Glad to hear that. I'll discuss a more automated solution with Roy and see about getting that rolled in to the next release.
----- Original Message ----- From: luyi <[EMAIL PROTECTED]> To: Kirk, Benjamin (JSC-EG); [EMAIL PROTECTED] <[EMAIL PROTECTED]>; [email protected] <[email protected]> Sent: Mon May 19 21:15:42 2008 Subject: Re: [Libmesh-users] about the long time for the assemble of first step! Hello,Ben, I change my libmesh to SVN code and the implicit part works well with --implicit_neighbor_dofs on the command line, the first step is much faster than libmesh0.62. Thank you very much. Kirk, Benjamin (JSC-EG) 写道: > Can you try running the code with --implicit_neighbor_dofs on the > command line? > By default the matrix sparsity is computed assuming continuous shape > functions, but for a DG forumation this is not a good assumption. > Let me know if that works for you, if so we can look at putting > something more sophisticated in the library to detect this situation > and handling it automatically. > -Ben > > ------------------------------------------------------------------------ > *From:* [EMAIL PROTECTED] on behalf of Roy > Stogner > *Sent:* Sun 5/11/2008 10:31 PM > *To:* luyi > *Cc:* [email protected] > *Subject:* Re: [Libmesh-users] about the long time for the assemble of > first step! > > > On Mon, 12 May 2008, luyi wrote: > > > Now I write a DG programme on libmesh, when I use the FIRST or higher > > order basis, the first step is too slow, > > besides this everything is ok, I use XYZ basis and I check the code by > > perf_log, the time ratio is up to snuff. > > The first time step (especially in a non-adaptive run) has some > overhead that later timesteps don't, but it's also possible you've > found a bug. If we're not preallocating the sparse matrix with the > correct sparsity pattern, for example, PETSc might take quite a long > time on the first assembly. > > Could you run your code for only one time step and check the perf_log > on that, to see what part of the code is taking too long? > --- > Roy > > ------------------------------------------------------------------------- > This SF.net email is sponsored by the 2008 JavaOne(SM) Conference > Don't miss this year's exciting event. There's still time to save $100. > Use priority code J8TL2D2. > http://ad.doubleclick.net/clk;198757673;13503038;p?http://java.sun.com/javaone > _______________________________________________ > Libmesh-users mailing list > [email protected] > https://lists.sourceforge.net/lists/listinfo/libmesh-users > ------------------------------------------------------------------------- This SF.net email is sponsored by: Microsoft Defy all challenges. Microsoft(R) Visual Studio 2008. http://clk.atdmt.com/MRT/go/vse0120000070mrt/direct/01/ _______________________________________________ Libmesh-users mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/libmesh-users
