Re: Parallel algorithm for tridiagonal matrix


It looks to me like each line is uncoupled from the others. Thus, you
have 1000^2 independent systems of size 1000, and parallelism is easy.


You are absolutely right. However, the data is also distributed.
I am using 2D grid topology.
Nx (dimension of problem in x) is distributed on Px processors.
So they dont have the complete line.


.