<html>
  <head>
    <meta content="text/html; charset=windows-1252"
      http-equiv="Content-Type">
  </head>
  <body bgcolor="#FFFFFF" text="#000000">
    Am 8/16/17 um 5:38 PM schrieb Stephan Hoyer:<br>
    <blockquote
cite="mid:CAEQ_Tve-fSKsZg5PPfWFiz058e5_uH-mfdAXHuwXupKu0fY-Lg@mail.gmail.com"
      type="cite">
      <div dir="ltr">
        <div class="gmail_extra">
          <div class="gmail_quote">On Wed, Aug 16, 2017 at 2:39 AM, Paul
            Springer <span dir="ltr"><<a moz-do-not-send="true"
                href="mailto:pavdev@gmx.de" target="_blank">pavdev@gmx.de</a>></span>
            wrote:<br>
            <blockquote class="gmail_quote" style="margin:0px 0px 0px
              0.8ex;border-left:1px solid
              rgb(204,204,204);padding-left:1ex">
              <div bgcolor="#FFFFFF"><span class="gmail-"> <br>
                  <blockquote type="cite">
                    <div dir="ltr">
                      <div class="gmail_extra">
                        <div class="gmail_quote">
                          <div>What version of Numpy are you comparing
                            to? Note that in 1.13 you can enable some
                            optimization in einsum, and the coming 1.14
                            makes that the default and uses CBLAS when
                            possible.<br>
                          </div>
                        </div>
                      </div>
                    </div>
                  </blockquote>
                </span> I was using 1.10.4; however, I am currently
                running the benchmark with 1.13.1 and 'optimize=True';
                this, however, seems to yield even worse performance
                (see attached).<br>
                If you are interested, you can check the performance
                difference yourself via: ./benchmark/python/bechmark.sh</div>
            </blockquote>
            <div><br>
            </div>
            <div>This sounds like you may be using relatively small
              matrices, where the overhead of calculating the optimal
              strategy dominates. Can you try with a few bigger test
              cases?</div>
          </div>
        </div>
      </div>
      <br>
    </blockquote>
    The sizes of the tensors varies form ~5MB up to ~100MB towards the
    far right of the plot; this corresponds to matrices of size ~1000^2
    to ~5000^2, thus the sizes should be large enough to amortize any
    overhead associated to calculating the optimal strategy.<br>
    <br>
  </body>
</html>