Running Functions in Parallel

Hi,

I'm working on a project where I need to call a function from within a script. The problem is that the function takes ~2 sec to execute. The main script is also controlling a quadcopter, and when run independently(without calling the said function), runs at ~50Hz. Is there a way in which I do not have to wait for the function output, and keep executing the main script. You can refer to the relevant code snippet below:

        % Main Script 

        %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
        %-------------------- Main Loop for Simulation -----------------------%
        %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

        %set up the domain
        x_coord=[0:.1:4.0]; %meters
        y_coord=[0:.1:4.0]; %meters
        z_trash=[0:.1:2.5]; %meters
        z_coord(1,1,:)=z_trash;
        clear z_trash;

        Q=zeros(length(y_coord),length(x_coord),length(z_coord));

        x_mat = permute(repmat(x_coord',1,length(y_coord),length(z_coord)),[2 1 3]);
        y_mat = permute(repmat(y_coord,length(x_coord),1,length(z_coord)),[2 1 3]);
        z_mat = repmat(z_coord,length(x_coord),length(y_coord),1);

        R=10;
        %Define the human head state x_h y_h z_h pitch_h roll_h yaw_h
        x_h=3; y_h=2; z_h=1.8; pitch_h=0; roll_h=0; yaw_h=pi/2;

        points=[x_mat(:),y_mat(:),z_mat(:)];

        while (true)
            tic
            phi=10;
            c1=1;
            a=-60; %deg
            b=60; %deg
            sigma=8; %deg

            c2 = 5;
            B=(1/max(0,c1))+(1/max(0,c2));
            Si=1./B;
            Si=Si./max(max(max(Si)));
            Q=Q+Si;
            output = resample([points,Q(:)]);
            save('results.mat','output','-v6');
            %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
            %--------------- problematic part - runs at ~ 0.5Hz ------------------%
            %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%        
            code = gmm();
            %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%       

            ctrl_msg(1).Translation.X = vx;
            ctrl_msg(1).Translation.Y = vy;
            ctrl_msg(1).Translation.Z = vz;
            ctrl_msg(1).Rotation.X = roll;
            ctrl_msg(1).Rotation.Y = pitch;
            ctrl_msg(1).Rotation.Z = yaw;

            send(ctrl_pub(1),ctrl_msg(1));
            drawnow
            toc
        end

The secondary script (gmm.m) calls a python function which computes EM using a CUDA implementation and takes around 2 sec on my GPU. The code for gmm.m can be found below. inp and out are .mat files which I write to communicate between python and matlab. If I use '&' at the end of commandStr in the script below, the main scripts initializes multiple instances of Python causing the CPU to get overloaded. If I don't use it, the function waits till the command is executed and returns a returncode of 1.

function returncode = gmm()
    global inp out thresh num_iter verbose K;
    if  exist(fullfile(cd, inp), 'file') == 2
        commandStr = ['python emgmm.py ' inp ' ' out ' ' int2str(num_iter) ' ' int2str(thresh) ' ' int2str(verbose) ' ' int2str(K) '&'];
        returncode = system(commandStr);     
    end
end

Is there a way to keep executing the gmm function in the background till it exits with returncode 1 (after ~2sec) and let the main loop keep on executing. Any suggestions would really help me out alot!

Answer 1

[Walter Roberson]

parfeval with a pool size of 1 (so that the gpu is not overloaded) and NOT using & on the command (so that it waits for the results)

You can queue multiple parfeval and they will execute in sequence. But of course if you issue them faster than an average of two seconds intervals then you risk developing a growing backlog.