Is this parfor fixable?

aid = 100;

% matrices to fill
iass = NaN(mma,aid,2,3,aid+run);

% dummy to make parfor work
s = aid+1;

% part A: recalculating paths for those already alive
parfor (i = 2:aid,6)
    wag = wages(1:aid+1-i);
    r = rint(1:aid+1-i);
    gamma = CoSurv(i,:,1);
    iass(:,:,:,:,s-i) = recalc_paths(aid, pars, i-1);
end

With error "unable classify variable iass". It /should/ be properly sliced - each loop sets a "row" of iass, completely independently of all the other loops (but which does depend on the loop number i). Is there a way to fix this that I'm missing? I introduced the dummy "s" because it gave me a different error when I inserted aid+1 directly into the index, but clearly that wasn't the silver bullet I thought it would be.

Answer 1

[Gaurav Garg]

Hi Jacob,

Apparently, you cannot subtract the loop index (within parfor loop) from any other scalar/vector quantity.

As a workaround in your case, you can assign the final values returned by the function "recalc_paths" in a temporary variable indexed by i and then write another for loop (parfor won't work here, again due to the same reason/error of classify variable) to assign the values to iass(:,:,:,:,s-i).

Pseudo-Code:

% Declaring temp as a matrix of suitbale size and type
parfor (i = 2:aid,6)
    % Finding gamma and other values
    temp(i) = recalc_paths(aid, pars, i-1);
end
for (i = 2:aid,6)
    iass(:,:,:,:,s-i) = temp(i);
end

To know more about the error, you can refer to the doc here.

[Walter Roberson]

According to https://www.mathworks.com/help/coder/ug/classification-of-variables-in-parfor-loops.html#btfitu2-19

Form of Indexing. Within the list of indices for a sliced variable, one index is of the form i, i+k, i-k, k+i, or k-i.

• i is the loop variable.
• k is a constant or a simple (nonindexed) variable.
• Every other index is a constant, a simple variable, colon, or end.

In the user's case, s is a "simple (nonindexed) variable", and k-i form is being used.

The user is using R2019b, but the rules for subtraction where the same then; https://www.mathworks.com/help/releases/R2019b/coder/ug/classification-of-variables-in-parfor-loops.html

Therefor, the operation is documented as being permitted.

[Jacob Stevens]

Thank you both - this is indeed the solution I went for, although I used iass(:,:,:,2:end) = temp(:,:,:,end:-1:2) instead of the proposed for loop. But as Walter says, the document does indeed suggest this kind of indexing is allowed, so it seems strange that this workaround is necessary.