You can not select more than 25 topics
Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
|
|
<!DOCTYPE html PUBLIC "-//W3C//DTD HTML 4.01 Transitional//EN" "http://www.w3.org/TR/html4/loose.dtd"><!-- saved from url=(0014)about:internet --><html xmlns:MSHelp="http://www.microsoft.com/MSHelp/" lang="en-us" xml:lang="en-us"><head><meta http-equiv="Content-Type" content="text/html; charset=UTF-8">
<meta name="DC.Type" content="topic"><meta name="DC.Title" content="Elementwise"><meta name="DC.subject" content="Elementwise"><meta name="keywords" content="Elementwise"><meta name="DC.Relation" scheme="URI" content="../../tbb_userguide/Design_Patterns/Design_Patterns.htm"><meta name="DC.Relation" scheme="URI" content="Agglomeration.htm#Agglomeration"><meta name="DC.Relation" scheme="URI" content="Reduction.htm#Reduction"><meta name="DC.Format" content="XHTML"><meta name="DC.Identifier" content="Elementwise"><link rel="stylesheet" type="text/css" href="../../intel_css_styles.css"><title>Elementwise</title><xml><MSHelp:Attr Name="DocSet" Value="Intel"></MSHelp:Attr><MSHelp:Attr Name="Locale" Value="kbEnglish"></MSHelp:Attr><MSHelp:Attr Name="TopicType" Value="kbReference"></MSHelp:Attr></xml></head><body id="Elementwise"> <!-- ==============(Start:NavScript)================= --> <script src="..\..\NavScript.js" language="JavaScript1.2" type="text/javascript"></script> <script language="JavaScript1.2" type="text/javascript">WriteNavLink(2);</script> <!-- ==============(End:NavScript)================= --><a name="Elementwise"><!-- --></a>
<h1 class="topictitle1">Elementwise</h1> <div> <div class="section"><h2 class="sectiontitle">Problem</h2> <p>Initiate similar independent computations across items in a data set, and wait until all complete. </p> </div> <div class="section"><h2 class="sectiontitle">Context</h2> <p>Many serial algorithms sweep over a set of items and do an independent computation on each item. However, if some kind of summary information is collected, use the Reduction pattern instead. </p> </div> <div class="section"><h2 class="sectiontitle">Forces</h2> <p>No information is carried or merged between the computations. </p> </div> <div class="section"><h2 class="sectiontitle">Solution</h2> <p>If the number of items is known in advance, use <samp class="codeph">tbb::parallel_for</samp>. If not, consider using <samp class="codeph">tbb::parallel_do</samp>. </p> <p>Use agglomeration if the individual computations are small relative to scheduler overheads. </p> <p>If the pattern is followed by a reduction on the same data, consider doing the element-wise operation as part of the reduction, so that the combination of the two patterns is accomplished in a single sweep instead of two sweeps. Doing so may improve performance by reducing traffic through the memory hierarchy. </p> </div> <div class="section"><h2 class="sectiontitle">Example</h2> <p>Convolution is often used in signal processing. The convolution of a filter <var>c</var> and signal <var>x</var> is computed as: </p><br><img width="99" height="29" src="Images/image004.jpg"><br> <p>Serial code for this computation might look like: </p> <pre>// Assumes c[0..clen-1] and x[1-clen..xlen-1] are definedfor( int i=0; i<xlen+clen-1; ++i ) { float tmp = 0; for( int j=0; j<clen; ++j ) tmp += c[j]*x[i-j]; y[i] = tmp;}</pre> <p>For simplicity, the fragment assumes that <samp class="codeph">x</samp> is a pointer into an array padded with zeros such that <samp class="codeph">x[k]</samp>returns zero when <samp class="codeph">k<0</samp> or <samp class="codeph">k≥xlen</samp>. </p> <p>The inner loop does not fit the elementwise pattern, because each iteration depends on the previous iteration. However, the outer loop fits the elementwise pattern. It is straightforward to render it using <samp class="codeph">tbb::parallel_for</samp> as shown: </p> <pre>tbb::parallel_for( 0, xlen+clen-1, [=]( int i ) { float tmp = 0; for( int j=0; j<clen; ++j ) tmp += c[j]*x[i-j]; y[i] = tmp;});</pre> <p><samp class="codeph">tbb::parallel_for</samp> does automatic agglomeration by implicitly using <samp class="codeph">tbb::auto_partitioner</samp> in its underlying implementation. If there is reason to agglomerate explicitly, use the overload of <samp class="codeph">tbb::parallel_for</samp> that takes an explicit range argument. The following shows the example transformed to use the overload. </p> <pre>tbb::parallel_for( tbb::blocked_range<int>(0,xlen+clen-1,1000), [=]( tbb::blocked_range<int> r ) { int end = r.end(); for( int i=r.begin(); i!=end; ++i ) { float tmp = 0; for( int j=0; j<clen; ++j ) tmp += c[j]*x[i-j]; y[i] = tmp; } });</pre> <p> </p> </div> </div> <div class="familylinks"><div class="parentlink"><strong>Parent topic:</strong> <a href="../../tbb_userguide/Design_Patterns/Design_Patterns.htm">Design Patterns</a></div></div><div class="See Also"><h2>See Also</h2><div class="linklist"><div><a href="Agglomeration.htm#Agglomeration">Agglomeration </a></div><div><a href="Reduction.htm#Reduction">Reduction </a></div></div></div>
</body></html>
|
