<html>
  <head>
    <meta http-equiv="Content-Type" content="text/html; charset=utf-8">
  </head>
  <body text="#000000" bgcolor="#FFFFFF">
    <div class="moz-cite-prefix">I don't think that is a good idea.<br>
      <br>
      Ideally GTT should now have the same performance as VRAM on APUs
      and we should use VRAM only for things where we absolutely have to
      and to actually use up the otherwise unused VRAM.<br>
      <br>
      Can you run some tests with all BOs forced to GTT and see if there
      is any performance regression?<br>
      <br>
      Christian.<br>
      <br>
      Am 20.03.2018 um 15:51 schrieb Marek Olšák:<br>
    </div>
    <blockquote type="cite"
cite="mid:CAAxE2A7ffe2fkOQG0mFpPfP4uRKZ1BVTKWQ3Q+FhLWQnYU4nww@mail.gmail.com">
      <div dir="ltr">
        <div class="gmail_extra">
          <div class="gmail_quote">On Tue, Mar 20, 2018 at 9:55 AM,
            Christian König <span dir="ltr"><<a
                href="mailto:ckoenig.leichtzumerken@gmail.com"
                target="_blank" moz-do-not-send="true">ckoenig.leichtzumerken@gmail.<wbr>com</a>></span>
            wrote:<br>
            <blockquote class="gmail_quote" style="margin:0 0 0
              .8ex;border-left:1px #ccc solid;padding-left:1ex">
              <div text="#000000" bgcolor="#FFFFFF">
                <div
                  class="m_-3838486020630618689m_8572536124089128592moz-cite-prefix">Yes,
                  exactly. And if I remember correctly Mesa used to
                  always set GTT as fallback on APUs, correct?<br>
                </div>
              </div>
            </blockquote>
            <div><br>
            </div>
            <div>"used to" is the key part. Mesa doesn't force GTT on
              APUs anymore. It expects that the combination of BO
              priorities and BO move throttling will result in optimal
              BO placements over time.<br>
              <br>
            </div>
            <div>Marek<br>
            </div>
            <div> </div>
            <blockquote class="gmail_quote" style="margin:0 0 0
              .8ex;border-left:1px #ccc solid;padding-left:1ex">
              <div text="#000000" bgcolor="#FFFFFF">
                <div
                  class="m_-3838486020630618689m_8572536124089128592moz-cite-prefix">
                  <br>
                  The problem seems to be that this fallback isn't set
                  for displayable BOs.<br>
                  <br>
                  So what needs to be done is to just enable this
                  fallback for displayable BOs as well if the kernel can
                  handle it.<span class="m_-3838486020630618689HOEnZb"><font
                      color="#888888"><br>
                      <br>
                      Christian.</font></span>
                  <div>
                    <div class="m_-3838486020630618689h5"><br>
                      <br>
                      Am 20.03.2018 um 00:01 schrieb Marek Olšák:<br>
                    </div>
                  </div>
                </div>
                <blockquote type="cite">
                  <div>
                    <div class="m_-3838486020630618689h5">
                      <div dir="ltr">
                        <div class="gmail_extra">
                          <div class="gmail_quote">In theory, Mesa
                            doesn't have to do anything. It can continue
                            setting VRAM and if the kernel has to put a
                            display buffer into GTT, it doesn't matter
                            (for Mesa). Whether the VRAM placement is
                            really used is largely determined by BO
                            priorities.<br>
                          </div>
                          <div class="gmail_quote"><br>
                          </div>
                          <div class="gmail_quote">The way I understand
                            scather/gather is that it only allows the
                            GTT placement. It doesn't force the GTT
                            placement. Mesa also doesn't force the GTT
                            placement.<br>
                          </div>
                          <div class="gmail_quote"><br>
                          </div>
                          <div class="gmail_quote">Marek<br>
                          </div>
                          <div class="gmail_quote"><br>
                          </div>
                          <div class="gmail_quote">On Mon, Mar 19, 2018
                            at 5:12 PM, Alex Deucher <span dir="ltr"><<a
                                href="mailto:alexdeucher@gmail.com"
                                target="_blank" moz-do-not-send="true">alexdeucher@gmail.com</a>></span>
                            wrote:<br>
                            <blockquote class="gmail_quote"
                              style="margin:0 0 0 .8ex;border-left:1px
                              #ccc solid;padding-left:1ex"><span>On Mon,
                                Mar 19, 2018 at 4:29 PM, Li, Samuel <<a
                                  href="mailto:Samuel.Li@amd.com"
                                  target="_blank" moz-do-not-send="true">Samuel.Li@amd.com</a>>
                                wrote:<br>
                                >>to my earlier point, there may
                                be cases where it is advantageous to put<br>
                                >> display buffers in vram even if
                                s/g display is supported<br>
                                ><br>
                                > Agreed. That is also why the patch
                                has the options to let user select where<br>
                                > to put display buffers.<br>
                                ><br>
                                > As whether to put the option in
                                Mesa or kernel, it seems the difference
                                is<br>
                                > not much. Also, since amdgpufb can
                                request even without mesa, kernel might<br>
                                > be a better choice. In addition,
                                putting in the kernel can save client’s<br>
                                > duplicate work(mesa, ogl, vulkan,
                                2d, kernel…)<br>
                                <br>
                              </span>Why do we even expose different
                              memory pools to the UMDs in the first<br>
                              place ;)  Each pool has performance
                              characteristics that may be<br>
                              relevant for a particular work load.  Only
                              the UMDs really know the<br>
                              finer points of those workloads. In
                              general, you don't want the kernel<br>
                              dictating policy if you can avoid it.  The
                              kernel exposes<br>
                              functionality and userspace sets the
                              policy.  With the location set in<br>
                              userspace, each app/user can have whatever
                              policy makes sense for<br>
                              their use case all at the same time
                              without needing to tweak their<br>
                              kernel for every use case.<br>
                              <span
class="m_-3838486020630618689m_8572536124089128592m_4947926278827078055HOEnZb"><font
                                  color="#888888"><br>
                                  Alex</font></span><br>
                            </blockquote>
                          </div>
                        </div>
                      </div>
                      <br>
                      <fieldset
                        class="m_-3838486020630618689m_8572536124089128592mimeAttachmentHeader"></fieldset>
                      <br>
                    </div>
                  </div>
                  <span>
                    <pre>______________________________<wbr>_________________
amd-gfx mailing list
<a class="m_-3838486020630618689m_8572536124089128592moz-txt-link-abbreviated" href="mailto:amd-gfx@lists.freedesktop.org" target="_blank" moz-do-not-send="true">amd-gfx@lists.freedesktop.org</a>
<a class="m_-3838486020630618689m_8572536124089128592moz-txt-link-freetext" href="https://lists.freedesktop.org/mailman/listinfo/amd-gfx" target="_blank" moz-do-not-send="true">https://lists.freedesktop.org/<wbr>mailman/listinfo/amd-gfx</a>
</pre>
                  </span></blockquote>
                <br>
              </div>
            </blockquote>
          </div>
          <br>
        </div>
      </div>
      <br>
      <fieldset class="mimeAttachmentHeader"></fieldset>
      <br>
      <pre wrap="">_______________________________________________
amd-gfx mailing list
<a class="moz-txt-link-abbreviated" href="mailto:amd-gfx@lists.freedesktop.org">amd-gfx@lists.freedesktop.org</a>
<a class="moz-txt-link-freetext" href="https://lists.freedesktop.org/mailman/listinfo/amd-gfx">https://lists.freedesktop.org/mailman/listinfo/amd-gfx</a>
</pre>
    </blockquote>
    <br>
  </body>
</html>