<html>
  <head>
    <meta content="text/html; charset=utf-8" http-equiv="Content-Type">
  </head>
  <body bgcolor="#FFFFFF" text="#000000">
    <p><br>
    </p>
    <div class="moz-cite-prefix">On 02/27/2017 11:47 AM, Adam Nemet
      wrote:<br>
    </div>
    <blockquote
      cite="mid:1808DF46-BCA7-4F08-B707-8996564A9C76@apple.com"
      type="cite">
      <meta http-equiv="Content-Type" content="text/html; charset=utf-8">
      <br class="">
      <div>
        <blockquote type="cite" class="">
          <div class="">On Feb 27, 2017, at 9:39 AM, Daniel Berlin <<a
              moz-do-not-send="true" href="mailto:dberlin@dberlin.org"
              class="">dberlin@dberlin.org</a>> wrote:</div>
          <br class="Apple-interchange-newline">
          <div class="">
            <div dir="ltr" class=""><br class="">
              <div class="gmail_extra"><br class="">
                <div class="gmail_quote">On Mon, Feb 27, 2017 at 9:29
                  AM, Adam Nemet <span dir="ltr" class=""><<a
                      moz-do-not-send="true"
                      href="mailto:anemet@apple.com" target="_blank"
                      class="">anemet@apple.com</a>></span> wrote:<br
                    class="">
                  <blockquote class="gmail_quote" style="margin:0px 0px
                    0px 0.8ex;border-left:1px solid
                    rgb(204,204,204);padding-left:1ex">
                    <div style="word-wrap:break-word" class=""><br
                        class="">
                      <div class=""><span class="gmail-">
                          <blockquote type="cite" class="">
                            <div class="">On Feb 27, 2017, at 7:27 AM,
                              Hal Finkel <<a moz-do-not-send="true"
                                href="mailto:hfinkel@anl.gov"
                                target="_blank" class="">hfinkel@anl.gov</a>>
                              wrote:</div>
                            <br
                              class="gmail-m_8239107560126363242Apple-interchange-newline">
                            <div class="">
                              <div
                                class="gmail-m_8239107560126363242moz-cite-prefix"
style="font-family:helvetica;font-size:10px;font-style:normal;font-variant-caps:normal;font-weight:normal;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;background-color:rgb(255,255,255)"><br
class="gmail-m_8239107560126363242Apple-interchange-newline">
                                On 02/27/2017 06:29 AM, Nema, Ashutosh
                                wrote:<br class="">
                              </div>
                              <blockquote type="cite"
style="font-family:helvetica;font-size:10px;font-style:normal;font-variant-caps:normal;font-weight:normal;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;background-color:rgb(255,255,255)"
                                class="">
                                <div
                                  class="gmail-m_8239107560126363242WordSection1">
                                  <div style="margin:0in 0in
                                    0.0001pt;font-size:12pt;font-family:"times
                                    new roman",serif" class="">Thanks
                                    for looking into this.</div>
                                  <div style="margin:0in 0in
                                    0.0001pt;font-size:12pt;font-family:"times
                                    new roman",serif" class=""> </div>
                                  <div style="margin:0in 0in
                                    0.0001pt;font-size:12pt;font-family:"times
                                    new roman",serif" class="">1)
                                    Issues with re running vectorizer:</div>
                                  <div style="margin:0in 0in
                                    0.0001pt;font-size:12pt;font-family:"times
                                    new roman",serif" class="">Vectorizer
                                    might generate redundant alias
                                    checks while vectorizing epilog
                                    loop.</div>
                                  <div style="margin:0in 0in
                                    0.0001pt;font-size:12pt;font-family:"times
                                    new roman",serif" class="">Redundant
                                    alias checks are expensive, we like
                                    to reuse the results of already
                                    computed alias checks.</div>
                                  <div style="margin:0in 0in
                                    0.0001pt;font-size:12pt;font-family:"times
                                    new roman",serif" class="">With
                                    metadata we can limit the width of
                                    epilog loop, but not sure about
                                    reusing alias check result.</div>
                                  <div style="margin:0in 0in
                                    0.0001pt;font-size:12pt;font-family:"times
                                    new roman",serif" class="">Any
                                    thoughts on rerunning vectorizer
                                    with reusing the alias check result
                                    ?</div>
                                </div>
                              </blockquote>
                              <br
style="font-family:helvetica;font-size:10px;font-style:normal;font-variant-caps:normal;font-weight:normal;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;background-color:rgb(255,255,255)"
                                class="">
                              <span
style="font-family:helvetica;font-size:10px;font-style:normal;font-variant-caps:normal;font-weight:normal;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;background-color:rgb(255,255,255);float:none;display:inline"
                                class="">One way of looking at this is:
                                Reusing the alias-check result is really
                                just a conditional propagation problem;
                                if we don't already have an optimization
                                that can combine these after the fact,
                                then we should.</span><br
style="font-family:helvetica;font-size:10px;font-style:normal;font-variant-caps:normal;font-weight:normal;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;background-color:rgb(255,255,255)"
                                class="">
                            </div>
                          </blockquote>
                          <div class=""><br class="">
                          </div>
                        </span>
                        <div class="">+Danny</div>
                        <div class=""><br class="">
                        </div>
                        <div class="">Isn’t Extended SSA supposed to
                          help with this?</div>
                      </div>
                    </div>
                  </blockquote>
                  <div class=""><br class="">
                  </div>
                  <div class="">Yes, it will solve this with no issue
                    already.  GVN probably does already too.</div>
                  <div class=""><br class="">
                  </div>
                  <div class="">even if if you have</div>
                  <div class=""><br class="">
                  </div>
                  <div class="">if (a == b)</div>
                  <div class="">
                    <div class="">if (a == c)</div>
                  </div>
                  <div class="">
                    <div class=""> if (a == d)</div>
                  </div>
                  <div class="">
                    <div class=""> if (a == e)</div>
                  </div>
                  <div class="">
                    <div class=""> if (a == g)</div>
                  </div>
                  <div class=""><br class="">
                  </div>
                  <div class=""><br class="">
                  </div>
                  <div class="">and  we can prove a ... g equivalent,
                    newgvn will eliminate them all and set all the
                    branches true.</div>
                  <div class=""><br class="">
                  </div>
                  <div class="">If you need a simpler clean up pass, we
                    could run it on sub-graphs.</div>
                </div>
              </div>
            </div>
          </div>
        </blockquote>
        <div><br class="">
        </div>
        <div>Yes we probably don’t want to run a full GVN after the
          “loop-scheduling” passes.</div>
      </div>
    </blockquote>
    <br>
    FWIW, we could, just without the memory-dependence analysis enabled
    (i.e. set the NoLoads constructor parameter to true). GVN is pretty
    fast in that mode.<br>
    <br>
     -Hal<br>
    <br>
    <blockquote
      cite="mid:1808DF46-BCA7-4F08-B707-8996564A9C76@apple.com"
      type="cite">
      <div>
        <div><br class="">
        </div>
        <div>I guess the pipeline to experiment with for now is opt
          -loop-vectorize -loop-vectorize -newgvn.</div>
        <div><br class="">
        </div>
        <div>Adam</div>
        <br class="">
        <blockquote type="cite" class="">
          <div class="">
            <div dir="ltr" class="">
              <div class="gmail_extra">
                <div class="gmail_quote">
                  <div class="">The only thing you'd have to do is write
                    some code to set "live on entry" subgraph variables
                    in their own congruence classes.</div>
                  <div class="">We already do this for incoming
                    arguments.</div>
                  <div class=""><br class="">
                  </div>
                  <div class="">Otherwise, it's trivial to make it only
                    walk things in the subgraph.</div>
                  <div class=""><br class="">
                  </div>
                  <div class=""><br class="">
                  </div>
                </div>
              </div>
            </div>
          </div>
        </blockquote>
      </div>
      <br class="">
    </blockquote>
    <br>
    <pre class="moz-signature" cols="72">-- 
Hal Finkel
Lead, Compiler Technology and Programming Languages
Leadership Computing Facility
Argonne National Laboratory</pre>
  </body>
</html>