On Thu, 31 Jan 2013, Vikram Garg wrote:
Roy and I are tracking down an extremely nasty bug in residual
evaluations after solution storage and recovery. Hopefully, we will
get this bug hunted down today and be able to finish the remaining
work today.
So we've found the bug, and it's *not* a SolutionHistory bug, it's an
UnsteadySolver bug. Basically we've been doing localization of the
old solution vector before each solve(), which is the most efficient
time to do it when we're solving but which can mess up the results of
anything that tries to use a plain assembly() of a transient problem
without having done a solve() at the same time step immediately prior.
We're also not doing the resizing of the old solution vector at the
best place either, so while I'm shuffling things around tonight I'm
going to try to avoid some of the massive redundancy there. This may
involve API changes, in the worst case, but nothing
backwards-incompatible.
It'll be a fundamental enough and important enough change that I'd
like to get it done and tested with GRINS before we tag 0.9.0-rc1,
though.
---
Roy
------------------------------------------------------------------------------
Everyone hates slow websites. So do we.
Make your web apps faster with AppDynamics
Download AppDynamics Lite for free today:
http://p.sf.net/sfu/appdyn_d2d_jan
_______________________________________________
Libmesh-devel mailing list
Libmesh-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/libmesh-devel
