<!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_SSE3.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_SSE3.cpp</div> </div>
</div>
<div class="contents">
<a href="_c_image___s_s_e3_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="comment">// ---------------------------------------------------------------------------</span>
<a name="l00032"></a>00032 <span class="comment">// This file contains the SSE3/SSSE3 optimized functions for mrpt::utils::CImage</span>
<a name="l00033"></a>00033 <span class="comment">// See the sources and the doxygen documentation page "sse_optimizations" for more details.</span>
<a name="l00034"></a>00034 <span class="comment">// ---------------------------------------------------------------------------</span>
<a name="l00035"></a>00035 <span class="preprocessor">#if MRPT_HAS_SSE3</span>
<a name="l00036"></a>00036 <span class="preprocessor"></span>
<a name="l00037"></a>00037 <span class="preprocessor">#include <<a class="code" href="_c_image_8h.html">mrpt/utils/CImage.h</a>></span>
<a name="l00038"></a>00038 <span class="preprocessor">#include <<a class="code" href="_s_s_e__types_8h.html">mrpt/utils/SSE_types.h</a>></span>
<a name="l00039"></a>00039 <span class="preprocessor">#include <<a class="code" href="_s_s_e__macros_8h.html">mrpt/utils/SSE_macros.h</a>></span>
<a name="l00040"></a>00040 <span class="preprocessor">#include "CImage_SSEx.h"</span>
<a name="l00041"></a>00041
<a name="l00042"></a>00042 <span class="comment"></span>
<a name="l00043"></a>00043 <span class="comment">/** \addtogroup sse_optimizations</span>
<a name="l00044"></a>00044 <span class="comment"> * SSE optimized functions</span>
<a name="l00045"></a>00045 <span class="comment"> * @{</span>
<a name="l00046"></a>00046 <span class="comment"> */</span>
<a name="l00047"></a>00047 <span class="comment"></span>
<a name="l00048"></a>00048 <span class="comment">/** Subsample each 2x2 pixel block into 1x1 pixel, taking the first pixel & ignoring the other 3</span>
<a name="l00049"></a>00049 <span class="comment"> * - <b>Input format:</b> uint8_t, 3 channels (RGB or BGR)</span>
<a name="l00050"></a>00050 <span class="comment"> * - <b>Output format:</b> uint8_t, 3 channels (RGB or BGR)</span>
<a name="l00051"></a>00051 <span class="comment"> * - <b>Preconditions:</b> in & out aligned to 16bytes, w = k*16 (w=width in pixels), widthStep=w*3</span>
<a name="l00052"></a>00052 <span class="comment"> * - <b>Notes:</b></span>
<a name="l00053"></a>00053 <span class="comment"> * - <b>Requires:</b> SSSE3</span>
<a name="l00054"></a>00054 <span class="comment"> * - <b>Invoked from:</b> mrpt::utils::CImage::scaleHalf()</span>
<a name="l00055"></a>00055 <span class="comment"> */</span>
<a name="l00056"></a><a class="code" href="group__sse__optimizations.html#ga4829c652864d5e5bfc8d91e1993930aa">00056</a> <span class="keywordtype">void</span> <a class="code" href="group__sse__optimizations.html#ga4829c652864d5e5bfc8d91e1993930aa" title="Subsample each 2x2 pixel block into 1x1 pixel, taking the first pixel & ignoring the other 3...">image_SSSE3_scale_half_3c8u</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="l00057"></a>00057 {
<a name="l00058"></a>00058 EIGEN_ALIGN16 <span class="keyword">const</span> <span class="keywordtype">unsigned</span> <span class="keywordtype">long</span> <span class="keywordtype">long</span> mask0[2] = { 0x0D0C080706020100ull, 0x808080808080800Eull }; <span class="comment">// Long words are in inverse order due to little endianness</span>
<a name="l00059"></a>00059 EIGEN_ALIGN16 <span class="keyword">const</span> <span class="keywordtype">unsigned</span> <span class="keywordtype">long</span> <span class="keywordtype">long</span> mask1[2] = { 0x8080808080808080ull, 0x0E0A090804030280ull };
<a name="l00060"></a>00060 EIGEN_ALIGN16 <span class="keyword">const</span> <span class="keywordtype">unsigned</span> <span class="keywordtype">long</span> <span class="keywordtype">long</span> mask2[2] = { 0x0C0B0A0605040080ull, 0x8080808080808080ull };
<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> mask3[2] = { 0x808080808080800Full, 0x8080808080808080ull };
<a name="l00062"></a>00062
<a name="l00063"></a>00063 <span class="keyword">const</span> __m128i m0 = _mm_load_si128((<span class="keyword">const</span> __m128i*)mask0);
<a name="l00064"></a>00064 <span class="keyword">const</span> __m128i m1 = _mm_load_si128((<span class="keyword">const</span> __m128i*)mask1);
<a name="l00065"></a>00065 <span class="keyword">const</span> __m128i m2 = _mm_load_si128((<span class="keyword">const</span> __m128i*)mask2);
<a name="l00066"></a>00066 <span class="keyword">const</span> __m128i m3 = _mm_load_si128((<span class="keyword">const</span> __m128i*)mask3);
<a name="l00067"></a>00067
<a name="l00068"></a>00068 <span class="keyword">const</span> <span class="keywordtype">int</span> sw = w >> 4; <span class="comment">// This are the number of 3*16 blocks in each row</span>
<a name="l00069"></a>00069 <span class="keyword">const</span> <span class="keywordtype">int</span> sh = h >> 1;
<a name="l00070"></a>00070
<a name="l00071"></a>00071 <span class="keywordtype">int</span> odd_row = 0;
<a name="l00072"></a>00072
<a name="l00073"></a>00073 <span class="keywordflow">for</span> (<span class="keywordtype">int</span> i=0; i<sh; i++)
<a name="l00074"></a>00074 {
<a name="l00075"></a>00075 <span class="keywordflow">for</span> (<span class="keywordtype">int</span> j=0; j<sw; j++)
<a name="l00076"></a>00076 {
<a name="l00077"></a>00077 <span class="comment">// 16-byte blocks #0,#1,#2:</span>
<a name="l00078"></a>00078 __m128i d0 = _mm_load_si128((<span class="keyword">const</span> __m128i*)in); in += 16;
<a name="l00079"></a>00079 __m128i d1 = _mm_load_si128((<span class="keyword">const</span> __m128i*)in); in += 16;
<a name="l00080"></a>00080
<a name="l00081"></a>00081 <span class="comment">// First 16 bytes:</span>
<a name="l00082"></a>00082 __m128i shuf0 = _mm_shuffle_epi8(d0,m0);
<a name="l00083"></a>00083 __m128i shuf1 = _mm_shuffle_epi8(d1,m1);
<a name="l00084"></a>00084
<a name="l00085"></a>00085 __m128i res0 = _mm_or_si128(shuf0,shuf1);
<a name="l00086"></a>00086
<a name="l00087"></a>00087 <span class="keywordflow">if</span> ((odd_row&0x1)!=0)
<a name="l00088"></a>00088 _mm_storeu_si128((__m128i*)out,res0); <span class="comment">// unaligned output</span>
<a name="l00089"></a>00089 <span class="keywordflow">else</span> _mm_store_si128 ((__m128i*)out,res0); <span class="comment">// aligned output</span>
<a name="l00090"></a>00090 out += 16;
<a name="l00091"></a>00091
<a name="l00092"></a>00092 <span class="comment">// Last 8 bytes:</span>
<a name="l00093"></a>00093 __m128i d2 = _mm_load_si128((<span class="keyword">const</span> __m128i*)in); in += 16;
<a name="l00094"></a>00094
<a name="l00095"></a>00095 _mm_storel_epi64( <span class="comment">// Write lower 8 bytes only</span>
<a name="l00096"></a>00096 (__m128i*)out,
<a name="l00097"></a>00097 _mm_or_si128(_mm_shuffle_epi8(d2,m2),_mm_shuffle_epi8(d1,m3))
<a name="l00098"></a>00098 );
<a name="l00099"></a>00099 odd_row++;
<a name="l00100"></a>00100 out += 8;
<a name="l00101"></a>00101 }
<a name="l00102"></a>00102 in += 3*w;
<a name="l00103"></a>00103 }
<a name="l00104"></a>00104 }
<a name="l00105"></a>00105
<a name="l00106"></a>00106
<a name="l00107"></a>00107 <span class="comment">// This is the actual function behind both: image_SSSE3_rgb_to_gray_8u() and image_SSSE3_bgr_to_gray_8u():</span>
<a name="l00108"></a>00108 <span class="keyword">template</span> <<span class="keywordtype">bool</span> IS_RGB>
<a name="l00109"></a><a class="code" href="group__sse__optimizations.html#ga23e43b45f41b2d80acf48341f986e4af">00109</a> <span class="keywordtype">void</span> <a class="code" href="group__sse__optimizations.html#ga23e43b45f41b2d80acf48341f986e4af">private_image_SSSE3_rgb_or_bgr_to_gray_8u</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="l00110"></a>00110 {
<a name="l00111"></a>00111 <span class="comment">// Masks: 0 1 2 3 4 5 6 7 8 9 A B C D E F</span>
<a name="l00112"></a>00112 <a class="code" href="_s_s_e__macros_8h.html#a3f1c2515952ba96acfd2c755987e4eac">BUILD_128BIT_CONST</a>(mask0, 80,00, 80,03, 80,06, 80,09, 80,0C, 80,0F, 80,80, 80,80) <span class="comment">// reds[0-7] from D0</span>
<a name="l00113"></a>00113 <a class="code" href="_s_s_e__macros_8h.html#a3f1c2515952ba96acfd2c755987e4eac">BUILD_128BIT_CONST</a>(mask1, 80,80, 80,80, 80,80, 80,80, 80,80, 80,80, 80,02, 80,05) <span class="comment">// reds[0-7] from D1</span>
<a name="l00114"></a>00114
<a name="l00115"></a>00115 <a class="code" href="_s_s_e__macros_8h.html#a3f1c2515952ba96acfd2c755987e4eac">BUILD_128BIT_CONST</a>(mask2, 80,01, 80,04, 80,07, 80,0A, 80,0D, 80,80, 80,80, 80,80) <span class="comment">// greens[0-7] from D0</span>
<a name="l00116"></a>00116 <a class="code" href="_s_s_e__macros_8h.html#a3f1c2515952ba96acfd2c755987e4eac">BUILD_128BIT_CONST</a>(mask3, 80,80, 80,80, 80,80, 80,80, 80,80, 80,00, 80,03, 80,06) <span class="comment">// greens[0-7] from D1</span>
<a name="l00117"></a>00117
<a name="l00118"></a>00118 <a class="code" href="_s_s_e__macros_8h.html#a3f1c2515952ba96acfd2c755987e4eac">BUILD_128BIT_CONST</a>(mask4, 80,02, 80,05, 80,08, 80,0B, 80,0E, 80,80, 80,80, 80,80) <span class="comment">// blues[0-7] from D0</span>
<a name="l00119"></a>00119 <a class="code" href="_s_s_e__macros_8h.html#a3f1c2515952ba96acfd2c755987e4eac">BUILD_128BIT_CONST</a>(mask5, 80,80, 80,80, 80,80, 80,80, 80,80, 80,01, 80,04, 80,07) <span class="comment">// blues[0-7] from D1</span>
<a name="l00120"></a>00120
<a name="l00121"></a>00121
<a name="l00122"></a>00122 <a class="code" href="_s_s_e__macros_8h.html#a3f1c2515952ba96acfd2c755987e4eac">BUILD_128BIT_CONST</a>(mask6, 80,08, 80,0B, 80,0E, 80,80, 80,80, 80,80, 80,80, 80,80) <span class="comment">// reds[8-15] from D1</span>
<a name="l00123"></a>00123 <a class="code" href="_s_s_e__macros_8h.html#a3f1c2515952ba96acfd2c755987e4eac">BUILD_128BIT_CONST</a>(mask7, 80,80, 80,80, 80,80, 80,01, 80,04, 80,07, 80,0A, 80,0D) <span class="comment">// reds[8-15] from D2</span>
<a name="l00124"></a>00124
<a name="l00125"></a>00125 <a class="code" href="_s_s_e__macros_8h.html#a3f1c2515952ba96acfd2c755987e4eac">BUILD_128BIT_CONST</a>(mask8, 80,09, 80,0C, 80,0F, 80,80, 80,80, 80,80, 80,80, 80,80) <span class="comment">// greens[8-15] from D1</span>
<a name="l00126"></a>00126 <a class="code" href="_s_s_e__macros_8h.html#a3f1c2515952ba96acfd2c755987e4eac">BUILD_128BIT_CONST</a>(mask9, 80,80, 80,80, 80,80, 80,02, 80,05, 80,08, 80,0B, 80,0E) <span class="comment">// greens[8-15] from D2</span>
<a name="l00127"></a>00127
<a name="l00128"></a>00128 <a class="code" href="_s_s_e__macros_8h.html#a3f1c2515952ba96acfd2c755987e4eac">BUILD_128BIT_CONST</a>(mask10,80,0A, 80,0D, 80,80, 80,80, 80,80, 80,80, 80,80, 80,80) <span class="comment">// blues[8-15] from D1</span>
<a name="l00129"></a>00129 <a class="code" href="_s_s_e__macros_8h.html#a3f1c2515952ba96acfd2c755987e4eac">BUILD_128BIT_CONST</a>(mask11,80,80, 80,80, 80,00, 80,03, 80,06, 80,09, 80,0C, 80,0F) <span class="comment">// blues[8-15] from D2</span>
<a name="l00130"></a>00130
<a name="l00131"></a>00131
<a name="l00132"></a>00132 <a class="code" href="_s_s_e__macros_8h.html#a3f1c2515952ba96acfd2c755987e4eac">BUILD_128BIT_CONST</a>(mask_to_low, 01,03,05,07,09,0B,0D,0F, 80,80,80,80,80,80,80,80)
<a name="l00133"></a>00133
<a name="l00134"></a>00134
<a name="l00135"></a>00135 <span class="comment">// Conversion factors for RGB->Y</span>
<a name="l00136"></a>00136 <a class="code" href="_s_s_e__macros_8h.html#a3f1c2515952ba96acfd2c755987e4eac">BUILD_128BIT_CONST</a>(val_red , 00,1D, 00,1D, 00,1D, 00,1D, 00,1D, 00,1D, 00,1D, 00,1D)
<a name="l00137"></a>00137 <a class="code" href="_s_s_e__macros_8h.html#a3f1c2515952ba96acfd2c755987e4eac">BUILD_128BIT_CONST</a>(val_green , 00,96, 00,96, 00,96, 00,96, 00,96, 00,96, 00,96, 00,96)
<a name="l00138"></a>00138 <a class="code" href="_s_s_e__macros_8h.html#a3f1c2515952ba96acfd2c755987e4eac">BUILD_128BIT_CONST</a>(val_blue , 00,4D, 00,4D, 00,4D, 00,4D, 00,4D, 00,4D, 00,4D, 00,4D)
<a name="l00139"></a>00139
<a name="l00140"></a>00140 <span class="keyword">const</span> __m128i m0 = _mm_load_si128( IS_RGB ? (<span class="keyword">const</span> __m128i*)mask4 : (<span class="keyword">const</span> __m128i*)mask0);
<a name="l00141"></a>00141 <span class="keyword">const</span> __m128i m1 = _mm_load_si128( IS_RGB ? (<span class="keyword">const</span> __m128i*)mask5 : (<span class="keyword">const</span> __m128i*)mask1);
<a name="l00142"></a>00142 <span class="keyword">const</span> __m128i m2 = _mm_load_si128((<span class="keyword">const</span> __m128i*)mask2);
<a name="l00143"></a>00143 <span class="keyword">const</span> __m128i m3 = _mm_load_si128((<span class="keyword">const</span> __m128i*)mask3);
<a name="l00144"></a>00144 <span class="keyword">const</span> __m128i m4 = _mm_load_si128( IS_RGB ? (<span class="keyword">const</span> __m128i*)mask0 : (<span class="keyword">const</span> __m128i*)mask4);
<a name="l00145"></a>00145 <span class="keyword">const</span> __m128i m5 = _mm_load_si128( IS_RGB ? (<span class="keyword">const</span> __m128i*)mask1 : (<span class="keyword">const</span> __m128i*)mask5);
<a name="l00146"></a>00146
<a name="l00147"></a>00147 <span class="keyword">const</span> __m128i m6 = _mm_load_si128( IS_RGB ? (<span class="keyword">const</span> __m128i*)mask10 : (<span class="keyword">const</span> __m128i*)mask6);
<a name="l00148"></a>00148 <span class="keyword">const</span> __m128i m7 = _mm_load_si128( IS_RGB ? (<span class="keyword">const</span> __m128i*)mask11 : (<span class="keyword">const</span> __m128i*)mask7);
<a name="l00149"></a>00149 <span class="keyword">const</span> __m128i m8 = _mm_load_si128((<span class="keyword">const</span> __m128i*)mask8);
<a name="l00150"></a>00150 <span class="keyword">const</span> __m128i m9 = _mm_load_si128((<span class="keyword">const</span> __m128i*)mask9);
<a name="l00151"></a>00151 <span class="keyword">const</span> __m128i m10= _mm_load_si128( IS_RGB ? (<span class="keyword">const</span> __m128i*)mask6 : (<span class="keyword">const</span> __m128i*)mask10);
<a name="l00152"></a>00152 <span class="keyword">const</span> __m128i m11= _mm_load_si128( IS_RGB ? (<span class="keyword">const</span> __m128i*)mask7 : (<span class="keyword">const</span> __m128i*)mask11);
<a name="l00153"></a>00153
<a name="l00154"></a>00154 <span class="keyword">const</span> __m128i mask_low= _mm_load_si128((<span class="keyword">const</span> __m128i*)mask_to_low);
<a name="l00155"></a>00155
<a name="l00156"></a>00156 <span class="keyword">const</span> __m128i VAL_R = _mm_load_si128((<span class="keyword">const</span> __m128i*)val_red);
<a name="l00157"></a>00157 <span class="keyword">const</span> __m128i VAL_G = _mm_load_si128((<span class="keyword">const</span> __m128i*)val_green);
<a name="l00158"></a>00158 <span class="keyword">const</span> __m128i VAL_B = _mm_load_si128((<span class="keyword">const</span> __m128i*)val_blue);
<a name="l00159"></a>00159
<a name="l00160"></a>00160 <span class="keyword">const</span> <span class="keywordtype">int</span> sw = w >> 4; <span class="comment">// This are the number of 3*16 blocks in each row</span>
<a name="l00161"></a>00161 <span class="keyword">const</span> <span class="keywordtype">int</span> sh = h ;
<a name="l00162"></a>00162
<a name="l00163"></a>00163 <span class="keywordflow">for</span> (<span class="keywordtype">int</span> i=0; i<sh; i++)
<a name="l00164"></a>00164 {
<a name="l00165"></a>00165 <span class="keywordflow">for</span> (<span class="keywordtype">int</span> j=0; j<sw; j++)
<a name="l00166"></a>00166 {
<a name="l00167"></a>00167 <span class="comment">// We process RGB data in blocks of 3 x 16byte blocks:</span>
<a name="l00168"></a>00168 <span class="keyword">const</span> __m128i d0 = _mm_load_si128((<span class="keyword">const</span> __m128i*)in); in += 16;
<a name="l00169"></a>00169 <span class="keyword">const</span> __m128i d1 = _mm_load_si128((<span class="keyword">const</span> __m128i*)in); in += 16;
<a name="l00170"></a>00170 <span class="keyword">const</span> __m128i d2 = _mm_load_si128((<span class="keyword">const</span> __m128i*)in); in += 16;
<a name="l00171"></a>00171
<a name="l00172"></a>00172 <span class="comment">// First 8 bytes of gray levels:</span>
<a name="l00173"></a>00173 {
<a name="l00174"></a>00174 <span class="keyword">const</span> __m128i BLUES_0_7 = _mm_or_si128(_mm_shuffle_epi8(d0,m0),_mm_shuffle_epi8(d1,m1));
<a name="l00175"></a>00175 <span class="keyword">const</span> __m128i GREENS_0_7 = _mm_or_si128(_mm_shuffle_epi8(d0,m2),_mm_shuffle_epi8(d1,m3));
<a name="l00176"></a>00176 <span class="keyword">const</span> __m128i REDS_0_7 = _mm_or_si128(_mm_shuffle_epi8(d0,m4),_mm_shuffle_epi8(d1,m5));
<a name="l00177"></a>00177
<a name="l00178"></a>00178 <span class="comment">// _mm_mulhi_epu16(): Multiplies the 8 unsigned 16-bit integers from a by the 8 unsigned 16-bit integers from b.</span>
<a name="l00179"></a>00179 <span class="comment">//r0 := (a0 * b0)[31:16]</span>
<a name="l00180"></a>00180 <span class="comment">//r1 := (a1 * b1)[31:16]</span>
<a name="l00181"></a>00181 <span class="comment">//...</span>
<a name="l00182"></a>00182 <span class="comment">//r7 := (a7 * b7)[31:16]</span>
<a name="l00183"></a>00183 <span class="comment">//</span>
<a name="l00184"></a>00184 <span class="keyword">const</span> __m128i GRAYS_0_7 =
<a name="l00185"></a>00185 _mm_adds_epu16(
<a name="l00186"></a>00186 _mm_mulhi_epu16(REDS_0_7, VAL_R),
<a name="l00187"></a>00187 _mm_adds_epu16(
<a name="l00188"></a>00188 _mm_mulhi_epu16(GREENS_0_7, VAL_G),
<a name="l00189"></a>00189 _mm_mulhi_epu16(BLUES_0_7, VAL_B)
<a name="l00190"></a>00190 ));
<a name="l00191"></a>00191
<a name="l00192"></a>00192 _mm_storel_epi64((__m128i*)out, _mm_shuffle_epi8(GRAYS_0_7,mask_low));
<a name="l00193"></a>00193 out+=8;
<a name="l00194"></a>00194 }
<a name="l00195"></a>00195
<a name="l00196"></a>00196 <span class="comment">// Second 8 bytes of gray levels:</span>
<a name="l00197"></a>00197 {
<a name="l00198"></a>00198 <span class="keyword">const</span> __m128i BLUES_8_15 = _mm_or_si128(_mm_shuffle_epi8(d1,m6),_mm_shuffle_epi8(d2,m7));
<a name="l00199"></a>00199 <span class="keyword">const</span> __m128i GREENS_8_15 = _mm_or_si128(_mm_shuffle_epi8(d1,m8),_mm_shuffle_epi8(d2,m9));
<a name="l00200"></a>00200 <span class="keyword">const</span> __m128i REDS_8_15 = _mm_or_si128(_mm_shuffle_epi8(d1,m10),_mm_shuffle_epi8(d2,m11));
<a name="l00201"></a>00201
<a name="l00202"></a>00202 <span class="keyword">const</span> __m128i GRAYS_8_15 =
<a name="l00203"></a>00203 _mm_adds_epu16(
<a name="l00204"></a>00204 _mm_mulhi_epu16(REDS_8_15, VAL_R),
<a name="l00205"></a>00205 _mm_adds_epu16(
<a name="l00206"></a>00206 _mm_mulhi_epu16(GREENS_8_15, VAL_G),
<a name="l00207"></a>00207 _mm_mulhi_epu16(BLUES_8_15, VAL_B)
<a name="l00208"></a>00208 ));
<a name="l00209"></a>00209
<a name="l00210"></a>00210 _mm_storel_epi64((__m128i*)out, _mm_shuffle_epi8(GRAYS_8_15,mask_low));
<a name="l00211"></a>00211 out+=8;
<a name="l00212"></a>00212 }
<a name="l00213"></a>00213 }
<a name="l00214"></a>00214 }
<a name="l00215"></a>00215
<a name="l00216"></a>00216 } <span class="comment">// end private_image_SSSE3_rgb_or_bgr_to_gray_8u()</span>
<a name="l00217"></a>00217
<a name="l00218"></a>00218 <span class="comment"></span>
<a name="l00219"></a>00219 <span class="comment">/** Convert a RGB image (3cu8) into a GRAYSCALE (1c8u) image, using Y=77*R+150*G+29*B</span>
<a name="l00220"></a>00220 <span class="comment"> * - <b>Input format:</b> uint8_t, 3 channels (BGR order)</span>
<a name="l00221"></a>00221 <span class="comment"> * - <b>Output format:</b> uint8_t, 1 channel</span>
<a name="l00222"></a>00222 <span class="comment"> * - <b>Preconditions:</b> in & out aligned to 16bytes, w = k*16 (w=width in pixels), widthStep=w*3 and w*1</span>
<a name="l00223"></a>00223 <span class="comment"> * - <b>Notes:</b></span>
<a name="l00224"></a>00224 <span class="comment"> * - <b>Requires:</b> SSSE3</span>
<a name="l00225"></a>00225 <span class="comment"> * - <b>Invoked from:</b> mrpt::utils::CImage::grayscale(), mrpt::utils::CImage::grayscaleInPlace()</span>
<a name="l00226"></a>00226 <span class="comment"> */</span>
<a name="l00227"></a><a class="code" href="group__sse__optimizations.html#ga55b830ae32a3393bf57af1c782096eed">00227</a> <span class="keywordtype">void</span> <a class="code" href="group__sse__optimizations.html#ga55b830ae32a3393bf57af1c782096eed" title="Convert a RGB image (3cu8) into a GRAYSCALE (1c8u) image, using Y=77*R+150*G+29*B.">image_SSSE3_bgr_to_gray_8u</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="l00228"></a>00228 {
<a name="l00229"></a>00229 private_image_SSSE3_rgb_or_bgr_to_gray_8u<false>(in,out,w,h);
<a name="l00230"></a>00230 }
<a name="l00231"></a>00231 <span class="comment"></span>
<a name="l00232"></a>00232 <span class="comment">/** Convert a RGB image (3cu8) into a GRAYSCALE (1c8u) image, using Y=77*R+150*G+29*B</span>
<a name="l00233"></a>00233 <span class="comment"> * - <b>Input format:</b> uint8_t, 3 channels (RGB order)</span>
<a name="l00234"></a>00234 <span class="comment"> * - <b>Output format:</b> uint8_t, 1 channel</span>
<a name="l00235"></a>00235 <span class="comment"> * - <b>Preconditions:</b> in & out aligned to 16bytes, w = k*16 (w=width in pixels), widthStep=w*3 and w*1</span>
<a name="l00236"></a>00236 <span class="comment"> * - <b>Notes:</b></span>
<a name="l00237"></a>00237 <span class="comment"> * - <b>Requires:</b> SSSE3</span>
<a name="l00238"></a>00238 <span class="comment"> * - <b>Invoked from:</b> mrpt::utils::CImage::grayscale(), mrpt::utils::CImage::grayscaleInPlace()</span>
<a name="l00239"></a>00239 <span class="comment"> */</span>
<a name="l00240"></a><a class="code" href="group__sse__optimizations.html#ga45eca9cc60ed408bc72735f47c2f88d1">00240</a> <span class="keywordtype">void</span> <a class="code" href="group__sse__optimizations.html#ga45eca9cc60ed408bc72735f47c2f88d1" title="Convert a RGB image (3cu8) into a GRAYSCALE (1c8u) image, using Y=77*R+150*G+29*B.">image_SSSE3_rgb_to_gray_8u</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="l00241"></a>00241 {
<a name="l00242"></a>00242 private_image_SSSE3_rgb_or_bgr_to_gray_8u<true>(in,out,w,h);
<a name="l00243"></a>00243 }
<a name="l00244"></a>00244
<a name="l00245"></a>00245 <span class="comment"></span>
<a name="l00246"></a>00246 <span class="comment">/** @} */</span>
<a name="l00247"></a>00247
<a name="l00248"></a>00248 <span class="preprocessor">#endif // end of MRPT_HAS_SSE3</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>
