MATLAB: Sliced variables in parfor loop (restricted indexing)

form of indexingParallel Computing Toolboxparforsliced variables

Hi,
I am having trouble with some parallel for-loop coding. I believe my essential difficulties can be captured in the following example code.
The function "do_something(x)" takes an array as input and returns an array of equal size ("rand(size(pSlice))" would suffice for example), but this should be irrelevant. Matlab doesn't like the way I am indexing "p". After reading "documentation-center/classification-of-variables/sliced-variables", I understand (in principle) why Matlab doesnt accept "p" (incorrect form of indexing ), but what are my options to resolve this problem?
CellOccupations=round(rand(50,1)*20);
p=rand([sum(CellOccupations) 3]);
csO=cumsum(CellOccupations);
parfor iter=1:50
    StartingPointIndex=csO(iter)-CellOccupations(iter)+1;
    EndPointIndex=csO(iter);
    pSlice=p(StartingPointIndex:EndPointIndex,:);
    pSlice=do_something(pSlice);
    p(StartingPointIndex:EndPointIndex,:)=pSlice;
end
I'll be happy to provide additional details and description of the problem if someone thinks they can help.
Thanks very much in advance for any help and effort!

Best Answer

Although you are dividing up 'p' so that each loop iteration works on a different piece, unfortunately you aren't doing it in the simple manner required for a PARFOR sliced variable. I think in this case because each iteration works on a differently sized 'slice' of 'p', you need to divide up 'p' to be a cell array. Something like this might work:
    CellOccupations = round(rand(50, 1) * 20);
    p = rand([sum(CellOccupations), 3]);
    pSliced = mat2cell(p, CellOccupations);
    parfor idx = 1:numel(pSliced)
        pSlice = pSliced{idx};
        pSlice = do_something(pSlice);
        pSliced{idx} = pSlice;
    end
    p = cell2mat(pSliced)
Related Question