<div dir="ltr"><div dir="ltr"><div dir="ltr">On Thu, Feb 21, 2019 at 1:19 PM Thibaut Appel <<a href="mailto:t.appel17@imperial.ac.uk">t.appel17@imperial.ac.uk</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 bgcolor="#FFFFFF">
    <p>Hi Matthew,</p>
    <p>Is your first part of your answer (using DMDASetBlockFills) valid
      only in the case I create a DMDA object?</p>
    <p>Yes I think that is the kind of stencil I am using. I could know
      how the stencil looks like exactly, but I preallocate looping, for
      each process on all the elements of the stencil, grid node by grid
      node (which is not that costly, and "exact")</p>
    <p>If I do NOT use a DMDA object and create my MPIAIJ matrix myself,
      how do I get the global row indices owned by the process (the
      "DMDA-like" ones you mentioned)?</p></div></blockquote><div>I think you want</div><div><br></div><div>  <a href="https://www.mcs.anl.gov/petsc/petsc-current/docs/manualpages/DMDA/DMDAGetLocalInfo.html">https://www.mcs.anl.gov/petsc/petsc-current/docs/manualpages/DMDA/DMDAGetLocalInfo.html</a></div><div><br></div><div>This tells you exactly what patch is owned by the process, and also how big the overlap is. Its not the same as G2L, but do you need</div><div>the whole G2L, or just the sizes?</div><div><br></div><div>  Thanks,</div><div><br></div><div>    Matt</div><div> </div><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex"><div bgcolor="#FFFFFF"><p> The problem is that
      MatGetOwnershipRange cannot be called if the matrix hasn't been
      preallocated, and I need the global indices to preallocate.<br>
    </p>
    <p>Thibaut<br>
    </p>
    <p><br>
    </p>
    <div class="gmail-m_-4123279978571850831moz-cite-prefix">On 21/02/2019 17:49, Matthew Knepley
      wrote:<br>
    </div>
    <blockquote type="cite">
      
      <div dir="ltr">
        <div dir="ltr">
          <div dir="ltr">On Thu, Feb 21, 2019 at 11:16 AM Thibaut Appel
            via petsc-users <<a href="mailto:petsc-users@mcs.anl.gov" target="_blank">petsc-users@mcs.anl.gov</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 bgcolor="#FFFFFF"> <font size="-1" face="Arial,
                  sans-serif"><span style="font-size:10pt">Dear PETSc
                    developers/users,</span></font><br>
                <font size="-1" face="Arial, sans-serif"> </font><br>
                <font size="-1" face="Arial, sans-serif"><span style="font-size:10pt">I’m solving linear PDEs on a
                    regular grid with high-order finite differences,
                    assembling an MPIAIJ matrix to solve linear systems
                    or eigenvalue problems. I’ve been using vertex
                    major, natural ordering for the parallelism with
                    PetscSplitOwnership (yielding rectangular slices of
                    the physical domain) and wanted to move to DMDA to
                    have a more square-ish domain decomposition and
                    minimize communication between processes.</span></font><br>
                <font size="-1" face="Arial, sans-serif"><span style="font-size:10pt"></span></font><br>
                <font size="-1" face="Arial, sans-serif"><span style="font-size:10pt">However, my application is
                    memory critical, and I have finely-tuned matrix
                    preallocation routines for allocating memory
                    “optimally”. It seems the memory of a DMDA matrix is
                    allocated along the value of the stencil width of
                    DMDACreate and the manual says about it</span></font><br>
                <font size="-1" face="Arial, sans-serif"> </font><br>
                <font size="-1" face="Arial, sans-serif"><span style="font-size:10pt">“</span>These DMDA stencils
                  have nothing directly to do with any finite difference
                  stencils one might chose to use for a discretization”</font><br>
                <font size="-1" face="Arial, sans-serif"> </font><br>
                <font size="-1" face="Arial, sans-serif">And despite
                  reading the manual pages there must be something I do
                  not understand in the DM topology, what is that
                  "stencil width" for then? I will not use ghost values
                  for my FD-method, right?</font></div>
            </blockquote>
            <div><br>
            </div>
            <div>What this is saying is, "You might be using some
              stencil that is not STAR or BOX, but we are preallocating
              according to one of those".</div>
            <div>If you really care about how much memory is
              preallocated, which it seems you do, then you might be
              able to use</div>
            <div><br>
            </div>
            <div>  <a href="https://www.mcs.anl.gov/petsc/petsc-current/docs/manualpages/DMDA/DMDASetBlockFills.html" target="_blank">https://www.mcs.anl.gov/petsc/petsc-current/docs/manualpages/DMDA/DMDASetBlockFills.html</a></div>
            <div><br>
            </div>
            <div>to tell use exactly how to preallocate.</div>
            <div> </div>
            <blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex">
              <div bgcolor="#FFFFFF"> <font size="-1" face="Arial,
                  sans-serif">I was then wondering if I could just
                  create a MPIAIJ matrix, and with a PETSc routine get
                  the global indices of the domain for each process: in
                  other words, an equivalent of PetscSplitOwnership that
                  gives me the DMDA unknown ordering. So I can feed and
                  loop on that in my preallocation and assembly
                  routines.</font></div>
            </blockquote>
            <div><br>
            </div>
            <div>You can make an MPIAIJ matrix yourself of course. It
              should have the same division of rows as the DMDA division
              of dofs. Also, MatSetValuesStencil() will not work for a
              custom matrix.</div>
            <div><br>
            </div>
            <div>  Thanks,</div>
            <div><br>
            </div>
            <div>     Matt</div>
            <div> </div>
            <blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex">
              <div bgcolor="#FFFFFF"> <font size="-1" face="Arial,
                  sans-serif">T</font><font size="-1" face="Arial,
                  sans-serif">hanks very much,<span style="font-size:10pt"></span></font><br>
                <font size="-1" face="Arial, sans-serif"><span style="font-size:10pt"> </span></font><br>
                <font size="-1" face="Arial, sans-serif"> <span style="font-size:10pt">Thibaut</span></font> </div>
            </blockquote>
          </div>
          <br clear="all">
          <div><br>
          </div>
          -- <br>
          <div dir="ltr" class="gmail-m_-4123279978571850831gmail_signature">
            <div dir="ltr">
              <div>
                <div dir="ltr">
                  <div>
                    <div dir="ltr">
                      <div>What most experimenters take for granted
                        before they begin their experiments is
                        infinitely more interesting than any results to
                        which their experiments lead.<br>
                        -- Norbert Wiener</div>
                      <div><br>
                      </div>
                      <div><a href="http://www.cse.buffalo.edu/~knepley/" target="_blank">https://www.cse.buffalo.edu/~knepley/</a><br>
                      </div>
                    </div>
                  </div>
                </div>
              </div>
            </div>
          </div>
        </div>
      </div>
    </blockquote>
  </div>

</blockquote></div><br clear="all"><div><br></div>-- <br><div dir="ltr" class="gmail_signature"><div dir="ltr"><div><div dir="ltr"><div><div dir="ltr"><div>What most experimenters take for granted before they begin their experiments is infinitely more interesting than any results to which their experiments lead.<br>-- Norbert Wiener</div><div><br></div><div><a href="http://www.cse.buffalo.edu/~knepley/" target="_blank">https://www.cse.buffalo.edu/~knepley/</a><br></div></div></div></div></div></div></div></div></div>