Parallel Computing (Job and task)

hi i try to run matlab code that runs parallel . my problem is .i have 8 for loops and i want these loops to run at the same time because they are independent .any way to do that.my code is

a=imread('x.jpg');
[xx,yy]=size(a);
  while x-1>0 && y-1>0
      r=mean-a(x-1,y-1);
      x=x-1;
      y=y-1;
  end
  %north
  while x-1>0 && y<yy
      r=mean-a(x-1,y+1);
      x=x-1;
  end
  %north west
  while x-1>0 && y+1<yy
      r=mean-a(x-1,y+2);
      x=x-1;
      y=y+1;
  end
  %west
  while x>0 && y+3<yy
      r=mean-a(x+1,y+3);
      y=y+1;
  end
  %north west
  while x+3<xx && y+3<yy
      r=mean-a(x+3,y+3);
      x=x+1;
      y=y+1;
  end
  %norht
  while x-1>0 && y-1>0
      r=mean-a(x+3,y+1);
      x=x-1;
      y=y-1;
  end
  %north east
  while x+3<xx && y-1>0
      r=mean-a(x+3,y-1);
      x=x+1;
      y=y-1;
  end
  %eset
    while x>0 && y-1>0
        r=mean-a(x+1,y-1);
          y=y-1;
    end

Answer 1

[Walter Roberson]

Those loops are not independent: the initial x and y values for any one of the loops depends upon the outcome of the previous loops.

If you want to execute different code per worker, then my understanding is that you would use smpd and use the labindex to switch() to the work to be done. (Or use an array of function handles and index in that by the labindex, I suppose.)

Looking at your code structure, it seems to me likely that the amount of work done would be unbalanced between the workers, with some doing little or no work. There is a high overhead cost to starting a worker, so it is best if each worker does sufficient work that the startup overhead is relatively negligible.

I cannot tell at the moment what your code is intended to do. Please do not use "mean" as a variable name, as it conflicts with the name of the important MATLAB routine mean()

[Isee You]

thank u.

but please can you tell me how to use (Job and task)and if it can help me. please can you tell me how?

(smpd)what do you mean by it?

[Walter Roberson]

Sorry, 'smpd' should have been written 'spmd'.

http://www.mathworks.com/help/toolbox/distcomp/spmd.html

There might be a way to do that work using the Job and Task structure http://www.mathworks.com/help/toolbox/distcomp/bqur7ev-2.html but I don't think that would be the usual way to approach matters.

[Isee You]

"array of function handles and index in that by the labindex""

i do not familiar with that. any help will be useful !!!

[Walter Roberson]

See Example 3

http://www.mathworks.com/help/techdoc/ref/function_handle.html

[Isee You]

can you see my code and tell me your opinion?

Answer 2

[Isee You]

i try this code , do you think it true??

function do_test(avg,x,y)
         matlabpool open
    a=imread('x.jpg');
    % [xx,yy]=size(a);

      c = {@northest, @north, @northwest,@west, @southwest, @south,@southeset,@eset};
      parfor f=1:8
         c{f}(avg,a,x,y) ;

      end

      function  northest(avg,a,x,y) 
      %north eset
      while x-1>0 && y-1>0
          r=avg-a(x-1,y-1);
          x=x-1;
          y=y-1;
      end

      function north(avg,a,x,y) 
      [xx,yy]=size(a);
      %north
      while x-1>0 && y<yy
          r=avg-a(x-1,y+1);
          x=x-1;
      end

      function northwest(avg,a,x,y) 
      [xx,yy]=size(a);
      %north west
      while x-1>0 && y+1<yy
          r=avg-a(x-1,y+2);
          x=x-1;
          y=y+1;
      end

      function west(avg,a,x,y) 
      [xx,yy]=size(a);

      %west
      while x>0 && y+3<yy
          r=avg-a(x+1,y+3);
          y=y+1;
      end

      function southwest(avg,a,x,y) 
      [xx,yy]=size(a);

      %south west
      while x+3<xx && y+3<yy
          r=avg-a(x+3,y+3);
          x=x+1;
          y=y+1;
      end

      function south(avg,a,x,y) 
      [xx,yy]=size(a);

      % south
      while x-1>0 && y-1>0
          r=avg-a(x+3,y+1);
          x=x-1;
          y=y-1;
      end

      function southeset(avg,a,x,y) 
      [xx,yy]=size(a);
      %south east
      while x+3<xx && y-1>0
          r=avg-a(x+3,y-1);
          x=x+1;
          y=y-1;
      end

      function eset(avg,a,x,y) 
      [xx,yy]=size(a);
      %eset
      while x>0 && y-1>0
          r=avg-a(x+1,y-1);
            y=y-1;
      end

[Walter Roberson]

Well, that code will do something in parallel.

It won't do anything _useful_ as you do not return or store any values in the functions.

It will also not do the equivalent of the serialized code, as in the serialized code, the input of each stage is the output of the previous stage.

But in terms of setting up parallelism... yes, those routines should get executed in parallel.

[Isee You]

thank u "Walter Roberson".

my code is a step in many steps .