Can parfor be used in my case?

Hi, I'm a new to parallel computing and was wondering if there is any way around to use parfor in my nested for-loop

for l=1:size(xSample,2) 
    for m=1:size(ySample,2) 
        for n=1:size(zSample,2) 
            shiftedDoseDist=imtranslate(doseDist,[xSample(l),ySample(m),zSample(n)]); %%3D image translation
            dose=shiftedDoseDist(index);
            TCP_n(n)=function(dose) %%function I'm calculating
        end
        TCP_m(m)=1/sum(zWeighting)*dot(zWeighting,TCP_n);
    end
    TCP_l(l)=1/sum(yWeighting)*dot(yWeighting,TCP_m);
end

I'm evaluating a function for each x,y,z translation of a 3D matrix and then doing weighted sum for normalisation (expectation value). I tried just replacing for with parfor for the outerloop and it comes up with an error that the way TCP_n,TCP_l...are defined it is not possible.

Any help would be appreciated!

Answer 1

[Edric Ellis]

The problem here turns out to be quite simple to work around. You're assigning to elements of TCP_m in the loop over m, but unfortunately the parfor machinery cannot tell that you're completely overwriting TCP_m to make it into a parfor temporary variable.

In other words, parfor cannot tell that you aren't doing something using TCP_m that depends on the order of iterations of the parfor loop. The fix is simply to completely assign to TCP_m at the start of each iteration of the parfor loop:

parfor l=1:size(xSample,2) 
    TCP_m = zeros(1, size(ySample,2)); % forces TCP_m to be "temporary"
    for m=1:size(ySample,2)
        TCP_n = zeros(1, size(zSample,2));
        for n ...
        end
        ...
        TCP_m(m)=1/sum(zWeighting)*dot(zWeighting,TCP_n);
    end
    TCP_l(l)=1/sum(yWeighting)*dot(yWeighting,TCP_m);
end

( EDIT just realised that the same applies to TCP_n )