You can try making a copy of this function, and editing it, so you avoid the sparse function, as that is causing the problem here:
i=[1 2 1];j=[1 1 2];s=[3 2 1];
D_original_code=sparse(i,j,s);
D=zeros(max(i),max(j));
D(sub2ind(size(D),i,j))=s;
isequal(D,full(D_original_code))
The tricky thing is that the sparse function is there for a reason: it makes things more efficient. You only need to hope either method (using full or filling a zeros) will not decrease the efficiency so much that generating C code and running the compiled code is not useful anymore.
Last note: don't edit the actual Matlab function, but create a copy. That might mean you are no longer allowed to share your compiled function, as it contains code by the Mathworks.
