX function is called.

Thank you again!

[André C. D.]

P.S.: what exactly do you mean by working with "explicit input and output arguments"? I couldn't find anything online regarding that.

[Edric Ellis]

What I meant was that instead of using setVariable and getVariable, you should write your MEX function to accept input parameters and return output parameters using the argument list.

[André C. D.]

Thank you, it's (more or less) working now!