Understanding parfor restrictions with indexing

I was attempting to parallelize a for loop in MATLAB with parfor, and came across a strange restriction with regards to indexing. Specifically, the fact that you seemingly cannot use more than one different index into a variable. For example, something like

parfor ii=1:10
    A(ii, 1) = 1;
    A(ii, 2) = 2;
end

does not work, whereas the code

parfor ii=1:10
    A(ii, :) = [1,2]
end

does. Why is this the case? I can make my code accomodate this restriction, but I can't understand why it's necessary. The two different methods seem equivalent to me with respect to being able to cause unintended side effects in parfor.

Answer 1

[Edric Ellis]

The indexing restrictions in parfor range from those which are required to make the loop iterations provably order-independent (well... modulo non-standard implementations of subsref or subsasgn), to those which are limitations of the implementation. In this case, you have two indexing expressions either of which on its own would constitute a valid "slicing" operation, but the current implementation cannot support that.

Also note that the following is allowed:

parfor ii=1:10
    for jj = 1:2
        A(ii, jj) = jj;
    end
end