Hi, I wanna if there is any method to efficiently solve a 1-d tridiagonal linear system quation on a vector machine. Any idea?

Thanks in advance.

Solution from tridiagonal linear equations could be computed with Thoma's algorithm, wich is "pivot de Gauss" method for this particular case.

Do you really need a vector machine for this ?

Please, ask for the algorithm.

Best regards,

Christophe

The vector rate of Thoma's algorithm is too low when it runs on a parallel machine. I want to improve the vectorization. of the following do-loop.

<DD> do I=1,IMAX <DD> D(I) = D(I) -W(I)*D(I-1) <DD> enddo </DD>

Thanks Zhong Lei

Dear Zhong Lei,

Apologises for this late answer due to sparse internet connection.

After a read of J.C. Chien's post about vector machines, I could not guess how to vectorize a loop which requires previous iteration step result.

As I am mostly newbie to this forum, I can not confirm own opinion that meaningless requests may be a reason for none, or very few answers.

If you need running some parallel machines, couldn't you run some domain decomposition method (ddm) ?

I had an internet link to those methods, but could not find it before this post.

Maybe someone else here knows where to find informations about ddm(s).

Yours,

Christophe