<?xml version="1.0" encoding="UTF-8"?>
<rss xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:rdf="http://www.w3.org/1999/02/22-rdf-syntax-ns#" xmlns:taxo="http://purl.org/rss/1.0/modules/taxonomy/" version="2.0">
  <channel>
    <title>topic Re: nondeterministic performance of IPP call in Intel® Integrated Performance Primitives</title>
    <link>https://community.intel.com/t5/Intel-Integrated-Performance/nondeterministic-performance-of-IPP-call/m-p/895935#M12094</link>
    <description>&lt;DIV style="margin:0px;"&gt;
&lt;DIV id="quote_reply" style="width: 100%; margin-top: 5px;"&gt;
&lt;DIV style="margin-left:2px;margin-right:2px;"&gt;Quoting - &lt;A href="https://community.intel.com/en-us/profile/336498"&gt;Vladimir Dudnik (Intel)&lt;/A&gt;&lt;/DIV&gt;
&lt;DIV style="background-color:#E5E5E5; padding:5px;border: 1px; border-style: inset;margin-left:2px;margin-right:2px;"&gt;&lt;EM&gt;
&lt;DIV style="margin:0px;"&gt;I'm affraid I did not get your question. Is that about different performance you get from sequence of the same IPP calls? It mightcaused bysystem warm-up.. loading data into processor cache and so on.&lt;BR /&gt;&lt;BR /&gt;Vladimir&lt;/DIV&gt;
&lt;BR /&gt;&lt;/EM&gt;&lt;/DIV&gt;
&lt;/DIV&gt;
&lt;/DIV&gt;
thats it. the same sequence, the same input parameters. each time the first performance after lunching the programm gives me a result that differs from all other performances.&lt;BR /&gt;&lt;BR /&gt;lets say in other words.&lt;BR /&gt;i have some 2D signal. I apply some filter for each row and write the result to file.&lt;BR /&gt;next i load the data once again and apply the same filtration and write the result to file.&lt;BR /&gt;i get diffrent results. if i reload the data again and apply the filter, I get the same result as in the second time. and so on as much as &lt;BR /&gt;&lt;BR /&gt;if I draw the result using color autoscaling i cant see the difference. only if i save the data and compare the digits. it means that the profile of the signal is the same. it differs be some scale.&lt;BR /&gt;&lt;BR /&gt;such performance is the same for any computer i'v tested.&lt;BR /&gt;</description>
    <pubDate>Wed, 27 May 2009 13:24:56 GMT</pubDate>
    <dc:creator>chobanu_p_m</dc:creator>
    <dc:date>2009-05-27T13:24:56Z</dc:date>
    <item>
      <title>nondeterministic performance of IPP call</title>
      <link>https://community.intel.com/t5/Intel-Integrated-Performance/nondeterministic-performance-of-IPP-call/m-p/895933#M12092</link>
      <description>I have some problems with calling a FIR filtring function (ippsFIR_32f_I)&lt;BR /&gt;&lt;BR /&gt;it has the same input, including the delay line (i'v checked the input by saving it) for any call.&lt;BR /&gt;&lt;BR /&gt;the symptoms are the following: the result of evaluating the FIR-filtring differs from any other evaluating. and all other calling (exept the first one) are the same.&lt;BR /&gt;&lt;BR /&gt;how could it be? how can I fix it?&lt;BR /&gt;regards,&lt;BR /&gt;Petr</description>
      <pubDate>Mon, 25 May 2009 18:08:05 GMT</pubDate>
      <guid>https://community.intel.com/t5/Intel-Integrated-Performance/nondeterministic-performance-of-IPP-call/m-p/895933#M12092</guid>
      <dc:creator>chobanu_p_m</dc:creator>
      <dc:date>2009-05-25T18:08:05Z</dc:date>
    </item>
    <item>
      <title>Re: nondeterministic performance of IPP call</title>
      <link>https://community.intel.com/t5/Intel-Integrated-Performance/nondeterministic-performance-of-IPP-call/m-p/895934#M12093</link>
      <description>&lt;DIV style="margin:0px;"&gt;I'm affraid I did not get your question. Is that about different performance you get from sequence of the same IPP calls? It mightcaused bysystem warm-up.. loading data into processor cache and so on.&lt;BR /&gt;&lt;BR /&gt;Vladimir&lt;/DIV&gt;
&lt;BR /&gt;</description>
      <pubDate>Wed, 27 May 2009 08:49:52 GMT</pubDate>
      <guid>https://community.intel.com/t5/Intel-Integrated-Performance/nondeterministic-performance-of-IPP-call/m-p/895934#M12093</guid>
      <dc:creator>Vladimir_Dudnik</dc:creator>
      <dc:date>2009-05-27T08:49:52Z</dc:date>
    </item>
    <item>
      <title>Re: nondeterministic performance of IPP call</title>
      <link>https://community.intel.com/t5/Intel-Integrated-Performance/nondeterministic-performance-of-IPP-call/m-p/895935#M12094</link>
      <description>&lt;DIV style="margin:0px;"&gt;
&lt;DIV id="quote_reply" style="width: 100%; margin-top: 5px;"&gt;
&lt;DIV style="margin-left:2px;margin-right:2px;"&gt;Quoting - &lt;A href="https://community.intel.com/en-us/profile/336498"&gt;Vladimir Dudnik (Intel)&lt;/A&gt;&lt;/DIV&gt;
&lt;DIV style="background-color:#E5E5E5; padding:5px;border: 1px; border-style: inset;margin-left:2px;margin-right:2px;"&gt;&lt;EM&gt;
&lt;DIV style="margin:0px;"&gt;I'm affraid I did not get your question. Is that about different performance you get from sequence of the same IPP calls? It mightcaused bysystem warm-up.. loading data into processor cache and so on.&lt;BR /&gt;&lt;BR /&gt;Vladimir&lt;/DIV&gt;
&lt;BR /&gt;&lt;/EM&gt;&lt;/DIV&gt;
&lt;/DIV&gt;
&lt;/DIV&gt;
thats it. the same sequence, the same input parameters. each time the first performance after lunching the programm gives me a result that differs from all other performances.&lt;BR /&gt;&lt;BR /&gt;lets say in other words.&lt;BR /&gt;i have some 2D signal. I apply some filter for each row and write the result to file.&lt;BR /&gt;next i load the data once again and apply the same filtration and write the result to file.&lt;BR /&gt;i get diffrent results. if i reload the data again and apply the filter, I get the same result as in the second time. and so on as much as &lt;BR /&gt;&lt;BR /&gt;if I draw the result using color autoscaling i cant see the difference. only if i save the data and compare the digits. it means that the profile of the signal is the same. it differs be some scale.&lt;BR /&gt;&lt;BR /&gt;such performance is the same for any computer i'v tested.&lt;BR /&gt;</description>
      <pubDate>Wed, 27 May 2009 13:24:56 GMT</pubDate>
      <guid>https://community.intel.com/t5/Intel-Integrated-Performance/nondeterministic-performance-of-IPP-call/m-p/895935#M12094</guid>
      <dc:creator>chobanu_p_m</dc:creator>
      <dc:date>2009-05-27T13:24:56Z</dc:date>
    </item>
    <item>
      <title>Re: nondeterministic performance of IPP call</title>
      <link>https://community.intel.com/t5/Intel-Integrated-Performance/nondeterministic-performance-of-IPP-call/m-p/895936#M12095</link>
      <description>&lt;DIV style="margin:0px;"&gt;&lt;/DIV&gt;
I think it is usual case when data processing time is much less than data loading time. First time you read file into memory it get cached by OS for some time, so the next run data are already in OS file cache.&lt;BR /&gt;&lt;BR /&gt;Another reason is IPP DLLs, we usedelayed load linker option to build IPP DLLs. So the libraries get's loaded not at the start of your program but only after you call the first IPP function. Then again, loaded DLLs stays in OS file cache for some time. So next run will have better performance.&lt;BR /&gt;&lt;BR /&gt;Regards,&lt;BR /&gt; Vladimir&lt;BR /&gt;</description>
      <pubDate>Wed, 27 May 2009 16:30:29 GMT</pubDate>
      <guid>https://community.intel.com/t5/Intel-Integrated-Performance/nondeterministic-performance-of-IPP-call/m-p/895936#M12095</guid>
      <dc:creator>Vladimir_Dudnik</dc:creator>
      <dc:date>2009-05-27T16:30:29Z</dc:date>
    </item>
    <item>
      <title>Re: nondeterministic performance of IPP call</title>
      <link>https://community.intel.com/t5/Intel-Integrated-Performance/nondeterministic-performance-of-IPP-call/m-p/895937#M12096</link>
      <description>&lt;DIV style="margin:0px;"&gt;
&lt;DIV id="quote_reply" style="width: 100%; margin-top: 5px;"&gt;
&lt;DIV style="margin-left:2px;margin-right:2px;"&gt;Quoting - &lt;A href="https://community.intel.com/en-us/profile/336498"&gt;Vladimir Dudnik (Intel)&lt;/A&gt;&lt;/DIV&gt;
&lt;DIV style="background-color:#E5E5E5; padding:5px;border: 1px; border-style: inset;margin-left:2px;margin-right:2px;"&gt;&lt;EM&gt; I think it is usual case when data processing time is much less than data loading time. First time you read file into memory it get cached by OS for some time, so the next run data are already in OS file cache.&lt;BR /&gt;&lt;BR /&gt;Another reason is IPP DLLs, we usedelayed load linker option to build IPP DLLs. So the libraries get's loaded not at the start of your program but only after you call the first IPP function. Then again, loaded DLLs stays in OS file cache for some time. So next run will have better performance.&lt;BR /&gt;&lt;BR /&gt;Regards,&lt;BR /&gt; Vladimir&lt;BR /&gt;&lt;/EM&gt;&lt;/DIV&gt;
&lt;/DIV&gt;
&lt;/DIV&gt;
&lt;BR /&gt;before using filtring I use a call of ippStaticInit. doesn't it load the library to the cache?&lt;BR /&gt;</description>
      <pubDate>Thu, 28 May 2009 06:19:40 GMT</pubDate>
      <guid>https://community.intel.com/t5/Intel-Integrated-Performance/nondeterministic-performance-of-IPP-call/m-p/895937#M12096</guid>
      <dc:creator>chobanu_p_m</dc:creator>
      <dc:date>2009-05-28T06:19:40Z</dc:date>
    </item>
    <item>
      <title>Re: nondeterministic performance of IPP call</title>
      <link>https://community.intel.com/t5/Intel-Integrated-Performance/nondeterministic-performance-of-IPP-call/m-p/895938#M12097</link>
      <description>&lt;DIV style="margin:0px;"&gt;&lt;/DIV&gt;
You do not need to call ippStaticInit when you link with IPP DLLs (it is no ops in DLL case). But I think it should be enough to load DLL.&lt;BR /&gt;&lt;BR /&gt;Regards,&lt;BR /&gt; Vladimir&lt;BR /&gt;</description>
      <pubDate>Fri, 29 May 2009 06:02:16 GMT</pubDate>
      <guid>https://community.intel.com/t5/Intel-Integrated-Performance/nondeterministic-performance-of-IPP-call/m-p/895938#M12097</guid>
      <dc:creator>Vladimir_Dudnik</dc:creator>
      <dc:date>2009-05-29T06:02:16Z</dc:date>
    </item>
  </channel>
</rss>

