It's nice to see DOE is still buying computers for 100s of millions of dollars that do not support a debugger.
> On Oct 23, 2023, at 11:22 AM, Mark Adams <[email protected]> wrote: > > > > On Mon, Oct 23, 2023 at 10:18 AM Barry Smith <[email protected] > <mailto:[email protected]>> wrote: >> >> Are they then not using any preallocation? > > I asked but have not gotten a response. > >> The "old dumb MatSetValues" used default preallocation. To reproduce that >> they can call MatXAIJSetPreallocation() with 10 and 3 as the local and >> nonlocal number of nonzeros per row. >> > > I am sure they use some sort of preallocation, or they would have seen bad > performance problems and they are pretty mature PETSc users, but we have to > wait and see. > >> Best to use -start_in_debugger or -on_error_attach_debugger to find the >> details of the crash >> > > They did give me a ddt --offline .hdftm file, attached > (nice when you can't open a window, eg, Frontier). > I asked them to try to get a line number. > > Thanks, > Mark > >> >> >> > On Oct 23, 2023, at 8:15 AM, Mark Adams <[email protected] >> > <mailto:[email protected]>> wrote: >> > >> > I have a Fortran user that is getting a segv in MatSetValues_MPIAIJ in >> > v3.19 and v3.20 and it works with v3.17. >> > They are trying to get a line number but I was thinking it might be worth >> > trying the old dumb MatSetValues. >> > Is that possible? >> > >> > Thanks, >> > Mark >> > <debug (50).html>
