Parfor - classification of variables in multiple nested loops

My variable "tmp" cannot be classified. How can this be edited for it to work in parfor form?

w = [1:5];
x = [1:5];
y = [1:5];
z = [1:5];

parfor i=1:length(w);
   for j=1:length(x);
      for k=1:length(y);
         count=0;
         for l=z;
           count=count+1;
           file = ['filepath.' k '-' l '.nc'];
           ncfile = netcdf.open(file);
           t = netcdf.getVar(ncfile,1);
           tmp(count,y+1) = t(w(i),x(j));
         end
      end
   end
end

[Walter Roberson]

Caution: ['filepath.' k '-' l '.nc'] is going to use the numeric value of k and l in the indicated positions, so the code is equivalent to

 file = ['filepath.' char(k) '-' char(l) '.nc'];

with k and l each being in the range 1 to 5, that is going to be using the characters SOH, STX, ETX, EOT, ENQ in those positions. If you want '1' through '5' to be used instead of char(1) through char(5) then you need to convert the numbers to text. For example,

file = sprintf('filepath.%d-%d.nc', k, l);

Answer 1

[Walter Roberson]

Your parfor is over "i". Essentially your output needs to be indexed over "i". Not indexed over w(i), over i directly. Cell arrays and struct arrays are acceptable. For example,

data_w{i} = ....

would be acceptable but not

data_w{w(i)}

It would be allowed to use

parfor i = w

and index by i, but on output to data_w{i} the cells used would not be, in the general case be adjacent. With those particular w values they would be adjacent.

You effectively only get one shot at assigning to something indexed at the parfor loop variable, so all nested "for" should be building into a variable and then the last statement inside the parfor should be to copy that variable into the locations indexed at the loop variable. It is acceptable to copy entire columns or sub-matrices, such as

data_w(:,:,:,i) = data_built_in_the_loop;

provided of course that the data is always exactly the same size.