<html>
  <head>
    <meta http-equiv="Content-Type" content="text/html; charset=utf-8">
  </head>
  <body text="#000000" bgcolor="#FFFFFF">
    <div class="moz-cite-prefix">See
      <a class="moz-txt-link-freetext" href="http://lists.llvm.org/pipermail/llvm-dev/2018-July/124792.html">http://lists.llvm.org/pipermail/llvm-dev/2018-July/124792.html</a> ?<br>
      <br>
      -Eli<br>
      <br>
      On 9/20/2018 3:40 PM, hameeza ahmed wrote:<br>
    </div>
    <blockquote type="cite"
cite="mid:CAFMPKeZX=pw2+KaRLoYCcLuHcAQyWrJ6sdJb3P0ZZ0DT+Ey0Kg@mail.gmail.com">
      <meta http-equiv="content-type" content="text/html; charset=utf-8">
      Thank you.
      <div><br>
      </div>
      <div>I am working on a machine with greater vector widths. How to
        enable the emission of greater and different vector widths in
        loop codes through pragma ? and automatically.<br>
        <br>
        On Friday, September 21, 2018, Friedman, Eli <<a
          href="mailto:efriedma@codeaurora.org" moz-do-not-send="true">efriedma@codeaurora.org</a>>
        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>On 9/20/2018 2:15 PM, hameeza ahmed wrote:<br>
            </div>
            <blockquote type="cite">
              <div dir="ltr">
                <div dir="ltr">
                  <div dir="ltr">
                    <div dir="ltr">
                      <div>Hello,</div>
                      <div>I m trying to set vector width using #pragma
                        clang loop vectorize_width(32) but i m getting
                        width 8 for the following kernel;</div>
                      <br>
                      <div><b>i m getting following output when i
                          compiled;</b></div>
                      <div><b><br>
                        </b></div>
                      <div><b>clang -O3  correlation.c  
                          -Rpass=loop-vectorize  -emit-llvm
                          -march=knl    -S  -o 1.ll<br>
                          correlation.c:38:9: remark: vectorized loop
                          (vectorization width: 8, interleaved count: 4)
                          [-Rpass=loop-vectorize]<br>
                                  for (j = 0; j < M; j++)<br>
                                  ^<br>
                        </b></div>
                    </div>
                  </div>
                </div>
              </div>
            </blockquote>
            <br>
            With AVX-512, an instruction can operate on at most 8
            double-precision lanes.  The vectorizer recognizes that, and
            interleaves the loop so you get 8*4==32 scalar iterations
            per iteration of the vectorized loop.<br>
            <br>
            -Eli<br>
            <pre cols="72">-- 
Employee of Qualcomm Innovation Center, Inc.
Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum, a Linux Foundation Collaborative Project</pre>
          </div>
        </blockquote>
      </div>
      -- <br>
      You received this message because you are subscribed to the Google
      Groups "Polly Development" group.<br>
      To unsubscribe from this group and stop receiving emails from it,
      send an email to <a
        href="mailto:polly-dev+unsubscribe@googlegroups.com"
        moz-do-not-send="true">polly-dev+unsubscribe@googlegroups.com</a>.<br>
      For more options, visit <a
        href="https://groups.google.com/d/optout" moz-do-not-send="true">https://groups.google.com/d/optout</a>.<br>
    </blockquote>
    <p><br>
    </p>
    <pre class="moz-signature" cols="72">-- 
Employee of Qualcomm Innovation Center, Inc.
Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum, a Linux Foundation Collaborative Project</pre>
  </body>
</html>