<div dir="ltr"><div class="gmail_quote"><div dir="ltr" class="gmail_attr">On Wed, Jan 16, 2019 at 9:43 AM Christian König <<a href="mailto:ckoenig.leichtzumerken@gmail.com">ckoenig.leichtzumerken@gmail.com</a>> wrote:<br></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">
    <div class="gmail-m_1079118432294564707moz-cite-prefix">Am 16.01.19 um 15:39 schrieb Marek
      Olšák:<br>
    </div>
    <blockquote type="cite">
      
      <div dir="auto">
        <div><br>
          <br>
          <div class="gmail_quote">
            <div dir="ltr">On Wed, Jan 16, 2019, 9:34 AM Koenig,
              Christian <<a href="mailto:Christian.Koenig@amd.com" target="_blank">Christian.Koenig@amd.com</a>
              wrote:<br>
            </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">
                <div class="gmail-m_1079118432294564707m_3799249069402976776moz-cite-prefix">Am
                  16.01.19 um 15:31 schrieb Marek Olšák:<br>
                </div>
                <blockquote type="cite">
                  <div dir="auto">
                    <div><br>
                      <br>
                      <div class="gmail_quote">
                        <div dir="ltr">On Wed, Jan 16, 2019, 7:55 AM
                          Christian König <<a href="mailto:ckoenig.leichtzumerken@gmail.com" rel="noreferrer" target="_blank">ckoenig.leichtzumerken@gmail.com</a>
                          wrote:<br>
                        </div>
                        <blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex">
                          Well if you ask me we should have the
                          following interface for <br>
                          negotiating memory management with the kernel:<br>
                          <br>
                          1. We have per process BOs which can't be
                          shared between processes.<br>
                          <br>
                          Those are always valid and don't need to be
                          mentioned in any BO list <br>
                          whatsoever.<br>
                          <br>
                          If we knew that a per process BO is currently
                          not in use we can <br>
                          optionally tell that to the kernel to make
                          memory management more efficient.<br>
                          <br>
                          In other words instead of a list of stuff
                          which is used we send down to <br>
                          the kernel a list of stuff which is not used
                          any more and that only when <br>
                          we know that it is necessary, e.g. when a game
                          or application overcommits.<br>
                        </blockquote>
                      </div>
                    </div>
                    <div dir="auto"><br>
                    </div>
                    <div dir="auto">Radeonsi doesn't use this because
                      this approach caused performance degradation and
                      also drops BO priorities.</div>
                  </div>
                </blockquote>
                <br>
                The performance degradation where mostly shortcomings
                with the LRU which by now have been fixed.<br>
                <br>
                BO priorities are a different topic, but could be added
                to per VM BOs as well.<br>
              </div>
            </blockquote>
          </div>
        </div>
        <div dir="auto"><br>
        </div>
        <div dir="auto">What's the minimum drm version that contains the
          fixes?</div>
      </div>
    </blockquote>
    <br>
    I've pushed the last optimization this morning. No idea when it
    really became useful, but the numbers from the closed source clients
    now look much better.<br>
    <br>
    We should probably test and bump the drm version when we are sure
    that this now works as expected.<br></div></blockquote><div><br></div><div>We should, but AMD Mesa guys don't have any time.</div><div><br></div><div>Marek</div></div></div>