Error using Parfor "Unable to classify the variable 'Sort'"

Please help, pulling my hair out;

The following code is meant to sift through arrays Sort and Sort2 looking for intersections, and then if it finds one it reassigns that line (k) to all zeros. Sort is 16.777E6 x 24, and Sort2 is 24x24 matrix. k runs from 1 to Np = 16.777E6. This segment of the whole program takes considerable time with very, very large versions of Sort, so I am trying to use parfor to speed things up. I assumed this could be done because the results of any line Sort(k) do not affect any other line of Sort(k).

The error specifically says "Error using Solver_V (line 647)

Error: Unable to classify the variable 'Sort' in the body of the parfor-loop. For more information, see Parallel for Loops in MATLAB, "Solve Variable Classification Issues in

parfor-Loops".

The offending portion of code is below, and I included the full files for context.

Thank you so much!

parfor k=1:Np
       for i=1:(Nu-1)
            for j=(i+1):Nu
                if Sort(k,i)*Sort(k,j)==1 && Sort2(i,j)==0
                        Sort(k,:)=zeros;
                        flag=1;
                        break
                end
            end
            if flag==1
                flag=0;
                break
            end
       end
       if mod(count2,10E6)==0
           count2-count1
           count2
       end
      
       count2=count2+1;
   end

Answer 1

[Raymond Norris]

Try this refactoring the for-loop a bit

parfor k=1:Np
    flag = 0; %#ok<PFTUSW> 
    for i=1:(Nu-1)
        for j=(i+1):Nu
%             if Sort(k,i)*Sort(k,j)==1 && Sort2(i,j)==0
%                 Sort(k,:)=zeros;
%                 flag=1;
%                 break
%             end
            tmp_S = Sort(k,:);
            if lFoundIntersection(tmp_S,i,j,Sort2)
                Sort(k,:) = zeros;
                flag = 1;
            end
        end
        if flag==1
            flag=0;
            break
        end
    end
    if mod(count2,10E6)==0
        count2-count1
        count2
    end

    count2=count2+1;
end

At the end of your script, add

function tf = lFoundIntersection(Sort,i,j,Sort2)
tf = Sort(:,i)*Sort(:,j)==1 && Sort2(i,j)==0;
end

I think you'll also need to comment out this

    if mod(count2,10E6)==0
        count2-count1
        count2
    end

[Erin Pratt]

Thank you so much, this works!

[Erin Pratt]

Well, it did work. I ran it with a 13 node project, and it worked fine. When I moved to a 26 node project, I got the following error (the parfor statement went from 1 to 16E6 to now 1 to 1.07E9).

Error using distcomp.remoteparfor/rebuildParforController (line 212)

Unexpected failure to add interval 5.

Error in distcomp.remoteparfor/handleIntervalErrorResult (line 258)

obj.rebuildParforController();

Error in distcomp.remoteparfor/getCompleteIntervals (line 395)

[r, err] = obj.handleIntervalErrorResult(r);

Error in Solver_V (line 647)

parfor k=1:Np

Any idea why this would suddenly not work with a larger matrix?

Thanks again

[Raymond Norris]

I would suggest two things:

1. Make Solver_V a function and not a script. This will scope what variables are in the script's workspace.
2. Change the parfor back to a for-loop with a 26 node project and see if it works. If not, resolve the issue and then go back to a parfor for a final test.

Answer 2

[Edric Ellis]

I just came here to mention that we tried to improve the error reporting for this case. Here's what happens now, hopefully it makes more sense.

parfor k=1:Np
       for i=1:(Nu-1)
            for j=(i+1):Nu
                if Sort(k,i)*Sort(k,j)==1 && Sort2(i,j)==0
                        Sort(k,:)=zeros;
                        flag=1;
                        break
                end
            end
            if flag==1
                flag=0;
                break
            end
       end
       if mod(count2,10E6)==0
           count2-count1
           count2
       end
      
       count2=count2+1;
   end

Error: The parfor-loop variable 'Sort' has multiple sliced accesses, but they do not all have the same list of subscripts. Each access to a sliced variable must use precisely the same list of subscripts. For more information, see Parallel for Loops in MATLAB, "Solve Variable Classification Issues in parfor-Loops".