Thanks for reporting this. There appears to be a bug in MatGetSubMatricesParallel(). I'll look into it and add a test for the case above.
In the meantime, maybe you could tell me more about what you are trying to do? PCGASM: it is rather experimental and doesn't get much use ordinarily. Depending on your particular use case, we can either try improving PCGASM, or find a different way to accomplish the same thing. Thanks. Dmitry. On Mon, Nov 25, 2013 at 9:07 AM, Jed Brown <[email protected]> wrote: > Laurent Berenguer <[email protected]> writes: > > >>How exactly did you set this subksp? Did you modify the code or change > >>the command line arguments? Did the test run without errors when you > >>don't modify anything? > > > > I tried both methods, the error is the same in both cases : > > - I added the option -user_set_subdomain_solvers and modified the code > > from > > ierr = PCSetType(subpc,PCILU);CHKERRQ(ierr); > > to > > ierr = PCSetType(subpc,PCJACOBI);CHKERRQ(ierr); > > - Or I only added -sub_ksp_type gmres without changing the code. > > > > The example runs correctly when I change nothing. But I'm not sure > > what is the default subksp. When I change nothing in the code but it > > seems to be KSPPREONLY. > > It should say if you run with -ksp_view. > > PCGASM is Dmitry's project and I don't think it has gotten much use yet. > I'll leave it to him to explain what's going on here. If your cases are > supposed to work, they should be added as tests. >
