Variable in parfor cannot be classified

I've looked for answers on this forum but each situation seems to different from mine. My code sends some data to a function which outputs many variables which are defined as sets of tau coefficients (or a tau by tau matrix) on a Nx by Ny mesh.

Here is my code:

speedmaxF = 0;
system_cell = NaN(tau,Nx,Ny);
trunc_east = NaN(tau,Nx,Ny);
trunc_west = NaN(tau,Nx,Ny);
Jac_trunc_east = NaN(tau,tau,Nx,Ny);
Jac_trunc_west = NaN(tau,tau,Nx,Ny);  
Jac_cell = NaN(tau,tau,Nx,Ny);
Jac_east_cell = NaN(tau,tau,Nx,Ny);
Jac_west_cell = NaN(tau,tau,Nx,Ny);
Jac_east_other = NaN(tau,tau,Nx,Ny);
Jac_west_other = NaN(tau,tau,Nx,Ny);
flux_east = NaN(tau,Nx,Ny);
flux_west = NaN(tau,Nx,Ny);

DGprev_ghosts_east = NaN(tau,Nx,Ny);
DGprev_ghosts_west = NaN(tau,Nx,Ny);
DGprev_ghosts_east(:,1:(end-1),:,:) = DGprev_ghosts(:,2:end,:,:);
DGprev_ghosts_west(:,2:end,:,:) = DGprev_ghosts(:,1:(end-1),:,:);

parfor index1 = 1:Nx
    for index2 = 1:Ny
        qstar = DGprev_ghosts(:,index1,index2);
        qeast = DGprev_ghosts_east(:,index1,index2);
        qwest = DGprev_ghosts_west(:,index1,index2);

          v1center = v1centers(index1); 
          v2center = v2centers(index2); 
          cellcenter = [v1center v2center];

          [trunc_east(:,index1,index2),trunc_west(:,index1,index2), ...
            flux_east(:,index1,index2), flux_west(:,index1,index2), ...
          Jac_east_cell(:,:,index1,index2),Jac_west_cell(:,:,index1,index2), ...
          Jac_east_other(:,:,index1,index2),Jac_west_other(:,:,index1,index2), ...
          Jac_trunc_east(:,:,index1,index2),Jac_trunc_west(:,:,index1,index2), ...
          Jac_cell(:,:,index1,index2),system_cell(:,index1,index2),maxspeedF] ...
          = predictor_precompute_eastwest(qstar,qeast,qwest,qstar,data,cellcenter);

          speedmaxF(iv1ghost) = max([speedmaxF(iv1ghost) maxspeedF]);
  end
  end

The error message is:

Error: File: predictor_main.m Line: 146 Column: 10
The variable trunceast in a parfor cannot be classified.
See Parallel for Loops in MATLAB, "Overview".

I have tried by adding things like truncteast = trunceast after the loop, but nothing seems to work. Suggestions?

[jgg]

I think the issue is that because you're using two indexes, both of which can be called in any loop iteration (but aren't, in reality) Matlab is having a hard time understanding the structure of your variables. Try this and see if it works:

 parfor index1 = 1:Nx
    trunc_i = NaN(1,Ny);
    for index2 = 1:Ny
           %Stuff

          [trunc_i(index2),%Stuff]  = predictor_precompute_eastwest(qstar,qeast,qwest,qstar,data,cellcenter);
          speedmaxF(iv1ghost) = max([speedmaxF(iv1ghost) maxspeedF]);
  end
  trunc_east(index1,:) = trunc_i;
  end

Basically, try using a temporary variable inside the inner loop to store the values you need to assign to trunc_east then assign it only once at the end of the outer loop.

You'll have to to do the same thing with your other variables as well.

I'd had success with this in the past; hopefully it will work here.

[Walter Roberson]

Yes, that is an appropriate approach: write into a temporary vector in the inner loop, and assign the vector as a whole to the appropriate output space in the outer loop.

[Pierson]

Thanks! Very helpful. Any ideas on how to make sure that the speedup in this code is optimal?