> Very nice!
> 
> I hope you got started on this because of the ParallelMesh work, not
> because you were running into the same out-of-memory errors that were
> killing Paul and I in FIN-S?  It turned out that the root problem was
> a memory leak in libablation svn trunk, now fixed there and in 0.25.0.

Yeah I started it because of the ParallelMesh stuff, but also going further
with a SerialMesh is always nice too ;-)

Plus if we do some stand-alone postprocessing of parallel runs it would be
nice to have the serial implementation be as lean as possible.

> Any notable changes in runtimes?  DofObject timing is in the noise on
> implicit non-matrix-free problems, but IIRC DofMap::dof_indices
> actually might show up on explicit and jacobian-free runs.

nothing notable yet.  I'm thinking if anything there will likely be a
reduction in runtime because of less indirect addressing - one contiguous
buffer that should hopefully fit in a cache line...  This will be
increasingly likely when I finally implement blocked variables.

-Ben


------------------------------------------------------------------------------
Achieve unprecedented app performance and reliability
What every C/C++ and Fortran developer should know.
Learn how Intel has extended the reach of its next-generation tools
to help boost performance applications - inlcuding clusters.
http://p.sf.net/sfu/intel-dev2devmay
_______________________________________________
Libmesh-devel mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/libmesh-devel

Reply via email to