Join GitHub today
GitHub is home to over 28 million developers working together to host and review code, manage projects, and build software together.Sign up
Jacobi, Gauss-Seidel & SOR iteration for SparseMatrixCSC using iterators #156
Unfortunately the simple stationary methods like Gauss-Seidel and SOR are a bit awkward to implement for column-major (sparse) matrices.
My goal with this PR is to support stationary methods without copying or transforming the underlying SparseMatrixCSC, while still exploiting in-place updates as much as possible.
* It should be avoided to compute
At this point the sparse version on full matrices is significantly faster... :p
The current / full matrix version computes
For dense matrices it makes sense to use the same column-wise approach as in the sparse implementation. See https://gist.github.com/haampie/6d24c91df406e9574f3d4cd451db2260
For matrices of size 300x300 you get a 2x speedup (Jacobi):
Or about 1.5x in SOR: