Posted by mike puso on February 18, 1998 at 16:28:00:
We've tried SuperLU and compared it against other sparse direct solvers on a few matrices. We tried both the A*A(transpose) and A+A(transpose)
reodering and have found that on some matrices the A*At solved the matrix but A+At gave us segmentation faults on both DEC and SGI unix systems.
We are usually solving symmetric indefinite matrices so we would like the A+At ordering
since it wont add any fill. We are using the threaded parallel version, how well does
this solver scale up to 5 or 6 processors? We didnt get great results. The sparse direct solvers
that I have tried such as the library version psldlt_factori on the SGI and libsparse on the HP scale quite well.
I would be interested in hearing others experiences on SuperLU.