<!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01 Transitional//EN">
<html><head><meta http-equiv="Content-Type" content="text/html;charset=iso-8859-1">
<title>CImage_SSE2.cpp Source File</title>
<link href="doxygen.css" rel="stylesheet" type="text/css">
<link href="tabs.css" rel="stylesheet" type="text/css">
</head><body>
<div align="left"><a href="http://www.mrpt.org/">Main MRPT website</a> > <b>C++ reference</b> </div>
<div align="right">
<a href="index.html"><img border="0" src="mrpt_logo.png" alt="MRPT logo"></a>
</div>
<!-- Generated by Doxygen 1.7.5 -->
<script type="text/javascript">
var searchBox = new SearchBox("searchBox", "search",false,'Search');
</script>
<div id="navrow1" class="tabs">
<ul class="tablist">
<li><a href="index.html"><span>Main Page</span></a></li>
<li><a href="pages.html"><span>Related Pages</span></a></li>
<li><a href="modules.html"><span>Modules</span></a></li>
<li><a href="namespaces.html"><span>Namespaces</span></a></li>
<li><a href="annotated.html"><span>Classes</span></a></li>
<li class="current"><a href="files.html"><span>Files</span></a></li>
<li>
<div id="MSearchBox" class="MSearchBoxInactive">
<div class="left">
<form id="FSearchBox" action="search.php" method="get">
<img id="MSearchSelect" src="search/mag.png" alt=""/>
<input type="text" id="MSearchField" name="query" value="Search" size="20" accesskey="S"
onfocus="searchBox.OnSearchFieldFocus(true)"
onblur="searchBox.OnSearchFieldFocus(false)"/>
</form>
</div><div class="right"></div>
</div>
</li>
</ul>
</div>
<div id="navrow2" class="tabs2">
<ul class="tablist">
<li><a href="files.html"><span>File List</span></a></li>
<li><a href="globals.html"><span>File Members</span></a></li>
</ul>
</div>
<div class="header">
<div class="headertitle">
<div class="title">CImage_SSE2.cpp</div> </div>
</div>
<div class="contents">
<a href="_c_image___s_s_e2_8cpp.html">Go to the documentation of this file.</a><div class="fragment"><pre class="fragment"><a name="l00001"></a>00001 <span class="comment">/* +---------------------------------------------------------------------------+</span>
<a name="l00002"></a>00002 <span class="comment"> | The Mobile Robot Programming Toolkit (MRPT) C++ library |</span>
<a name="l00003"></a>00003 <span class="comment"> | |</span>
<a name="l00004"></a>00004 <span class="comment"> | http://www.mrpt.org/ |</span>
<a name="l00005"></a>00005 <span class="comment"> | |</span>
<a name="l00006"></a>00006 <span class="comment"> | Copyright (C) 2005-2011 University of Malaga |</span>
<a name="l00007"></a>00007 <span class="comment"> | |</span>
<a name="l00008"></a>00008 <span class="comment"> | This software was written by the Machine Perception and Intelligent |</span>
<a name="l00009"></a>00009 <span class="comment"> | Robotics Lab, University of Malaga (Spain). |</span>
<a name="l00010"></a>00010 <span class="comment"> | Contact: Jose-Luis Blanco <jlblanco@ctima.uma.es> |</span>
<a name="l00011"></a>00011 <span class="comment"> | |</span>
<a name="l00012"></a>00012 <span class="comment"> | This file is part of the MRPT project. |</span>
<a name="l00013"></a>00013 <span class="comment"> | |</span>
<a name="l00014"></a>00014 <span class="comment"> | MRPT is free software: you can redistribute it and/or modify |</span>
<a name="l00015"></a>00015 <span class="comment"> | it under the terms of the GNU General Public License as published by |</span>
<a name="l00016"></a>00016 <span class="comment"> | the Free Software Foundation, either version 3 of the License, or |</span>
<a name="l00017"></a>00017 <span class="comment"> | (at your option) any later version. |</span>
<a name="l00018"></a>00018 <span class="comment"> | |</span>
<a name="l00019"></a>00019 <span class="comment"> | MRPT is distributed in the hope that it will be useful, |</span>
<a name="l00020"></a>00020 <span class="comment"> | but WITHOUT ANY WARRANTY; without even the implied warranty of |</span>
<a name="l00021"></a>00021 <span class="comment"> | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |</span>
<a name="l00022"></a>00022 <span class="comment"> | GNU General Public License for more details. |</span>
<a name="l00023"></a>00023 <span class="comment"> | |</span>
<a name="l00024"></a>00024 <span class="comment"> | You should have received a copy of the GNU General Public License |</span>
<a name="l00025"></a>00025 <span class="comment"> | along with MRPT. If not, see <http://www.gnu.org/licenses/>. |</span>
<a name="l00026"></a>00026 <span class="comment"> | |</span>
<a name="l00027"></a>00027 <span class="comment"> +---------------------------------------------------------------------------+ */</span>
<a name="l00028"></a>00028
<a name="l00029"></a>00029 <span class="preprocessor">#include <<a class="code" href="base_8h.html">mrpt/base.h</a>></span> <span class="comment">// Precompiled headers</span>
<a name="l00030"></a>00030
<a name="l00031"></a>00031 <span class="preprocessor">#if MRPT_HAS_SSE2</span>
<a name="l00032"></a>00032 <span class="preprocessor"></span><span class="comment">// ---------------------------------------------------------------------------</span>
<a name="l00033"></a>00033 <span class="comment">// This file contains the SSE2 optimized functions for mrpt::utils::CImage</span>
<a name="l00034"></a>00034 <span class="comment">// See the sources and the doxygen documentation page "sse_optimizations" for more details.</span>
<a name="l00035"></a>00035 <span class="comment">//</span>
<a name="l00036"></a>00036 <span class="comment">// Some functions here are derived from sources in libcvd, released</span>
<a name="l00037"></a>00037 <span class="comment">// under LGPL. See http://mi.eng.cam.ac.uk/~er258/cvd/</span>
<a name="l00038"></a>00038 <span class="comment">//</span>
<a name="l00039"></a>00039 <span class="comment">// ---------------------------------------------------------------------------</span>
<a name="l00040"></a>00040
<a name="l00041"></a>00041 <span class="preprocessor">#include <<a class="code" href="_c_image_8h.html">mrpt/utils/CImage.h</a>></span>
<a name="l00042"></a>00042 <span class="preprocessor">#include <<a class="code" href="_s_s_e__types_8h.html">mrpt/utils/SSE_types.h</a>></span>
<a name="l00043"></a>00043 <span class="preprocessor">#include <<a class="code" href="_s_s_e__macros_8h.html">mrpt/utils/SSE_macros.h</a>></span>
<a name="l00044"></a>00044 <span class="preprocessor">#include "CImage_SSEx.h"</span>
<a name="l00045"></a>00045 <span class="comment"></span>
<a name="l00046"></a>00046 <span class="comment">/** \addtogroup sse_optimizations</span>
<a name="l00047"></a>00047 <span class="comment"> * SSE optimized functions</span>
<a name="l00048"></a>00048 <span class="comment"> * @{</span>
<a name="l00049"></a>00049 <span class="comment"> */</span>
<a name="l00050"></a>00050 <span class="comment"></span>
<a name="l00051"></a>00051 <span class="comment">/** Subsample each 2x2 pixel block into 1x1 pixel, taking the first pixel & ignoring the other 3</span>
<a name="l00052"></a>00052 <span class="comment"> * - <b>Input format:</b> uint8_t, 1 channel</span>
<a name="l00053"></a>00053 <span class="comment"> * - <b>Output format:</b> uint8_t, 1 channel</span>
<a name="l00054"></a>00054 <span class="comment"> * - <b>Preconditions:</b> in & out aligned to 16bytes, w = k*16 (w=width in pixels), widthStep=w*1</span>
<a name="l00055"></a>00055 <span class="comment"> * - <b>Notes:</b></span>
<a name="l00056"></a>00056 <span class="comment"> * - <b>Requires:</b> SSE2</span>
<a name="l00057"></a>00057 <span class="comment"> * - <b>Invoked from:</b> mrpt::utils::CImage::scaleHalf()</span>
<a name="l00058"></a>00058 <span class="comment"> */</span>
<a name="l00059"></a><a class="code" href="group__sse__optimizations.html#ga2ea8698db2e2cb58774eac64b8db0721">00059</a> <span class="keywordtype">void</span> <a class="code" href="group__sse__optimizations.html#ga2ea8698db2e2cb58774eac64b8db0721" title="Subsample each 2x2 pixel block into 1x1 pixel, taking the first pixel & ignoring the other 3...">image_SSE2_scale_half_1c8u</a>(<span class="keyword">const</span> uint8_t* in, uint8_t* out, <span class="keywordtype">int</span> w, <span class="keywordtype">int</span> h)
<a name="l00060"></a>00060 {
<a name="l00061"></a>00061 EIGEN_ALIGN16 <span class="keyword">const</span> <span class="keywordtype">unsigned</span> <span class="keywordtype">long</span> <span class="keywordtype">long</span> mask[2] = {0x00FF00FF00FF00FFull, 0x00FF00FF00FF00FFull};
<a name="l00062"></a>00062 <span class="keyword">const</span> __m128i m = _mm_load_si128((<span class="keyword">const</span> __m128i*)mask);
<a name="l00063"></a>00063
<a name="l00064"></a>00064 <span class="keywordtype">int</span> sw = w >> 4;
<a name="l00065"></a>00065 <span class="keywordtype">int</span> sh = h >> 1;
<a name="l00066"></a>00066
<a name="l00067"></a>00067 <span class="keywordflow">for</span> (<span class="keywordtype">int</span> i=0; i<sh; i++)
<a name="l00068"></a>00068 {
<a name="l00069"></a>00069 <span class="keywordflow">for</span> (<span class="keywordtype">int</span> j=0; j<sw; j++)
<a name="l00070"></a>00070 {
<a name="l00071"></a>00071 <span class="keyword">const</span> __m128i here_sampled = _mm_and_si128( _mm_load_si128((<span class="keyword">const</span> __m128i*)in), m);
<a name="l00072"></a>00072 _mm_storel_epi64((__m128i*)out, _mm_packus_epi16(here_sampled,here_sampled));
<a name="l00073"></a>00073 in += 16;
<a name="l00074"></a>00074 out += 8;
<a name="l00075"></a>00075 }
<a name="l00076"></a>00076 in += w;
<a name="l00077"></a>00077 }
<a name="l00078"></a>00078 }
<a name="l00079"></a>00079
<a name="l00080"></a>00080 <span class="comment"></span>
<a name="l00081"></a>00081 <span class="comment">/** Average each 2x2 pixels into 1x1 pixel (arithmetic average)</span>
<a name="l00082"></a>00082 <span class="comment"> * - <b>Input format:</b> uint8_t, 1 channel</span>
<a name="l00083"></a>00083 <span class="comment"> * - <b>Output format:</b> uint8_t, 1 channel</span>
<a name="l00084"></a>00084 <span class="comment"> * - <b>Preconditions:</b> in & out aligned to 16bytes, w = k*16 (w=width in pixels), widthStep=w*1</span>
<a name="l00085"></a>00085 <span class="comment"> * - <b>Notes:</b></span>
<a name="l00086"></a>00086 <span class="comment"> * - <b>Requires:</b> SSE2</span>
<a name="l00087"></a>00087 <span class="comment"> * - <b>Invoked from:</b> mrpt::utils::CImage::scaleHalfSmooth()</span>
<a name="l00088"></a>00088 <span class="comment"> */</span>
<a name="l00089"></a><a class="code" href="group__sse__optimizations.html#gadd3e6ad655aed605c78f988688b8a986">00089</a> <span class="keywordtype">void</span> <a class="code" href="group__sse__optimizations.html#gadd3e6ad655aed605c78f988688b8a986" title="Average each 2x2 pixels into 1x1 pixel (arithmetic average)">image_SSE2_scale_half_smooth_1c8u</a>(<span class="keyword">const</span> uint8_t* in, uint8_t* out, <span class="keywordtype">int</span> w, <span class="keywordtype">int</span> h)
<a name="l00090"></a>00090 {
<a name="l00091"></a>00091 EIGEN_ALIGN16 <span class="keyword">const</span> <span class="keywordtype">unsigned</span> <span class="keywordtype">long</span> <span class="keywordtype">long</span> mask[2] = {0x00FF00FF00FF00FFull, 0x00FF00FF00FF00FFull};
<a name="l00092"></a>00092 <span class="keyword">const</span> uint8_t* nextRow = in + w;
<a name="l00093"></a>00093 __m128i m = _mm_load_si128((<span class="keyword">const</span> __m128i*)mask);
<a name="l00094"></a>00094 <span class="keywordtype">int</span> sw = w >> 4;
<a name="l00095"></a>00095 <span class="keywordtype">int</span> sh = h >> 1;
<a name="l00096"></a>00096
<a name="l00097"></a>00097 <span class="keywordflow">for</span> (<span class="keywordtype">int</span> i=0; i<sh; i++)
<a name="l00098"></a>00098 {
<a name="l00099"></a>00099 <span class="keywordflow">for</span> (<span class="keywordtype">int</span> j=0; j<sw; j++)
<a name="l00100"></a>00100 {
<a name="l00101"></a>00101 __m128i here = _mm_load_si128((<span class="keyword">const</span> __m128i*)in);
<a name="l00102"></a>00102 __m128i next = _mm_load_si128((<span class="keyword">const</span> __m128i*)nextRow);
<a name="l00103"></a>00103 here = _mm_avg_epu8(here,next);
<a name="l00104"></a>00104 next = _mm_and_si128(_mm_srli_si128(here,1), m);
<a name="l00105"></a>00105 here = _mm_and_si128(here,m);
<a name="l00106"></a>00106 here = _mm_avg_epu16(here, next);
<a name="l00107"></a>00107 _mm_storel_epi64((__m128i*)out, _mm_packus_epi16(here,here));
<a name="l00108"></a>00108 in += 16;
<a name="l00109"></a>00109 nextRow += 16;
<a name="l00110"></a>00110 out += 8;
<a name="l00111"></a>00111 }
<a name="l00112"></a>00112
<a name="l00113"></a>00113 in += w;
<a name="l00114"></a>00114 nextRow += w;
<a name="l00115"></a>00115 }
<a name="l00116"></a>00116 }
<a name="l00117"></a>00117
<a name="l00118"></a>00118
<a name="l00119"></a>00119 <span class="comment"></span>
<a name="l00120"></a>00120 <span class="comment">/** KLT score at a given point of a grayscale image.</span>
<a name="l00121"></a>00121 <span class="comment"> * - <b>Requires:</b> SSE2</span>
<a name="l00122"></a>00122 <span class="comment"> * - <b>Invoked from:</b> mrpt::utils::CImage::KLT_response()</span>
<a name="l00123"></a>00123 <span class="comment"> *</span>
<a name="l00124"></a>00124 <span class="comment"> * This function is not manually optimized for SSE2 but templatized for different</span>
<a name="l00125"></a>00125 <span class="comment"> * window sizes such as the compiler can optimize automatically for that size.</span>
<a name="l00126"></a>00126 <span class="comment"> *</span>
<a name="l00127"></a>00127 <span class="comment"> * Only for the most common window sizes this templates are instantiated (W=[2-16] and W=32 ),</span>
<a name="l00128"></a>00128 <span class="comment"> * falling back to</span>
<a name="l00129"></a>00129 <span class="comment"> * a generic implementation otherwise. The next figure shows the performance (time for</span>
<a name="l00130"></a>00130 <span class="comment"> * KLT_response() to compute the score for one single pixel) for different window sizes.</span>
<a name="l00131"></a>00131 <span class="comment"> *</span>
<a name="l00132"></a>00132 <span class="comment"> * <img src="KLT_response_performance_SSE2.png" ></span>
<a name="l00133"></a>00133 <span class="comment"> *</span>
<a name="l00134"></a>00134 <span class="comment"> */</span>
<a name="l00135"></a>00135 <span class="keywordtype">float</span> <a class="code" href="group__sse__optimizations.html#ga0546ed6bea3d509c18dabb603eb8bf94" title="KLT score at a given point of a grayscale image.">KLT_response_optimized</a>();
<a name="l00136"></a>00136
<a name="l00137"></a>00137 <span class="comment">// TODO:</span>
<a name="l00138"></a>00138 <span class="comment">// Sum of absolute differences: Use _mm_sad_epu8</span>
<a name="l00139"></a>00139 <span class="comment"></span>
<a name="l00140"></a>00140 <span class="comment">/** @} */</span>
<a name="l00141"></a>00141
<a name="l00142"></a>00142 <span class="preprocessor">#endif // end if MRPT_HAS_SSE2</span>
</pre></div></div>
</div>
<br><hr><br> <table border="0" width="100%"> <tr> <td> Page generated by <a href="http://www.doxygen.org" target="_blank">Doxygen 1.7.5</a> for MRPT 0.9.5 SVN: at Sun Sep 25 17:20:18 UTC 2011</td><td></td> <td width="100"> </td> <td width="150"> </td></tr> </table> </body></html>
