ILU, bicgstab,cgs,bcg and time optimalization

[R!]

Hi,

I sent you some procedure  last months. I hope that i will send two more
(bcg and ILU). But I also have some questions to you, mainly connected with
speed of these procedures.

I noticed that now they are terribly slow, I tracked the problem and I have
found that it 's due to these functions which I'm currently using
cond() - is really really very slow on sparse matrix on which this function
is used, when checking if preconditioner isn't ill-conditioned so I think
that I can remove it quite safely (and rely on user that will give me good
preconditioner)

Next weak point is inv function currently I'm calling inv on my
preconditioner. I hoped that it will speedup my algorithm little bit(in next
cycle I can use multipy insted of divide with my preconditioner), but in
fact the inv procedure takes so much time that its rather drawback.


Another problem  which I have is with ILU. Now I'm comuting ILU and MILU
(column modification) My function currenty has the same Params as matlab but
some of them I'm ignoring and I'm affraid that I will not be able to
implement them in near future.  But these procedure is really sllow.the 
main problem is this

function auto(A)
    for i=1:5000 is
        v = A(i,i)+i*i;
    endfor
endfunction
this simple function takes about 0,2s on my PC but in matlab it takes about
0,002s. I think  that it the reason why my ilu procedure takes about 10
minutes to compute L U factors on matrix with 5000 rows. I'm not sure if
it's useful in this time and Will this procedure be a benefit for octave ?

Have got any suggestion or advice for me?

Thanks

-- 
View this message in context: 
http://www.nabble.com/ILU%2C-bicgstab%2Ccgs%2Cbcg-and-time-optimalization-tp22376290p22376290.html
Sent from the Octave - Maintainers mailing list archive at Nabble.com.