valid slice are restricted in PARFOR loop

Dear All: For the sake of reducing the computation time, I decided to utilize the parallel computing toolbox, the framework of my code looks like following:

   numPoints = 68;
   correlations = zeros(1, numPoints);
   parfor j=1:numPoints
       pause(0);
        if((numPoints == 68 || numPoints == 29 )&& centres(2) == 0)

            if(numPoints == 68)
                mirrorInds =[1,17;2,16;3,15;4,14;5,13;6,12;7,11;8,10;18,27;19,26;20,25;21,24;22,23;...
                      32,36;33,35;37,46;38,45;39,44;40,43;41,48;42,47;49,55;50,54;51,53;60,56;59,57;...
                      61,65;62,64;68,66];     
            else
                mirrorInds = [1,2; 3,4; 5,7; 6,8; 9,10; 11,12; 13,15; 14,16; 17,18; 19,20; 23,24];                
            end
            mirror_idx = j;
            if(any(mirrorInds(:,1)==j))
                mirror_idx = mirrorInds(mirrorInds(:,1)==j,2);
            elseif(any(mirrorInds(:,2)==j))
                mirror_idx = mirrorInds(mirrorInds(:,2)==j,1);
            end

            if(mirror_idx~=j & correlations(1,mirror_idx) ~= 0) (it says the correlations are restricted in the parfor loop)
            ...
            correlations(1,j) = ...;
            ...

According to the sliced variable in parallel computing toolbox, there are 4 conditions on sliced variable

1. Type of First-Level Indexing
2. Fixed Index Listing
3. Form of Indexing
4. shape of array

I believe the correlations variable may already satisfy the above requirement. However, I could not understand why the matlab compiler shows that kind of information?

Could anybody give me some suggestions?

Thanks

[Adam]

mirror_idx is being calculated within the parfor loop and then used as an index into the correlations matrix. As far as I remember and understand, this is not possible in a parfor loop because it doesn't know in advance what parts of the matrix are needed by each of the workers.

[Allen]

Thanks for your answer, I changed my code into the following

 for i = 1:numPoints
         if(any(mirrorInds(:,1)==i))
                mirror_idxt(i) = mirrorInds(mirrorInds(:,1)==i,2);
           elseif(any(mirrorInds(:,2)==i))
                mirror_idxt(i) = mirrorInds(mirrorInds(:,2)==i,1);
          end
   end

parfor j=1:numPoints
      pause(0);
      % can only do mirroring if there is no yaw
      if((numPoints == 68 || numPoints == 29 )&& centres(2) == 0)
          % Do not redo a mirror feature (just flip them)

            mirror_idx = mirror_idxt(j);

            if(mirror_idx~=j & correlations(1,mirror_idx) ~= 0)
            .
            .
            .

It still shows the same message, I pre-compute the mirror_idx so it should be sliced. Could you give me some suggestions?

Answer 1

[Edric Ellis]

Adam's comment is correct. The parfor machinery cannot prove that your loop iterations are order-independent because you're indexing correlations in two different ways - you're reading from correlations(1,mirror_idx) as well as writing to correlations(1,j).

So, in this case you do not have a "fixed index listing" - you've got two different index listings. Also, correlations(1, mirror_idx) alone could never be "sliced" since you're not using the loop variable as one of the indices.

[Allen]

Thanks for your answer, I changed my code into the following

 for i = 1:numPoints
         if(any(mirrorInds(:,1)==i))
                mirror_idxt(i) = mirrorInds(mirrorInds(:,1)==i,2);
           elseif(any(mirrorInds(:,2)==i))
                mirror_idxt(i) = mirrorInds(mirrorInds(:,2)==i,1);
          end
   end
parfor j=1:numPoints
      pause(0);
      % can only do mirroring if there is no yaw
      if((numPoints == 68 || numPoints == 29 )&& centres(2) == 0)
          % Do not redo a mirror feature (just flip them)
            mirror_idx = mirror_idxt(j);
            if(mirror_idx~=j & correlations(1,mirror_idx) ~= 0)
            .
            .
            .
It still shows the same message, I pre-compute the mirror_idx so it should be sliced. Could you give me some suggestions?