Triangular Backsubstitution

[GoogleCodeExporter]

Is there an easy way to implement an efficient parallel sparse triangular
matrix backsubstitution step using cusp/thrust in the library? This would
enable the utilization of more elaborate preconditioners like SSOR or
custom incomplete Cholesky / LU to be easily integrated into the krylov pcg
solver.

Original issue reported on code.google.com by janick.m...@gmail.com on 19 Feb 2010 at 3:47

[GoogleCodeExporter]

Not to my knowledge.  In general, I don't know of any to implement 
backsubstitution
efficiently with 1000s of threads on a GPU.

However, if find any references that explain how it can be done, I'd be happy 
to look
at them.

Original comment by wnbell on 26 Feb 2010 at 3:00

• Added labels: Type-Enhancement
• Removed labels: Type-Defect

[GoogleCodeExporter]

Original comment by wnbell on 26 Mar 2010 at 3:28

• Added labels: Milestone-Unscheduled

[GoogleCodeExporter]

The algorithm in Cusparse works reasonably well.

Original comment by paulmull...@gmail.com on 4 Jan 2012 at 11:44

[GoogleCodeExporter]

There is version of the triangular solve method by Li and Saad in one of the 
clones.  In my opinion it isn't very robust without additional functionality to 
perform reordering to reduce the number of sequential stages.

Original comment by sdalton1@gmail.com on 28 Feb 2012 at 2:13