On Sun, 19 Apr 2009, David Knezevic wrote: > OK, so I've got a simple version of what we've discussed working. > > I've attached a patch, as well as some new files for the instantiation of > FEScalar. > > Some notes: > - As a first cut, I used lots of hacks in dof_map.h that will only work in > serial
Will only work in serial, and will only work when there's no more than one SCALAR variable added, right? Would it be inconvenient if we fixed those two things before adding the patch to SVN? If the answer is "yes", I'll add it now. (likewise if the answer is "shouldn't you be getting --enable-parmesh and --enable-ghosted working already?") Patch looks great otherwise. > - I used Roy's idea of letting the Order of SCALAR indicate how many scalar > variables we want -- I guess then it would be good to make sure that we only > add one SCALAR to a system, but I haven't done that yet... It could be more convenient in the long run if we enabled adding many SCALARs to a system - some multiphysics codes might be made up of parts that each want to add their own scalar variable(s) without first consulting with each other. > - I didn't (yet) make any changes to the sparsity pattern, so adding to the > last row(s) and column(s) of the matrix will be slow Horrifyingly slow, from the users' reports we've had over the years. PETSc really was designed to work best if you get the memory allocation right the first time. But your priorities are right: get things working and tested now and we'll optimize later. > - Plotting the SCALAR variables gives garbage at the moment, although I guess > it should be easy to change GMVIO etc to plot the constant value of the > SCALAR Changing GMVIO would only be the start - then we have to change VTKIO, then every other output format. For formats that allow "metadata" scalars to be attached, that's how we need to do it, but for general visualization formats, it might be convenient to just make FE<SCALAR> behave enough like FE<MONOMIAL> that it works on its own (for low-Order scalars). --- Roy ------------------------------------------------------------------------------ Crystal Reports - New Free Runtime and 30 Day Trial Check out the new simplified licensign option that enables unlimited royalty-free distribution of the report engine for externally facing server and web deployment. http://p.sf.net/sfu/businessobjects _______________________________________________ Libmesh-devel mailing list Libmesh-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/libmesh-devel
