Alternative for unefficient code using corrcoef in a for-loop

Dear specialists, I have the following problem. The matrix G is a cross-realization cross-correlation matrix. It is calculated from correlation coefficients of large (~200,000 dimensional) vectors. G is symmetric, which is being accounted for by the if-loop. The following code seems to be very inefficient, as I know that for-loops are normally not the best method in MATLAB. However, I could not come up with an idea how to make it more efficient, so far. Maybe, one of you can give me a hint.

Best regards

Florian

G=zeros(nc*K);
for l=1:K
 for m=1:K
      if l<=m
          for i=1:nc
              for j=1:nc
                  [R,P]=corrcoef(x(l,i,:),x(m,j,:));
                  G((l-1)*nc+i,(m-1)*nc+j)=R(1,2);
              end
          end
      else
          G((l-1)*nc+1:l*nc,(m-1)*nc+1:m*nc)=G((m-1)*nc+1:m*nc,(l-1)*nc+1:l*nc)';
      end
  end
end

Definition of x:

for i=1:nc
      x(l,i,:)=double(reshape(nii.img(:,:,:,i),s,1));
end

where nii.img are three-dimensional images loaded in an earlier step. s here is ~ 200,000. K and nc are ~50.