alternatives to create a huge 9 diagonal matrix to spdiags

I have a problem I am working with a 9 diagonal linear system, what I am doing is finding the vectors of each diagonal, let me called Ai and then using spdiags function to form the coeficient matrix but my matrix is huge somenthing like 1 million x 1 million and spdiags is taking like 1.5 h just for doing the matrix is there any better way to do this. I would appraciate any help. Let me give you some code

    [row,col,bands]=size(I)
    Diag=[A1 A2 A3 A4 A5 A6 A7 A8 A9]; %Ai: col vector of ith diagonal 
    n=length(A5);%A5 main diagonal
    indx=[-row-1;-row;-row+1;-1;0;1;row-1;row;row+1];  
    A=spdiags(Diag,indx,n,n)

[David Kusnirak]

Nothing new here? I'm bearing the same problem.

Answer 1

[Jan]

Perhaps this helps:

[row,col,bands] = size(I)
Diag = [A1 A2 A3 A4 A5 A6 A7 A8 A9];
n    = length(A5);
indx = [-row-1;-row;-row+1;-1;0;1;row-1;row;row+1];
A = spalloc(numel(Diag), n, n);
for i = 1:9
  A = spdiags(Diag(:, i), indx(i), A);
end

All A1 to A9 have the same length, correct? I've reserved more memory for A than necessary. I assume, you can fix this according to the value of "row".

Does this create the wanted result? If so, I'm not sure, why SPDIAGS is so much slower when inserting all diagonals together.

[Maider Marin]

thanks, I will try it and let you know

[Maider Marin]

Yes all Ai have the same length

[Maider Marin]

It takes almost the same time...I think I just found matlab boundary...I will try to avoid to create the coefficient matrix by programming PCG on my own. Thanks

[Jan]

What a pitty.