<html>
  <head>
    <meta http-equiv="Content-Type" content="text/html; charset=utf-8">
  </head>
  <body text="#000000" bgcolor="#FFFFFF">
    <p>Hi, <br>
    </p>
    <p>Do you have any idea when Barry's fix
(<a class="moz-txt-link-freetext" href="https://bitbucket.org/petsc/petsc/pull-requests/1606/change-handling-of-matptap_mpiaij_mpimaij/diff">https://bitbucket.org/petsc/petsc/pull-requests/1606/change-handling-of-matptap_mpiaij_mpimaij/diff</a>)
      will be released? I can see it has been merged to the "next"
      branch. Does it mean it will be soon available on master? <br>
    </p>
    <p>+for your information, I plotted a summary of the scalings of
      interest (memory and time):<br>
      - using petsc-3.10.2 (ref "bad" scaling)<br>
      - using petsc-3.6.4 (ref "good" scaling)<br>
      - using commit d330a26 + Barry's fix and different algorithms
      (none, scalable, allatonce, allatonce_merged)</p>
    <p>Best regards,</p>
    <p>Myriam<br>
    </p>
    <br>
    <div class="moz-cite-prefix">Le 05/13/19 à 17:20, Fande Kong a
      écrit :<br>
    </div>
    <blockquote type="cite"
cite="mid:CAN5Wd-KMuSCHUQ8AhGNT3zFLiudz+NMZoe8RxjB8wo11gHZH6w@mail.gmail.com">
      <meta http-equiv="content-type" content="text/html; charset=utf-8">
      <div dir="ltr">
        <div dir="ltr">
          <div dir="ltr">
            <div dir="ltr">
              <div dir="ltr">
                <div dir="ltr">
                  <div dir="ltr">
                    <div dir="ltr">
                      <div dir="ltr">
                        <div>Hi Myriam,</div>
                        <div><br>
                        </div>
                        <div>Thanks for your report back.</div>
                        <br>
                        <div class="gmail_quote">
                          <div dir="ltr" class="gmail_attr">On Mon, May
                            13, 2019 at 2:01 AM Myriam Peyrounette <<a
                              href="mailto:myriam.peyrounette@idris.fr"
                              moz-do-not-send="true">myriam.peyrounette@idris.fr</a>>
                            wrote:<br>
                          </div>
                          <blockquote class="gmail_quote"
                            style="margin:0px 0px 0px
0.8ex;border-left-width:1px;border-left-style:solid;border-left-color:rgb(204,204,204);padding-left:1ex">Hi
                            all,<br>
                            <br>
                            I tried with 3.11.1 version and Barry's fix.
                            The good scaling is back!<br>
                            See the green curve in the plot attached. It
                            is even better than PETSc<br>
                            3.6! And it runs faster (10-15s instead of
                            200-300s with 3.6).<br>
                          </blockquote>
                          <div><br>
                          </div>
                          <div>We are glad your issue was resolved
                            here. </div>
                          <div> </div>
                          <blockquote class="gmail_quote"
                            style="margin:0px 0px 0px
0.8ex;border-left-width:1px;border-left-style:solid;border-left-color:rgb(204,204,204);padding-left:1ex"><br>
                            So you were right. It seems that not all the
                            PtAPs used the scalable<br>
                            version.<br>
                            <br>
                            I was a bit confused about the options to
                            set... I used the options:<br>
                            -matptap_via scalable and
                            -mat_freeintermediatedatastructures 1. Do
                            you<br>
                            think it would be even better with
                            allatonce?<br>
                          </blockquote>
                          <div><br>
                          </div>
                          <div>"scalable" and "allatonce" correspond to
                            different algorithms respectively.
                            ``allatonce" should be using less memory
                            than "scalable". The "allatonce" algorithm
                             would be a good alternative if your
                            application is memory sensitive and the
                            problem size is large. </div>
                          <div>We are definitely curious about the
                            memory usage of ``allatonce" in your test
                            cases but don't feel obligated to do these
                            tests since your concern were resolved now.
                            In case you are also interested in how our
                            new algorithms perform, I post petsc options
                            here that are used to </div>
                          <div>choose these algorithms:</div>
                          <div><br>
                          </div>
                          <div>algorithm 1: ``allatonce" </div>
                          <div><br>
                          </div>
                          <div>-matptap_via allatonce<br>
                          </div>
                          <div>-mat_freeintermediatedatastructures 1<br>
                          </div>
                          <div><br>
                          </div>
                          <div>algorithm 2: ``allatonce_merged" <br>
                          </div>
                          <div><br>
                          </div>
                          <div>
                            <div>-matptap_via allatonce_merged<br>
                            </div>
                            <div>-mat_freeintermediatedatastructures 1</div>
                          </div>
                          <div><br>
                          </div>
                          <div><br>
                          </div>
                          <div>Again, thanks for your report that help
                            us improve PETSc.</div>
                          <div><br>
                          </div>
                          <div>Fande,</div>
                          <div> </div>
                          <blockquote class="gmail_quote"
                            style="margin:0px 0px 0px
0.8ex;border-left-width:1px;border-left-style:solid;border-left-color:rgb(204,204,204);padding-left:1ex"><br>
                            It is unfortunate that this fix can't be
                            merged with the master branch.<br>
                            But the patch works well and I can consider
                            the issue as solved now.<br>
                            <br>
                            Thanks a lot for your time!<br>
                            <br>
                            Myriam<br>
                            <br>
                            <br>
                            Le 05/04/19 à 06:54, Smith, Barry F. a
                            écrit :<br>
                            >    Hmm, I had already fixed this, I
                            think, <br>
                            ><br>
                            >    <a
href="https://bitbucket.org/petsc/petsc/pull-requests/1606/change-handling-of-matptap_mpiaij_mpimaij/diff"
                              rel="noreferrer" target="_blank"
                              moz-do-not-send="true">https://bitbucket.org/petsc/petsc/pull-requests/1606/change-handling-of-matptap_mpiaij_mpimaij/diff</a><br>
                            ><br>
                            >    but unfortunately our backlog of
                            pull requests kept it out of master. We are
                            (well Satish and Jed) working on a new CI
                            infrastructure that will hopefully be more
                            stable than the current CI that we are
                            using.<br>
                            ><br>
                            >    Fande,<br>
                            >       Sorry you had to spend time on
                            this. <br>
                            ><br>
                            ><br>
                            >    Barry<br>
                            ><br>
                            ><br>
                            ><br>
                            >> On May 3, 2019, at 11:20 PM, Fande
                            Kong via petsc-users <<a
                              href="mailto:petsc-users@mcs.anl.gov"
                              target="_blank" moz-do-not-send="true">petsc-users@mcs.anl.gov</a>>
                            wrote:<br>
                            >><br>
                            >> Hi Myriam,<br>
                            >><br>
                            >> I run the example you attached
                            earlier with "-mx 48 -my 48 -mz 48 -levels 3
                            -ksp_view  -matptap_via allatonce -log_view
                            ".  <br>
                            >><br>
                            >> There are six PtAPs. Two of them
                            are sill using the nonscalable version of
                            the algorithm (this might explain why the
                            memory still exponentially increases) even
                            though we have asked PETSc to use the
                            ``allatonce" algorithm. This is happening
                            because MATMAIJ does not honor the petsc
                            option, instead, it uses the default setting
                            of MPIAIJ.  I have a fix at <a
href="https://bitbucket.org/petsc/petsc/pull-requests/1623/choose-algorithms-in/diff"
                              rel="noreferrer" target="_blank"
                              moz-do-not-send="true">https://bitbucket.org/petsc/petsc/pull-requests/1623/choose-algorithms-in/diff</a>.
                            The PR should fix the issue.<br>
                            >><br>
                            >> Thanks again for your report,<br>
                            >><br>
                            >> Fande,<br>
                            >><br>
                            >>  <br>
                            <br>
                            -- <br>
                            Myriam Peyrounette<br>
                            CNRS/IDRIS - HLST<br>
                            --<br>
                            <br>
                            <br>
                          </blockquote>
                        </div>
                      </div>
                    </div>
                  </div>
                </div>
              </div>
            </div>
          </div>
        </div>
      </div>
    </blockquote>
    <br>
    <pre class="moz-signature" cols="72">-- 
Myriam Peyrounette
CNRS/IDRIS - HLST
--
</pre>
  </body>
</html>