<html><head>
<meta http-equiv="Content-Type" content="text/html; charset=utf-8">
  </head>
  <body>
    <p>Dear all</p>
    <p>    please ignore my previous email and read this one: I have
      better localized the problem. Maybe DMSwarmMigrate is designed to
      migrate particles only to first neighbouring ranks?<br>
    </p>
    <div class="moz-cite-prefix">Il 22/12/22 11:44, Matteo Semplice ha
      scritto:<br>
    </div>
    <blockquote type="cite" cite="mid:a87d961b-31be-9a2c-4d5f-4bac09180788@uninsubria.it">
      
      <p>Dear everybody,</p>
      <p>    I have bug a bit into the code and I am able to add more
        information.<br>
      </p>
      <div class="moz-cite-prefix">Il 02/12/22 12:48, Matteo Semplice ha
        scritto:<br>
      </div>
      <blockquote type="cite" cite="mid:35ebfa58-eed5-39fa-8b3d-918ff9d7e633@uninsubria.it">
        
        <div class="moz-cite-prefix">Hi.</div>
        <div class="moz-cite-prefix">I am sorry to take this up again,
          but further tests show that it's not right yet.<br>
        </div>
        <div class="moz-cite-prefix"><br>
        </div>
        <div class="moz-cite-prefix">Il 04/11/22 12:48, Matthew Knepley
          ha scritto:<br>
        </div>
        <blockquote type="cite" cite="mid:CAMYG4Gnznu1ct7GhjYNcBVRqPmpuVNnz1O_4OqM-_s9fZ28=vA@mail.gmail.com">
          <div dir="ltr">
            <div dir="ltr">On Fri, Nov 4, 2022 at 7:46 AM Matteo
              Semplice <<a href="mailto:matteo.semplice@uninsubria.it" moz-do-not-send="true" class="moz-txt-link-freetext">matteo.semplice@uninsubria.it</a>>
              wrote:<br>
            </div>
            <div class="gmail_quote">
              <blockquote class="gmail_quote" style="margin:0px 0px 0px
                0.8ex;border-left:1px solid
                rgb(204,204,204);padding-left:1ex">
                <div> On 04/11/2022 02:43, Matthew Knepley wrote:<br>
                  <blockquote type="cite">
                    <div dir="ltr">
                      <div dir="ltr">On Thu, Nov 3, 2022 at 8:36 PM
                        Matthew Knepley <<a href="mailto:knepley@gmail.com" target="_blank" moz-do-not-send="true" class="moz-txt-link-freetext">knepley@gmail.com</a>>
                        wrote:<br>
                      </div>
                      <div class="gmail_quote">
                        <blockquote class="gmail_quote" style="margin:0px 0px 0px
                          0.8ex;border-left:1px solid
                          rgb(204,204,204);padding-left:1ex">
                          <div dir="ltr">
                            <div dir="ltr">On Thu, Oct 27, 2022 at 11:57
                              AM Semplice Matteo <<a href="mailto:matteo.semplice@uninsubria.it" target="_blank" moz-do-not-send="true" class="moz-txt-link-freetext">matteo.semplice@uninsubria.it</a>>
                              wrote:<br>
                            </div>
                            <div class="gmail_quote">
                              <blockquote class="gmail_quote" style="margin:0px 0px 0px
                                0.8ex;border-left:1px solid
                                rgb(204,204,204);padding-left:1ex">
                                <div>
                                  <div dir="ltr">
                                    <div style="font-family:Calibri,Arial,Helvetica,sans-serif;font-size:12pt;color:rgb(0,0,0);background-color:rgb(255,255,255)">Dear
                                      Petsc developers,</div>
                                    <div style="font-family:Calibri,Arial,Helvetica,sans-serif;font-size:12pt;color:rgb(0,0,0);background-color:rgb(255,255,255)">   
                                      I am trying to use a DMSwarm to
                                      locate a cloud of points with
                                      respect to a background mesh. In
                                      the real application the points
                                      will be loaded from disk, but I
                                      have created a small demo in which</div>
                                    <div style="font-family:Calibri,Arial,Helvetica,sans-serif;font-size:12pt;color:rgb(0,0,0);background-color:rgb(255,255,255)">
                                      <ul>
                                        <li><span>each processor creates
                                            Npart particles, all within
                                            the domain covered by the
                                            mesh, but not all in the
                                            local portion of the mesh</span></li>
                                        <li><span>migrate the particles</span></li>
                                      </ul>
                                      <div>After migration most
                                        particles are not any more in
                                        the DMSwarm (how many and which
                                        ones seems to depend on the
                                        number of cpus, but it never
                                        happens that all particle
                                        survive the migration process).</div>
                                      <div><br>
                                      </div>
                                    </div>
                                  </div>
                                </div>
                              </blockquote>
                              <div>Thanks for sending this. I found the
                                problem. Someone has some overly fancy
                                code inside DMDA to figure out the local
                                bounding box from the coordinates.</div>
                              <div>It is broken for DM_BOUNDARY_GHOSTED,
                                but we never tested with this. I will
                                fix it.</div>
                            </div>
                          </div>
                        </blockquote>
                        <div><br>
                        </div>
                        <div>Okay, I think this fix is correct</div>
                        <div><br>
                        </div>
                        <div>  <a href="https://eur01.safelinks.protection.outlook.com/?url=https%3A%2F%2Fgitlab.com%2Fpetsc%2Fpetsc%2F-%2Fmerge_requests%2F5802&data=05%7C01%7Cmatteo.semplice%40uninsubria.it%7Cf4d64b09df1f438437ad08dad45b342b%7C9252ed8bdffc401c86ca6237da9991fa%7C0%7C0%7C638055785720875500%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C2000%7C%7C%7C&sdata=ISLaLhhnYU4njkYfod%2F3tEiIOIV5uZvmiAlKZ2PvhmE%3D&reserved=0" originalsrc="https://gitlab.com/petsc/petsc/-/merge_requests/5802" shash="PuvAgB7girkyJOr0PaoyM6ND/gKKaTLgVWtpHONXwmBNgX1xXkxIqoTLn/9EWhBzck+xBHuOHElpbZM0juMi1YgTJ/9UK2cwFTmlByMgSl3Aa14QI0SetAqu8s4vEzsJDt51uhIxJ0xSMxzpYWLN8lxcQmJ6IhM3n5rYrzps9QY=" target="_blank" moz-do-not-send="true">https://gitlab.com/petsc/petsc/-/merge_requests/5802</a></div>
                        <div><br>
                        </div>
                        <div>I incorporated your test as
                          src/dm/impls/da/tests/ex1.c. Can you take a
                          look and see if this fixes your issue?</div>
                      </div>
                    </div>
                  </blockquote>
                  <p>Yes, we have tested 2d and 3d, with various
                    combinations of DM_BOUNDARY_* along different
                    directions and it works like a charm.</p>
                  <p>On a side note, neither <span style="font-family:monospace"><span style="color:rgb(0,0,0);background-color:rgb(255,255,255)">DMSwarmViewXDMF</span></span>
                    nor <span style="font-family:monospace"><span style="color:rgb(0,0,0);background-color:rgb(255,255,255)">DMSwarmMigrate</span></span>
                    seem to be implemented for 1d: I get</p>
                  <p><span style="font-family:monospace"><span style="color:rgb(0,0,0);background-color:rgb(255,255,255)">[0]PETSC
                        ERROR: No support for this operation for this
                        object type</span><span style="color:rgb(178,24,24);background-color:rgb(255,255,255)">
                                                                                                                                           </span><span style="color:rgb(0,0,0);background-color:rgb(255,255,255)"> </span>[0]PETSC
                      ERROR: Support not provided for 1D<br>
                    </span></p>
                  <p>However, currently I have no need for this feature.<br>
                    <span style="font-family:monospace"></span></p>
                  <p>Finally, if the test is meant to stay in the
                    source, you may remove the call to <span style="font-family:monospace"><span style="color:rgb(0,0,0);background-color:rgb(255,255,255)">DMSwarmRegisterPetscDatatypeField</span></span>
                    as in the attached patch.<br>
                  </p>
                  <p>Thanks a lot!!</p>
                </div>
              </blockquote>
              <div>Thanks! Glad it works.</div>
              <div>
                <p>   Matt</p>
              </div>
            </div>
          </div>
        </blockquote>
        <p class="moz-signature" cols="72">There are still problems when
          not using 1,2 or 4 cpus. Any other number of cpus that I've
          tested does not work corectly.<br>
        </p>
      </blockquote>
      <p>I have now modified private_DMDALocatePointsIS_2D_Regular to
        print out some debugging information. I see that this is called
        twice during migration, once before and once after
        DMSwarmMigrate_DMNeighborScatter. If I understand correctly, the
        second call to private_DMDALocatePointsIS_2D_Regular should be
        able to locate all particles owned by the rank but it fails for
        some of them because they have been sent to the wrong rank
        (despite being well away from process boundaries).<br>
      </p>
      <p>For example, running the example src/dm/impls/da/tests/ex1.c
        with Nx=21 (20x20 Q1 elements on [-1,1]X[-1,1]) with 3
        processors,</p>
      <p>- the particles (-0.191,-0.462) and (0.191,-0.462) are sent
        cpu2 instead of cpu0</p>
      <p>- those at (-0.287,-0.693)and (0.287,-0.693) are sent to cpu1
        instead of cpu0</p>
      <p>- those at (0.191,0.462) and (-0.191,0.462) are sent to cpu0
        instead of cpu2</p>
      <p>(This is 2d and thus not affected by the 3d issue mentioned
        yesterday on petsc-dev. Tests were made based on the release
        branch pulled out this morning, i.e. on commit <span style="font-family:monospace"><span style="color:#b26818;background-color:#ffffff;">bebdc8d016f</span></span>).<br>
      </p>
    </blockquote>
    <p>I see: particles are sent "all around" and not only to the
      destination rank.</p>
    <p>Still however, running the example src/dm/impls/da/tests/ex1.c
      with Nx=21 (20x20 Q1 elements on [-1,1]X[-1,1]) with 3 processors,
      there are 2 particles initially owned by rank2 (at y=-0.6929 and
      x=+/-0.2870) that are sent only to rank1 and never make it to
      rank0 and are thus lost in the end since rank1, correctly,
      discards them.</p>
    <p>Thanks</p>
    <p>    Matteo<br>
    </p>
  </body>
</html>