Karl Rupp <[email protected]> writes:

Jed gave me a good hint with respect to D-ILU0, which I'll also add to PETSc. As with other GPU-accelerations using ILU, it will require a proper matrix ordering to give good performance.

Well, factorization time is not the bottleneck in ILU(0) so there's not much sense making the preconditioner weaker to get more parallelism in that part. It might be nice to have a CPU implementation just so we can compare strength.

I'm somewhat tempted to port the SA-AMG implementation in CUSP to OpenCL as well, but this certainly won't be in 1.5.0.

Rather than "porting" SACUSP to distributed memory, we are taking a different approach: providing the matrix-matrix operations for use with PCGAMG.