[petsc-dev] Shifting in ICC, changes?

[Jed Brown]

What exactly changed between Feb 1 and the petsc-3.1 release with respect to
shifting (including for (S)BAIJ).  In Feb, I saw negative pivots and blamed
certain performance characteristics on it.  Now I don't see them, despite
the fact that SBAIJ does not implement the same shifts that AIJ does.  The
performance now (or with 3.1) is significantly better, but I would like to
be able to explain what's going on.  Running with -pc_factor_shift_type none
is still working much better than Feb 1 -dev and I would like to know why.

Jed
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/petsc-dev/attachments/20101203/d6777327/attachment.html>