<div dir="ltr">Hi Walter<div><br><div> When I had DiFX running on early versions of Xeon Phi (2012?) I had to by-pass the IPP libraries (at the time Intel said they were not keen to carry those forwards). So all functionality was included in GENERIC. Some things had to be inline functions, but I could use the MKL libraries for others. <div style="color:rgb(34,34,34);font-family:arial,sans-serif;font-size:small;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;background-color:rgb(255,255,255);text-decoration-style:initial;text-decoration-color:initial">Has IPP been ported now? </div><br class="gmail-Apple-interchange-newline"></div><div><br></div><div> If not (and maybe if so) it could be worth looking to see where the speed up would be (it was dominated by the complex multiply accumulate at the time) and see if that could be improved. For example the intrinsic for <span style="color:rgb(0,0,0);white-space:pre-wrap">genericAddProduct_32fc could be replaced with the code example in:</span></div><div><div class="gmail-forum-post-wrapper1 gmail-D8CommentContainer gmail-row-with-author-and-content" style="width:841px;border:none;height:auto;overflow:hidden;padding:0px;margin:0px;float:left;display:table-row;color:rgb(83,87,94);font-family:Arial,宋体,Tahoma,Helvetica,sans-serif;font-size:13.008px;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;background-color:rgb(255,255,255);text-decoration-style:initial;text-decoration-color:initial"><div class="gmail-forum-post-panel-main1 gmail-clearfix gmail-D8Comment gmail-content-pane" style="zoom:1;float:none;height:auto;margin-left:0px;padding:5px 10px;color:rgb(96,96,96);box-sizing:border-box;width:780px;max-width:760px;display:table-cell"><div class="gmail-forum-post-content" style="min-height:80px"><div class="gmail-field gmail-field-name-comment-body gmail-field-type-text-long gmail-field-label-hidden"><div class="gmail-field-items"><div class="gmail-field-item even"><p style="word-wrap:break-word;margin:0px 0px 1.5em;padding:0px;font-family:Arial,宋体,Tahoma,Helvetica,sans-serif;font-size:12px;font-weight:normal;color:rgb(83,87,94)">In <a href="https://software.intel.com/en-us/forums/intel-isa-extensions/topic/74737">https://software.intel.com/en-us/forums/intel-isa-extensions/topic/74737</a></p><p style="word-wrap:break-word;margin:0px 0px 1.5em;padding:0px;font-family:Arial,宋体,Tahoma,Helvetica,sans-serif;font-size:12px;font-weight:normal;color:rgb(83,87,94)"> The comments include:</p><p style="word-wrap:break-word;margin:0px 0px 1.5em;padding:0px;font-family:Arial,宋体,Tahoma,Helvetica,sans-serif;font-size:12px;font-weight:normal;color:rgb(83,87,94)">SSE3 (included in SSE4.2) has specific instructions which the compiler will use to vectorize complex without requiring intrinsics or shuffle.  In order to use fully AVX2 or AVX512, as John said, it will be necessary to split the data. Compilers will not attempt to evaluate whether a mixture of SSE3 and AVX2 will prove faster.  The statistics produced by opt-report may help to evaluate this, and might lead you to use some SSE3 intrinsics in case the overhead of splitting the data would be significant.</p><p style="word-wrap:break-word;margin:0px 0px 1.5em;padding:0px;font-family:Arial,宋体,Tahoma,Helvetica,sans-serif;font-size:12px;font-weight:normal;color:rgb(83,87,94)"><br></p><p style="word-wrap:break-word;margin:0px 0px 1.5em;padding:0px;font-family:Arial,宋体,Tahoma,Helvetica,sans-serif;font-size:12px;font-weight:normal;color:rgb(83,87,94)">Page 5-26 of <a href="https://www.intel.com/content/dam/www/public/us/en/documents/manuals/64-ia-32-architectures-optimization-manual.pdf">https://www.intel.com/content/dam/www/public/us/en/documents/manuals/64-ia-32-architectures-optimization-manual.pdf</a> may have useful examples as well</p></div></div></div></div></div></div><div class="gmail-forum-post-footer gmail-clearfix" style="margin-top:0px;clear:both;border-bottom:1px solid rgb(153,153,153);padding-bottom:10px;zoom:1;color:rgb(83,87,94);font-family:Arial,宋体,Tahoma,Helvetica,sans-serif;font-size:13.008px;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;background-color:rgb(255,255,255);text-decoration-style:initial;text-decoration-color:initial"><br></div></div><div>       Richard</div><div class="gmail_extra"><br><div class="gmail_quote">On Tue, Apr 10, 2018 at 9:20 AM,  <span dir="ltr"><<a href="mailto:Chris.Phillips@csiro.au" target="_blank">Chris.Phillips@csiro.au</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">Hi Walter<br>
<br>
No I haven’t, but just a note that when I tried DIFX  on the KnightsLanding an year or so ago, IPP had not been ported to AVX512 at that stage<br>
<br>
Cheers<br>
<span class="gmail-m_7810411253554822398HOEnZb"><font color="#888888">Chris<br>
</font></span><div class="gmail-m_7810411253554822398HOEnZb"><div class="gmail-m_7810411253554822398h5"><br>
<br>
<br>
> On 10 Apr 2018, at 11:15, Walter Brisken <<a href="mailto:wbrisken@lbo.us" target="_blank">wbrisken@lbo.us</a>> wrote:<br>
><br>
><br>
> Hi DiFX Users,<br>
><br>
> I'm wondering if anyone on this list has tried DiFX on Intel XEON CPUs called "Scalable Platform".  These are distinguished by "bronze", "silver", "gold", or "platinum" varieties.  They are also labeled as "Skylake-SP" processors.  If anyone has such experience, I'm wondering if any benchmarks are available.  Also wondering if any attempt to evaluate improvement the AVX512 instructions offer in DiFX computing.<br>
><br>
> Thanks,<br>
><br>
> Walter<br>
><br>
> --<br>
> -------------------------<br>
> Walter Brisken<br>
> Director<br>
> Long Baseline Observatory<br>
> (575)-835-7133 (office)<br>
> (505)-234-5912 (cell)<br>
><br>
> ______________________________<wbr>_________________<br>
> Difx-users mailing list<br>
> <a href="mailto:Difx-users@listmgr.nrao.edu" target="_blank">Difx-users@listmgr.nrao.edu</a><br>
> <a href="https://listmgr.nrao.edu/mailman/listinfo/difx-users" rel="noreferrer" target="_blank">https://listmgr.nrao.edu/mailm<wbr>an/listinfo/difx-users</a><br>
<br>
<br>
______________________________<wbr>_________________<br>
Difx-users mailing list<br>
<a href="mailto:Difx-users@listmgr.nrao.edu" target="_blank">Difx-users@listmgr.nrao.edu</a><br>
<a href="https://listmgr.nrao.edu/mailman/listinfo/difx-users" rel="noreferrer" target="_blank">https://listmgr.nrao.edu/mailm<wbr>an/listinfo/difx-users</a><br>
</div></div></blockquote></div><br><br clear="all"><div><br></div>-- <br><div class="gmail-m_7810411253554822398gmail_signature">-------------------------<br>Dr Richard Dodson,<br>International Centre for Radio Astronomy Research<br>University of Western Australia<br>P: +8 6488 7842 E: <a href="mailto:richard.dodson@icrar.org" target="_blank">richard.dodson@icrar.org</a></div>
</div></div></div>