Matlab calculations slows to a crawl when executed from within a GUIDE GUI

Question

(The following takes place on a Macbook Pro 2.3GHz i7 Haswell with 16GB RAM)

Issue

In Matlab I have a 3D-data processing algorithm. In this case a three-way nested loop where the dimensions are on the scale of <1000 values per vector (ie z<1000, y<1000, x<1000). For each voxel a dynamic set of calculations are performed. When executed from the Matlab command line the speed of the algorithm is good. It runs as can be expected on a relatively high-end machine.

When I move this very same code (copy & paste) into a GUIDE GUI callback function (a Button) generated by Matlab the rate of execution slows to an absolute crawl. It is not 1/10th or even 1/100th the normal rate. Possibly it's closing in on 1/1000th the normal rate... or something of that order.

Thoughts

So my initial thoughts are that this radical performance hit must be comming from the GUI somehow. So I tried to implement the parfor Matlab keyword in order to run the computationally intensive for loop in parallel. This was not possible since any variable created inside the parfor-loop cannot be used outside of it. Or at least that is how I interpret the Matlab error that I get when trying to use that method.

I also tried spawning the calculation in a new thread but did not seem to improve performance in any way.

Question

My question: Is there a certain way that computationally intensive tasks should be executed when running from a Matlab GUI?

Addition

Pseudo code of this situation. The following Matlab functions are used inside the loop: max(), abs(), sqrt(), zeros(), mod(), fprintf() Please note that the use of fprintf() is NOT excessive and I am aware that excessive calls to functions like this could result in a performance penalty.

for-loop
    fprintf(1,'%d ',i);
    if (mod(i,20)==0) 
         fprintf(1,'\n'); % output iteration status
    end
    for-loop
        for-loop

            some variable assignments

            declare variables and init them to 1 or 0
            for-loop Xvar
                if ((condition)) <= 0.0
                    variable assignment
                end
                if ((condition)) <= 0.0
                    variable assignment
                end
            end
            for-loop Yvar
                if ((condition)) <= 0.0
                    variable assignment
                end
                if ((condition)) <= 0.0
                    variable assignment
                end
            end
            for-loop Zvar
                if ((condition)) <= 0.0
                    variable assignment
                end
                if ((condition)) <= 0.0
                    variable assignment
                end
            end
            somevariable=zeros(args);

            for-loop
                multiplication assignment to variable
                multiplication assignment to variable

                variable assignment

                for-loop
                    for-loop
                        for-loop
                            sqrt calculation with some divisions and squares
                            if-statement
                                subtraction addition assignment
                                if-statement
                                    var=some other var;
                                end
                            end
                        end
                    end
                end
                somevarindexed(m)=stuff calculated above;
            end

            variable assignment;
            3Dvector(x,y,z) assignment;
            3Dvector(x,y,z) assignment;
        end
    end
end

I cannot duplicate the behavior. Is your calculation recursive? Does the initial code run into a script or a function? If a function, then is the callback containing the calculating code as a nested function? — user2271770
I have added pseudo code to my post. The calling order is: GUIDE GUI Handler -> Button Callback with algorithm inside. No other function calls are made (aside from the Matlab function that I added above). — Peter
Have you used the profiler to identify differences in execution between the two cases? — excaza
Also check what's happening to the memory when you run it in the GUI vs. without the GUI. MATLAB GUIs aren't exactly super-optimized, so it wouldn't surprise me if the GUI is hogging up just enough memory to bring everything to a halt when your loop runs. — btmcnellis

Peter Peter · Accepted Answer · 2015-10-15T12:42:58

Global matrix variables that change inside a three-layer loop were the reason for this issue. After re-factoring the code everything now works as expected.

Matlab calculations slows to a crawl when executed from within a GUIDE GUI

1 Answers