<html>
  <head>
    <meta content="text/html; charset=utf-8" http-equiv="Content-Type">
  </head>
  <body bgcolor="#FFFFFF" text="#000000">
    +1 to the general idea of a target dependent hook for describing
    aliasing of address spaces.  I have no opinion on the particular
    implementation strategy proposed.  <br>
    <br>
    <div class="moz-cite-prefix">On 08/07/2015 11:35 AM, Jingyue Wu via
      llvm-dev wrote:<br>
    </div>
    <blockquote
cite="mid:CAMROOrEU8_BKcqW3-VudTL=gmF3ux2+aoC-b_+3__0PRh=jLtg@mail.gmail.com"
      type="cite">
      <div dir="ltr">+ the new llvm-dev</div>
      <div class="gmail_extra"><br>
        <div class="gmail_quote">On Fri, Aug 7, 2015 at 11:30 AM,
          Jingyue Wu <span dir="ltr"><<a moz-do-not-send="true"
              href="mailto:jingyue@google.com" target="_blank">jingyue@google.com</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">
              <div>Hi folks,</div>
              <div><br>
              </div>
              <div>Unsurprisingly, leveraging the fact that certain
                address spaces don't alias can significantly improve
                alias analysis precision and enhance (observably 2x
                performance gain) load/store optimizations such as LICM
                and DSE. </div>
              <div><br>
              </div>
              <div>This sounds to me an overdue feature. I saw several
                discussion threads on that direction, but none of them
                really happened. </div>
              <div><br>
              </div>
              (1) <a moz-do-not-send="true"
href="http://lists.cs.uiuc.edu/pipermail/llvm-commits/Week-of-Mon-20111010/129615.html"
                target="_blank">http://lists.cs.uiuc.edu/pipermail/llvm-commits/Week-of-Mon-20111010/129615.html</a>.
              Justin Holewinski proposed to add an address-space alias
              analysis that treats pointers in different address spaces
              not aliasing. This patch got shot down because, in some
              targets, address spaces may alias. For example, in
              CUDA+NVPTX, addrspace(0) aliases everyone. 
              <div>
                <div><br>
                </div>
                <div>(2) <a moz-do-not-send="true"
href="http://lists.cs.uiuc.edu/pipermail/llvmdev/2013-August/064620.html"
                    target="_blank">http://lists.cs.uiuc.edu/pipermail/llvmdev/2013-August/064620.html</a>.
                  Michele Scandale proposed address space extensions in
                  IR metadata which TBAA could then leverage to prove
                  non-aliasing. This approach didn't fly either because
                  address spaces are target-specific. The front end
                  doesn't know enough to decide aliasing. <br>
                </div>
                <br>
                So, can we make BasicAA to consider address spaces?
                Specifically, I am proposing:</div>
              <div>a) adding a new interface like
                TTI::addressSpacesAlias(unsigned, unsigned), and</div>
              <div>b) adding a little piece of logic in BasicAA that
                reports "no alias" if address spaces don't alias. </div>
              <div><br>
              </div>
              <div>This approach addresses the issue brought up in (2)
                because TTI can see the entire codegen. It also resolves
                the issue that shut down (1) because it allows address
                spaces to alias in a target-defined way. Actually, John
                Criswell did mention in that thread the idea of
                embedding alias info in TargetData. Now that we have
                TTI, it seems a better place to hold target-specific
                alias info than DataLayout. </div>
              <div><br>
              </div>
              <div>Any comments? </div>
              <span class="HOEnZb"><font color="#888888">
                  <div><br>
                  </div>
                  <div>Jingyue</div>
                </font></span></div>
          </blockquote>
        </div>
        <br>
      </div>
      <br>
      <fieldset class="mimeAttachmentHeader"></fieldset>
      <br>
      <pre wrap="">_______________________________________________
LLVM Developers mailing list
<a class="moz-txt-link-abbreviated" href="mailto:llvm-dev@lists.llvm.org">llvm-dev@lists.llvm.org</a>         <a class="moz-txt-link-freetext" href="http://llvm.cs.uiuc.edu">http://llvm.cs.uiuc.edu</a>
<a class="moz-txt-link-freetext" href="http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev">http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev</a>
</pre>
    </blockquote>
    <br>
  </body>
</html>