How to implement the "tsearchn" algorithm 2 in a parfor loop?

Hi!

I have been trying to implement the following for loop in the "tsearchn" function using a parfor.

t = nan(npt,1);            % Simplex containing corresponding input point
p = nan(npt,ndim+1);       % Barycentric coordinates for corresponding input point
X = [ones(size(x,1),1) x]; % Append 1s to vertex matrix
b = [ones(npt,1) xi];      % Append 1s to point matrix
parfor i = 1:ntri             % Return the largest simplex index
  % For each triangle
  q = b / X(tri(i,:),:);   % Compute barycentric coordinate of each point
  I = all(q > myeps,2);    % Find simplex where all coordinates are positive
  t(I) = i;                % Set simplex
  p(I,:) = q(I,:);         % Set barycentric coordinates
end

MATLAB gives me the following errors:

"The variable t in a parfor cannot be classified."

then in editor I get:

"Valid indices for t are restricted in PARFOR loops."

The same is true for the variable p.

Following the suggested links did not elucidate the matter for me.

Does anybody know how to convert it into a PARFOR loop?

Thanks!

Answer 1

[Edric Ellis]

The problem here is that the loop iterations are order-dependent. parfor can only operate when the iterations are independent - and the restrictions on indexing into t and p simply reflect that. Outputs from a parfor loop typically need to be sliced, as described in the documentation. Basically, this means that you must assign elements corresponding to the loop index only.

Answer 2

[DroneDoctor]

Thanks for the answer. The Parallel Toolbox staff explained the issue in more detail in my service request. I was able to parallelize it and do some vectorizations. Thanks for your answer though. Highly appreciate it!

[James Cairns]

Is there any change someone could explain how this was achieved? I realise this is an old thread but I'm trying to achieve exactly the same as above.