Hi Wolfgang, thank you for spending so much time on that issue; I am happy I could at least contribute a tiny bit! If I happen to implement that functionality in the future, I will simply put in on github ; ) Kind regards, Richard
Wolfgang Bangerth schrieb am Mittwoch, 19. August 2020 um 00:13:59 UTC+2: > > Hi Richard, > > > I do have approx. the same numbers, but on 4 cores (see screenshot). > > > > Fortunately, I was able to adapt our little program here with success > > - which I append here just for the sake of completeness, since I went on > > github and tried my best to propose that change, see > > https://github.com/dealii/dealii/pull/10838 > > < > https://nam01.safelinks.protection.outlook.com/?url=https%3A%2F%2Fgithub.com%2Fdealii%2Fdealii%2Fpull%2F10838&data=02%7C01%7CWolfgang.Bangerth%40colostate.edu%7Cda1531178e7b4705abac08d843a89035%7Cafb58802ff7a4bb1ab21367ff2ecfc8b%7C0%7C0%7C637333738540558308&sdata=1VdEvclgxPPonuEzf1nrSpc0TGmge9QTTXlkJ1snjPk%3D&reserved=0 > > > > I hope that worked out, but I used git for the first time. > > It's the first time at some time for everyone -- but we hope it won't be > your > last patch! > > > > Nonetheless, some concerns of mine remain: > > (i) if I allow in my application-code to actually change the row scaling > of > > the right matrix in the product (assuming the vector has all nonzero > entries), > > I will actually get away with only 1 additional matrix being the product > > instead of 2 including the product and a temporary one (which was the > one not > > destroyed) > > by the cost of scaling and rescaling the right matrix in the product. > > Right. You might not actually need this function, but it's still worth > fixing :-) > > > > (ii) The documentation says, that both Trilinos and PETSc actually > rebuilding > > the sparsity pattern of the product, > > so using compute_mmult_pattern in the setup of the product would be of > no use, > > right? > > For the moment that's correct. But the PETSc documentation provides a way > how > one could re-use a previously computed sparsity pattern. One could add a > flag > to the mmult function that switches between either telling PETSc to create > a > new sparsity pattern, or re-using the existing one. > > If you only call the function once, I have no doubt that there is not much > of > a difference between using the deal.II and PETSc ways of creating sparsity > patterns. It's a different story if you plan on calling this function > multiple > times using the same sparsity pattern. > > Best > W. > > > -- > ------------------------------------------------------------------------ > Wolfgang Bangerth email: bang...@colostate.edu > www: http://www.math.colostate.edu/~bangerth/ > > -- The deal.II project is located at http://www.dealii.org/ For mailing list/forum options, see https://groups.google.com/d/forum/dealii?hl=en --- You received this message because you are subscribed to the Google Groups "deal.II User Group" group. To unsubscribe from this group and stop receiving emails from it, send an email to dealii+unsubscr...@googlegroups.com. To view this discussion on the web visit https://groups.google.com/d/msgid/dealii/627e7d2f-d6ce-41c0-8e72-c63a2b7fdef1n%40googlegroups.com.
