<div dir="ltr"><br><div class="gmail_extra"><br><div class="gmail_quote">On Wed, Oct 25, 2017 at 10:19 AM, Hal Finkel <span dir="ltr"><<a href="mailto:hfinkel@anl.gov" target="_blank">hfinkel@anl.gov</a>></span> wrote:<br><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">
  
    
  
  <div bgcolor="#FFFFFF" text="#000000"><span class="">
    <p><br>
    </p>
    <div class="m_3151644750661815301moz-cite-prefix">On 10/25/2017 12:16 PM, Xinliang David
      Li via llvm-dev wrote:<br>
    </div>
    <blockquote type="cite">
      
      <div dir="ltr"><br>
        <div class="gmail_extra"><br>
          <div class="gmail_quote">On Tue, Oct 17, 2017 at 11:50 PM,
            Chandler Carruth via llvm-dev <span dir="ltr"><<a href="mailto:llvm-dev@lists.llvm.org" target="_blank">llvm-dev@lists.llvm.org</a>></span>
            wrote:<br>
            <blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">
              <div dir="ltr">Greetings everyone!
                <div><br>
                </div>
                <div>The new pass manager is getting extremely close to
                  the point where I'm not aware of any significant
                  outstanding work needed, and I'd like to see what else
                  would be needed to enable it by default. Here are the
                  current functionality I'm aware of outstanding:</div>
                <div><br>
                </div>
                <div>1) Does not do non-trivial loop unswitching.
                  Majority of this is in <a href="https://reviews.llvm.org/D34200" target="_blank">https://reviews.llvm.org/D3420<wbr>0</a> but
                  will need one or two small follow-ups.</div>
                <div><br>
                </div>
                <div>2) Currently, sanitizers don't work correctly with
                  it. Thanks to the work of others, the missing
                  infrastructure has been added and I'll send a patch to
                  wire this up this week.</div>
                <div><br>
                </div>
                <div>3) Missing support for 'optnone'. I've been working
                  on this, but the existing testing wasn't as thorough
                  as I wanted, so it is going slowly. I've got about 1/4
                  of this implemented and should have patches this week
                  or next.</div>
                <div><br>
                </div>
                <div>4) Missing opt-bisect (or similar) facility. This
                  looks pretty trivial to add, but I've not even
                  started. If anyone is interested in it, go for it. We
                  might even be able to do something simpler using the
                  generic debug counters and get equivalent
                  functionality.</div>
                <div><br>
                </div>
                <div>... that's it?</div>
              </div>
            </blockquote>
            <div><br>
            </div>
            <div><br>
            </div>
            <div>Missing support of 'print-after-all' or
              'print-after-passX'  is another major usability loss.</div>
            <div><br>
            </div>
            <div>Regarding the default switch, maybe do it in two
              steps: </div>
            <div><br>
            </div>
            <div>1) Switch the default for PGO build first</div>
            <div>2) Switch the default for all modes.</div>
          </div>
        </div>
      </div>
    </blockquote>
    <br></span>
    Why?<span class="HOEnZb"><font color="#888888"><br>
    <br></font></span></div></blockquote><div><br></div><div>1) PGO users benefit from the PM change the most</div><div>2) I suppose incremental changes are always better and causes fewer churns?</div><div><br></div><div>David</div><div> </div><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"><div bgcolor="#FFFFFF" text="#000000"><span class="HOEnZb"><font color="#888888">
     -Hal</font></span><span class=""><br>
    <br>
    <blockquote type="cite">
      <div dir="ltr">
        <div class="gmail_extra">
          <div class="gmail_quote">
            <div><br>
            </div>
            <div>David</div>
            <div><br>
            </div>
            <div><br>
            </div>
            <div> </div>
            <blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">
              <div dir="ltr">
                <div><br>
                </div>
                <div>Optimization quality / run-time performance:</div>
                <div>- We've been using it at Google extensively and are
                  very happy with the optimization quality. Benchmarks
                  look *very* good here.</div>
                <div>- More data from other users would be important.</div>
                <div>- You can try it out with `-fexperimental-new-pass-manag<wbr>er`
                  to Clang</div>
                <div><br>
                </div>
                <div>Compile-time performance:</div>
                <div>- Sometimes *much* better due to cached analyses.</div>
                <div>- Sometimes worse, typically due to more /
                  different inlining in turn running main pipeline
                  (GVN + InstCombine) more times or over more code.</div>
                <div>- Overall somewhat a wash, but the increased
                  compile times typically due to the optimizer "trying"
                  harder, so not too concerning on our end.</div>
                <div>- Again, more feedback from other users good:
                  `-fexperimental-new-pass-manag<wbr>er` to Clang</div>
                <div><br>
                </div>
                <div>Once the four missing things land, I'll also
                  happily work on collecting some of the basics on the
                  test-suite and CTMark. But I suspect more "in the
                  wild" data would really be useful here given the
                  significance of the change.</div>
                <div><br>
                </div>
                <div>Thoughts? What else (beyond the four items above
                  and feedback on run-time and compile-time) would folks
                  like to see?</div>
                <div><br>
                </div>
                <div>Once this happens, I'll also be preparing some
                  batch, mechanical updates to the test suite to
                  primarily use the new pass manager. Also there is lots
                  of documentation updates that will be needed here.</div>
                <span class="m_3151644750661815301HOEnZb"><font color="#888888">
                    <div><br>
                    </div>
                    <div>-Chandler</div>
                  </font></span>
                <div><br>
                </div>
                <div>PS: I'll be sending a note to cfe-dev as a "heads
                  up" about this discussion as in some ways, the default
                  flip is mostly a Clang default flip. But hopefully our
                  doc updates will trigger this being "perceived" as the
                  default for other frontends, and I'll try to reach out
                  to other major frontends as well (Swift and Rust are
                  on my radar, and I've already started talking with
                  Philip Reames about their Falcon JIT).</div>
              </div>
              <br>
              ______________________________<wbr>_________________<br>
              LLVM Developers mailing list<br>
              <a href="mailto:llvm-dev@lists.llvm.org" target="_blank">llvm-dev@lists.llvm.org</a><br>
              <a href="http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev" rel="noreferrer" target="_blank">http://lists.llvm.org/cgi-bin/<wbr>mailman/listinfo/llvm-dev</a><br>
              <br>
            </blockquote>
          </div>
          <br>
        </div>
      </div>
      <br>
      <fieldset class="m_3151644750661815301mimeAttachmentHeader"></fieldset>
      <br>
      <pre>______________________________<wbr>_________________
LLVM Developers mailing list
<a class="m_3151644750661815301moz-txt-link-abbreviated" href="mailto:llvm-dev@lists.llvm.org" target="_blank">llvm-dev@lists.llvm.org</a>
<a class="m_3151644750661815301moz-txt-link-freetext" href="http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev" target="_blank">http://lists.llvm.org/cgi-bin/<wbr>mailman/listinfo/llvm-dev</a>
</pre>
    </blockquote>
    <br>
    </span><span class=""><pre class="m_3151644750661815301moz-signature" cols="72">-- 
Hal Finkel
Lead, Compiler Technology and Programming Languages
Leadership Computing Facility
Argonne National Laboratory</pre>
  </span></div>

</blockquote></div><br></div></div>