MATLAB: Parfor loop variable cannot be classified!

classifiedparallel computingParallel Computing Toolboxparfor

When I try to execute the above code with parfor instead of using for, I get the following error :

Error: The variable study in a parfor cannot be classified. See Parallel for Loops in MATLAB, "Overview".

I think the "Form of Indexing : Within the list of indices for the variable, exactly one index involves the loop variable." rule for sliced variables in parfor is causing the issue. (Because study(i).run(j) has not been indexed according to parfor requirements.) But since I have a structure with different fields and I need to update the field 'run' in the structure 'study', is there any way i could make this code compatible with parfor ? Any help/suggestion is really appreciated!

Best Answer

No, a parallelization of this code is not possible. E.g. in this line:

study(i).run(j).result = ...

You write a specific value, which depends on the PARFOR loop counter j. Afterwards you try to save the struct study to a file:

save([study_path study(i).name '.mat'], 'study');

But then this file will be overwritten in each iteration and the contents of the struct depends on the currently processed thread. This is such confusing and the resulting file will contain the struct as created in the last processed thread - which is a kind of random.

At first I'd move the writing of the file outside the PARFOR loop, such that 1. the slow and colliding disk access does not slow down the program dramatically and 2. the final value of the file is well-defined and reproducible.

When the netsed struct is still a problem, try a cell (what is "i"?!):

runC = cell(1, size(study(i).J, 1));
...

Sorry, I give up here. The nested creation of nested structs looks too confusing. But the general idea would to create the different data separately in a cell array and insert them to the struct afterwards.

Related Solutions

MATLAB: Parfor Closed loop variable

You can use the parfor to parallelize computation in individual iterations, possibly reducing time per iteration. However, you cannot parallelize across iterations. The iterations lack the parallel independence assumed by PARFOR.

MATLAB: Am I unable to assign values to matrix blocks like A(10:20,i) inside the PARFOR loop in Parallel Computing Toolbox 4.1 (R2009a)

This error message obtained is the expected behavior when using PARFOR loops in Parallel Computing Toolbox 4.1 (R2009a).

PARFOR loops enforce stringent conditions on variables and expressions that appear in the loop body. As per the documentation of PARFOR, the variable A will be treated as a sliced variable as its value can be broken up into segments, each of which can be operated on separately by different workers.

The rules for indexing a sliced variable clearly state that within the list of indices for a sliced variable, one of these indices is of the form i, i+k, i-k, k+i, or k-i, where i is the loop variable and k is a constant or a simple (nonindexed) variable; and every other index is a constant, a simple variable, colon, or end.

Hence using an index of the form

 A(2:4, i)

is not permitted.

If the indices to be extracted are known apriori, the following workaround can be used

A = rand(10, 10);
b = (2:4);
parfor i = 1:4
      A(b, i) = ones(3, 1);
end

The above code snippet adheres to the conditions imposed on sliced variables.

Best Answer

Related Solutions

MATLAB: Parfor Closed loop variable

MATLAB: Am I unable to assign values to matrix blocks like A(10:20,i) inside the PARFOR loop in Parallel Computing Toolbox 4.1 (R2009a)

Related Question