documentation/direct__convolution__1x1__3x3__5x5__quantized_8cl_source.xhtml

   1 <!-- HTML header for doxygen 1.8.9.1-->
   2 <!DOCTYPE html PUBLIC "-//W3C//DTD XHTML 1.0 Transitional//EN" "http://www.w3.org/TR/xhtml1/DTD/xhtml1-transitional.dtd">
   3 <html xmlns="http://www.w3.org/1999/xhtml">
   4 <head>
   5 <meta http-equiv="Content-Type" content="text/xhtml;charset=UTF-8"/>
   6 <meta http-equiv="X-UA-Compatible" content="IE=9"/>
   7 <meta name="generator" content="Doxygen 1.8.11"/>
   8 <meta name="robots" content="NOINDEX, NOFOLLOW" /> <!-- Prevent indexing by search engines -->
   9 <title>Compute Library: src/core/CL/cl_kernels/direct_convolution_1x1_3x3_5x5_quantized.cl Source File</title>
  10 <link href="tabs.css" rel="stylesheet" type="text/css"/>
  11 <script type="text/javascript" src="jquery.js"></script>
  12 <script type="text/javascript" src="dynsections.js"></script>
  13 <link href="navtree.css" rel="stylesheet" type="text/css"/>
  14 <script type="text/javascript" src="resize.js"></script>
  15 <script type="text/javascript" src="navtreedata.js"></script>
  16 <script type="text/javascript" src="navtree.js"></script>
  17 <script type="text/javascript">
  18   $(document).ready(initResizable);
  19   $(window).load(resizeHeight);
  20 </script>
  21 <link href="search/search.css" rel="stylesheet" type="text/css"/>
  22 <script type="text/javascript" src="search/searchdata.js"></script>
  23 <script type="text/javascript" src="search/search.js"></script>
  24 <script type="text/javascript">
  25   $(document).ready(function() { init_search(); });
  26 </script>
  27 <script type="text/x-mathjax-config">
  28   MathJax.Hub.Config({
  29     extensions: ["tex2jax.js"],
  30     jax: ["input/TeX","output/HTML-CSS"],
  31 });
  32 </script><script type="text/javascript" src="http://cdn.mathjax.org/mathjax/latest/MathJax.js"></script>
  33 <link href="doxygen.css" rel="stylesheet" type="text/css" />
  34 </head>
  35 <body>
  36 <div id="top"><!-- do not remove this div, it is closed by doxygen! -->
  37 <div id="titlearea">
  38 <table cellspacing="0" cellpadding="0">
  39  <tbody>
  40  <tr style="height: 56px;">
  41   <td style="padding-left: 0.5em;">
  42    <div id="projectname">Compute Library
  43    &#160;<span id="projectnumber">18.02</span>
  44    </div>
  45   </td>
  46  </tr>
  47  </tbody>
  48 </table>
  49 </div>
  50 <!-- end header part -->
  51 <!-- Generated by Doxygen 1.8.11 -->
  52 <script type="text/javascript">
  53 var searchBox = new SearchBox("searchBox", "search",false,'Search');
  54 </script>
  55   <div id="navrow1" class="tabs">
  56     <ul class="tablist">
  57       <li><a href="index.xhtml"><span>Main&#160;Page</span></a></li>
  58       <li><a href="pages.xhtml"><span>Related&#160;Pages</span></a></li>
  59       <li><a href="namespaces.xhtml"><span>Namespaces</span></a></li>
  60       <li><a href="annotated.xhtml"><span>Data&#160;Structures</span></a></li>
  61       <li class="current"><a href="files.xhtml"><span>Files</span></a></li>
  62       <li>
  63         <div id="MSearchBox" class="MSearchBoxInactive">
  64         <span class="left">
  65           <img id="MSearchSelect" src="search/mag_sel.png"
  66                onmouseover="return searchBox.OnSearchSelectShow()"
  67                onmouseout="return searchBox.OnSearchSelectHide()"
  68                alt=""/>
  69           <input type="text" id="MSearchField" value="Search" accesskey="S"
  70                onfocus="searchBox.OnSearchFieldFocus(true)"
  71                onblur="searchBox.OnSearchFieldFocus(false)"
  72                onkeyup="searchBox.OnSearchFieldChange(event)"/>
  73           </span><span class="right">
  74             <a id="MSearchClose" href="javascript:searchBox.CloseResultsWindow()"><img id="MSearchCloseImg" border="0" src="search/close.png" alt=""/></a>
  75           </span>
  76         </div>
  77       </li>
  78     </ul>
  79   </div>
  80   <div id="navrow2" class="tabs2">
  81     <ul class="tablist">
  82       <li><a href="files.xhtml"><span>File&#160;List</span></a></li>
  83       <li><a href="globals.xhtml"><span>Globals</span></a></li>
  84     </ul>
  85   </div>
  86 </div><!-- top -->
  87 <div id="side-nav" class="ui-resizable side-nav-resizable">
  88   <div id="nav-tree">
  89     <div id="nav-tree-contents">
  90       <div id="nav-sync" class="sync"></div>
  91     </div>
  92   </div>
  93   <div id="splitbar" style="-moz-user-select:none;"
  94        class="ui-resizable-handle">
  95   </div>
  96 </div>
  97 <script type="text/javascript">
  98 $(document).ready(function(){initNavTree('direct__convolution__1x1__3x3__5x5__quantized_8cl_source.xhtml','');});
  99 </script>
 100 <div id="doc-content">
 101 <!-- window showing the filter options -->
 102 <div id="MSearchSelectWindow"
 103      onmouseover="return searchBox.OnSearchSelectShow()"
 104      onmouseout="return searchBox.OnSearchSelectHide()"
 105      onkeydown="return searchBox.OnSearchSelectKey(event)">
 106 </div>
 107
 108 <!-- iframe showing the search results (closed by default) -->
 109 <div id="MSearchResultsWindow">
 110 <iframe src="javascript:void(0)" frameborder="0"
 111         name="MSearchResults" id="MSearchResults">
 112 </iframe>
 113 </div>
 114
 115 <div class="header">
 116   <div class="headertitle">
 117 <div class="title">direct_convolution_1x1_3x3_5x5_quantized.cl</div>  </div>
 118 </div><!--header-->
 119 <div class="contents">
 120 <a href="direct__convolution__1x1__3x3__5x5__quantized_8cl.xhtml">Go to the documentation of this file.</a><div class="fragment"><div class="line"><a name="l00001"></a><span class="lineno">    1</span>&#160;<span class="comment">/*</span></div><div class="line"><a name="l00002"></a><span class="lineno">    2</span>&#160;<span class="comment"> * Copyright (c) 2017-2018 ARM Limited.</span></div><div class="line"><a name="l00003"></a><span class="lineno">    3</span>&#160;<span class="comment"> *</span></div><div class="line"><a name="l00004"></a><span class="lineno">    4</span>&#160;<span class="comment"> * SPDX-License-Identifier: MIT</span></div><div class="line"><a name="l00005"></a><span class="lineno">    5</span>&#160;<span class="comment"> *</span></div><div class="line"><a name="l00006"></a><span class="lineno">    6</span>&#160;<span class="comment"> * Permission is hereby granted, free of charge, to any person obtaining a copy</span></div><div class="line"><a name="l00007"></a><span class="lineno">    7</span>&#160;<span class="comment"> * of this software and associated documentation files (the &quot;Software&quot;), to</span></div><div class="line"><a name="l00008"></a><span class="lineno">    8</span>&#160;<span class="comment"> * deal in the Software without restriction, including without limitation the</span></div><div class="line"><a name="l00009"></a><span class="lineno">    9</span>&#160;<span class="comment"> * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or</span></div><div class="line"><a name="l00010"></a><span class="lineno">   10</span>&#160;<span class="comment"> * sell copies of the Software, and to permit persons to whom the Software is</span></div><div class="line"><a name="l00011"></a><span class="lineno">   11</span>&#160;<span class="comment"> * furnished to do so, subject to the following conditions:</span></div><div class="line"><a name="l00012"></a><span class="lineno">   12</span>&#160;<span class="comment"> *</span></div><div class="line"><a name="l00013"></a><span class="lineno">   13</span>&#160;<span class="comment"> * The above copyright notice and this permission notice shall be included in all</span></div><div class="line"><a name="l00014"></a><span class="lineno">   14</span>&#160;<span class="comment"> * copies or substantial portions of the Software.</span></div><div class="line"><a name="l00015"></a><span class="lineno">   15</span>&#160;<span class="comment"> *</span></div><div class="line"><a name="l00016"></a><span class="lineno">   16</span>&#160;<span class="comment"> * THE SOFTWARE IS PROVIDED &quot;AS IS&quot;, WITHOUT WARRANTY OF ANY KIND, EXPRESS OR</span></div><div class="line"><a name="l00017"></a><span class="lineno">   17</span>&#160;<span class="comment"> * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,</span></div><div class="line"><a name="l00018"></a><span class="lineno">   18</span>&#160;<span class="comment"> * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE</span></div><div class="line"><a name="l00019"></a><span class="lineno">   19</span>&#160;<span class="comment"> * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER</span></div><div class="line"><a name="l00020"></a><span class="lineno">   20</span>&#160;<span class="comment"> * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,</span></div><div class="line"><a name="l00021"></a><span class="lineno">   21</span>&#160;<span class="comment"> * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE</span></div><div class="line"><a name="l00022"></a><span class="lineno">   22</span>&#160;<span class="comment"> * SOFTWARE.</span></div><div class="line"><a name="l00023"></a><span class="lineno">   23</span>&#160;<span class="comment"> */</span></div><div class="line"><a name="l00024"></a><span class="lineno">   24</span>&#160;<span class="preprocessor">#include &quot;<a class="code" href="helpers__asymm_8h.xhtml">helpers_asymm.h</a>&quot;</span></div><div class="line"><a name="l00025"></a><span class="lineno">   25</span>&#160;</div><div class="line"><a name="l00026"></a><span class="lineno">   26</span>&#160;<span class="preprocessor">#undef CONVERT_SAT</span></div><div class="line"><a name="l00027"></a><span class="lineno">   27</span>&#160;</div><div class="line"><a name="l00028"></a><span class="lineno">   28</span>&#160;<span class="preprocessor">#if defined(DATA_TYPE) &amp;&amp; defined(STRIDE_X) &amp;&amp; defined(WEIGHTS_DEPTH)</span></div><div class="line"><a name="l00029"></a><span class="lineno">   29</span>&#160;</div><div class="line"><a name="l00030"></a><span class="lineno">   30</span>&#160;<span class="preprocessor">#if KERNEL_SIZE == 5</span></div><div class="line"><a name="l00031"></a><span class="lineno">   31</span>&#160;</div><div class="line"><a name="l00032"></a><span class="lineno">   32</span>&#160;<span class="preprocessor">#if STRIDE_X == 1</span></div><div class="line"><a name="l00033"></a><span class="lineno">   33</span>&#160;<span class="preprocessor">#define CONVOLUTION1x5(acc, src_row_ptr, weights_row_ptr) CONVOLUTION1x5_STRIDE1(acc, src_row_ptr, weights_row_ptr)</span></div><div class="line"><a name="l00034"></a><span class="lineno">   34</span>&#160;<span class="preprocessor">#elif STRIDE_X == 2</span></div><div class="line"><a name="l00035"></a><span class="lineno">   35</span>&#160;<span class="preprocessor">#define CONVOLUTION1x5(acc, src_row_ptr, weights_row_ptr) CONVOLUTION1x5_STRIDE2(acc, src_row_ptr, weights_row_ptr)</span></div><div class="line"><a name="l00036"></a><span class="lineno">   36</span>&#160;<span class="preprocessor">#else </span><span class="comment">/* STRIDE_X not equals 1 or 2 */</span><span class="preprocessor"></span></div><div class="line"><a name="l00037"></a><span class="lineno">   37</span>&#160;<span class="preprocessor">#error &quot;STRIDE_X larger than 2 is not supported&quot;</span></div><div class="line"><a name="l00038"></a><span class="lineno">   38</span>&#160;<span class="preprocessor">#endif </span><span class="comment">/* STRIDE_X */</span><span class="preprocessor"></span></div><div class="line"><a name="l00039"></a><span class="lineno">   39</span>&#160;</div><div class="line"><a name="l00040"></a><span class="lineno">   40</span>&#160;<span class="preprocessor">#define CONVOLUTION1x5_STRIDE1(acc, src_row_ptr, weights_row_ptr)                                                    \</span></div><div class="line"><a name="l00041"></a><span class="lineno">   41</span>&#160;<span class="preprocessor">    ({                                                                                                               \</span></div><div class="line"><a name="l00042"></a><span class="lineno">   42</span>&#160;<span class="preprocessor">        int4 weights_values0 = convert_int4(vload4(0, weights_row_ptr));                                             \</span></div><div class="line"><a name="l00043"></a><span class="lineno">   43</span>&#160;<span class="preprocessor">        int  weights_value1  = convert_int(*(weights_row_ptr + 4));                                                  \</span></div><div class="line"><a name="l00044"></a><span class="lineno">   44</span>&#160;<span class="preprocessor">        int8 src0            = convert_int8(vload8(0, src_row_ptr));                                                 \</span></div><div class="line"><a name="l00045"></a><span class="lineno">   45</span>&#160;<span class="preprocessor">        int4 src1            = convert_int4(vload4(0, src_row_ptr + 8));                                             \</span></div><div class="line"><a name="l00046"></a><span class="lineno">   46</span>&#160;<span class="preprocessor">        acc += (src0 + input_offset) * ((int8)weights_values0.s0 + weight_offset);                                   \</span></div><div class="line"><a name="l00047"></a><span class="lineno">   47</span>&#160;<span class="preprocessor">        acc += ((int8)(src0.s1234, src0.s567, src1.s0) + input_offset) * ((int8)weights_values0.s1 + weight_offset); \</span></div><div class="line"><a name="l00048"></a><span class="lineno">   48</span>&#160;<span class="preprocessor">        acc += ((int8)(src0.s234, src0.s567, src1.s01) + input_offset) * ((int8)weights_values0.s2 + weight_offset); \</span></div><div class="line"><a name="l00049"></a><span class="lineno">   49</span>&#160;<span class="preprocessor">        acc += ((int8)(src0.s345, src0.s67, src1.s012) + input_offset) * ((int8)weights_values0.s3 + weight_offset); \</span></div><div class="line"><a name="l00050"></a><span class="lineno">   50</span>&#160;<span class="preprocessor">        acc += ((int8)(src0.s45, src0.s67, src1.s0123) + input_offset) * ((int8)weights_value1 + weight_offset);     \</span></div><div class="line"><a name="l00051"></a><span class="lineno">   51</span>&#160;<span class="preprocessor">    })</span></div><div class="line"><a name="l00052"></a><span class="lineno">   52</span>&#160;</div><div class="line"><a name="l00053"></a><span class="lineno">   53</span>&#160;<span class="preprocessor">#define CONVOLUTION1x5_STRIDE2(acc, src_row_ptr, weights_row_ptr)                                                    \</span></div><div class="line"><a name="l00054"></a><span class="lineno">   54</span>&#160;<span class="preprocessor">    ({                                                                                                               \</span></div><div class="line"><a name="l00055"></a><span class="lineno">   55</span>&#160;<span class="preprocessor">        int4  weights_values0 = convert_int4(vload4(0, weights_row_ptr));                                            \</span></div><div class="line"><a name="l00056"></a><span class="lineno">   56</span>&#160;<span class="preprocessor">        int   weights_value1  = convert_int(*(weights_row_ptr + 4));                                                 \</span></div><div class="line"><a name="l00057"></a><span class="lineno">   57</span>&#160;<span class="preprocessor">        int16 src0            = convert_int16(vload16(0, src_row_ptr));                                              \</span></div><div class="line"><a name="l00058"></a><span class="lineno">   58</span>&#160;<span class="preprocessor">        int4  src1            = convert_int4(vload4(0, src_row_ptr + 16));                                           \</span></div><div class="line"><a name="l00059"></a><span class="lineno">   59</span>&#160;<span class="preprocessor">        acc += (src0.even + input_offset) * ((int8)weights_values0.s0 + weight_offset);                              \</span></div><div class="line"><a name="l00060"></a><span class="lineno">   60</span>&#160;<span class="preprocessor">        acc += ((int8)(src0.s1357, src0.s9BDF) + input_offset) * ((int8)weights_values0.s1 + weight_offset);         \</span></div><div class="line"><a name="l00061"></a><span class="lineno">   61</span>&#160;<span class="preprocessor">        acc += ((int8)(src0.s2468, src0.sACE, src1.s0) + input_offset) * ((int8)weights_values0.s2 + weight_offset); \</span></div><div class="line"><a name="l00062"></a><span class="lineno">   62</span>&#160;<span class="preprocessor">        acc += ((int8)(src0.s3579, src0.sBDF, src1.s1) + input_offset) * ((int8)weights_values0.s3 + weight_offset); \</span></div><div class="line"><a name="l00063"></a><span class="lineno">   63</span>&#160;<span class="preprocessor">        acc += ((int8)(src0.s468a, src0.sCE, src1.s02) + input_offset) * ((int8)weights_value1 + weight_offset);     \</span></div><div class="line"><a name="l00064"></a><span class="lineno">   64</span>&#160;<span class="preprocessor">    })</span></div><div class="line"><a name="l00065"></a><span class="lineno">   65</span>&#160;</div><div class="line"><a name="l00066"></a><span class="lineno">   66</span>&#160;<span class="preprocessor">#elif KERNEL_SIZE == 3</span></div><div class="line"><a name="l00067"></a><span class="lineno">   67</span>&#160;</div><div class="line"><a name="l00068"></a><span class="lineno">   68</span>&#160;<span class="preprocessor">#if STRIDE_X == 1</span></div><div class="line"><a name="l00069"></a><span class="lineno">   69</span>&#160;<span class="preprocessor">#define CONVOLUTION1x3(acc, src_row_ptr, weights_row_ptr) CONVOLUTION1x3_STRIDE1(acc, src_row_ptr, weights_row_ptr)</span></div><div class="line"><a name="l00070"></a><span class="lineno">   70</span>&#160;<span class="preprocessor">#elif STRIDE_X == 2</span></div><div class="line"><a name="l00071"></a><span class="lineno">   71</span>&#160;<span class="preprocessor">#define CONVOLUTION1x3(acc, src_row_ptr, weights_row_ptr) CONVOLUTION1x3_STRIDE2(acc, src_row_ptr, weights_row_ptr)</span></div><div class="line"><a name="l00072"></a><span class="lineno">   72</span>&#160;<span class="preprocessor">#else </span><span class="comment">/* STRIDE_X not equals 1 or 2 */</span><span class="preprocessor"></span></div><div class="line"><a name="l00073"></a><span class="lineno">   73</span>&#160;<span class="preprocessor">#error &quot;STRIDE_X larger than 2 is not supported&quot;</span></div><div class="line"><a name="l00074"></a><span class="lineno">   74</span>&#160;<span class="preprocessor">#endif </span><span class="comment">/* STRIDE_X */</span><span class="preprocessor"></span></div><div class="line"><a name="l00075"></a><span class="lineno">   75</span>&#160;</div><div class="line"><a name="l00076"></a><span class="lineno">   76</span>&#160;<span class="preprocessor">#define CONVOLUTION1x3_STRIDE1(acc, src_row_ptr, weights_row_ptr)                                                    \</span></div><div class="line"><a name="l00077"></a><span class="lineno">   77</span>&#160;<span class="preprocessor">    ({                                                                                                               \</span></div><div class="line"><a name="l00078"></a><span class="lineno">   78</span>&#160;<span class="preprocessor">        int3 weights_values0 = convert_int3(vload3(0, weights_row_ptr));                                             \</span></div><div class="line"><a name="l00079"></a><span class="lineno">   79</span>&#160;<span class="preprocessor">        int8 src0            = convert_int8(vload8(0, src_row_ptr));                                                 \</span></div><div class="line"><a name="l00080"></a><span class="lineno">   80</span>&#160;<span class="preprocessor">        int2 src1            = convert_int2(vload2(0, src_row_ptr + 8));                                             \</span></div><div class="line"><a name="l00081"></a><span class="lineno">   81</span>&#160;<span class="preprocessor">        acc += (src0 + input_offset) * ((int8)weights_values0.s0 + weight_offset);                                   \</span></div><div class="line"><a name="l00082"></a><span class="lineno">   82</span>&#160;<span class="preprocessor">        acc += ((int8)(src0.s1234, src0.s567, src1.s0) + input_offset) * ((int8)weights_values0.s1 + weight_offset); \</span></div><div class="line"><a name="l00083"></a><span class="lineno">   83</span>&#160;<span class="preprocessor">        acc += ((int8)(src0.s234, src0.s567, src1.s01) + input_offset) * ((int8)weights_values0.s2 + weight_offset); \</span></div><div class="line"><a name="l00084"></a><span class="lineno">   84</span>&#160;<span class="preprocessor">    })</span></div><div class="line"><a name="l00085"></a><span class="lineno">   85</span>&#160;</div><div class="line"><a name="l00086"></a><span class="lineno">   86</span>&#160;<span class="preprocessor">#define CONVOLUTION1x3_STRIDE2(acc, src_row_ptr, weights_row_ptr)                                                 \</span></div><div class="line"><a name="l00087"></a><span class="lineno">   87</span>&#160;<span class="preprocessor">    ({                                                                                                            \</span></div><div class="line"><a name="l00088"></a><span class="lineno">   88</span>&#160;<span class="preprocessor">        int3  weights_values0 = convert_int3(vload3(0, weights_row_ptr));                                         \</span></div><div class="line"><a name="l00089"></a><span class="lineno">   89</span>&#160;<span class="preprocessor">        int16 src0            = convert_int16(vload16(0, src_row_ptr));                                           \</span></div><div class="line"><a name="l00090"></a><span class="lineno">   90</span>&#160;<span class="preprocessor">        int   src1            = convert_int(*(src_row_ptr + 16));                                                 \</span></div><div class="line"><a name="l00091"></a><span class="lineno">   91</span>&#160;<span class="preprocessor">        acc += (src0.even + input_offset) * ((int8)weights_values0.s0 + weight_offset);                           \</span></div><div class="line"><a name="l00092"></a><span class="lineno">   92</span>&#160;<span class="preprocessor">        acc += ((int8)(src0.s1357, src0.s9BDF) + input_offset) * ((int8)weights_values0.s1 + weight_offset);      \</span></div><div class="line"><a name="l00093"></a><span class="lineno">   93</span>&#160;<span class="preprocessor">        acc += ((int8)(src0.s2468, src0.sACE, src1) + input_offset) * ((int8)weights_values0.s2 + weight_offset); \</span></div><div class="line"><a name="l00094"></a><span class="lineno">   94</span>&#160;<span class="preprocessor">    })</span></div><div class="line"><a name="l00095"></a><span class="lineno">   95</span>&#160;</div><div class="line"><a name="l00096"></a><span class="lineno">   96</span>&#160;<span class="preprocessor">#elif KERNEL_SIZE == 1</span></div><div class="line"><a name="l00097"></a><span class="lineno">   97</span>&#160;</div><div class="line"><a name="l00098"></a><span class="lineno">   98</span>&#160;<span class="preprocessor">#if STRIDE_X == 3</span></div><div class="line"><a name="l00099"></a><span class="lineno">   99</span>&#160;<span class="preprocessor">#define INPUT_PIXEL extract_input_stride3</span></div><div class="line"><a name="l00100"></a><span class="lineno">  100</span>&#160;<span class="preprocessor">#elif STRIDE_X == 2</span></div><div class="line"><a name="l00101"></a><span class="lineno">  101</span>&#160;<span class="preprocessor">#define INPUT_PIXEL extract_input_stride2</span></div><div class="line"><a name="l00102"></a><span class="lineno">  102</span>&#160;<span class="preprocessor">#elif STRIDE_X == 1</span></div><div class="line"><a name="l00103"></a><span class="lineno">  103</span>&#160;<span class="preprocessor">#define INPUT_PIXEL extract_input_stride1</span></div><div class="line"><a name="l00104"></a><span class="lineno">  104</span>&#160;</div><div class="line"><a name="l00105"></a><span class="lineno">  105</span>&#160;<span class="preprocessor">#else </span><span class="comment">/* STRIDE_X not equals 1, 2 or 3 */</span><span class="preprocessor"></span></div><div class="line"><a name="l00106"></a><span class="lineno">  106</span>&#160;<span class="preprocessor">#error &quot;Only support strides 1, 2 and 3&quot;</span></div><div class="line"><a name="l00107"></a><span class="lineno">  107</span>&#160;<span class="preprocessor">#endif </span><span class="comment">/* STRIDE_X */</span><span class="preprocessor"></span></div><div class="line"><a name="l00108"></a><span class="lineno">  108</span>&#160;</div><div class="line"><a name="l00115"></a><span class="lineno">  115</span>&#160;<span class="keyword">inline</span> uchar8 extract_input_stride1(__global <span class="keyword">const</span> uchar *input_pixel)</div><div class="line"><a name="l00116"></a><span class="lineno">  116</span>&#160;{</div><div class="line"><a name="l00117"></a><span class="lineno">  117</span>&#160;    <span class="keywordflow">return</span> vload8(0, input_pixel);</div><div class="line"><a name="l00118"></a><span class="lineno">  118</span>&#160;}</div><div class="line"><a name="l00119"></a><span class="lineno">  119</span>&#160;</div><div class="line"><a name="l00126"></a><span class="lineno">  126</span>&#160;<span class="keyword">inline</span> uchar8 extract_input_stride2(__global <span class="keyword">const</span> uchar *input_pixel)</div><div class="line"><a name="l00127"></a><span class="lineno">  127</span>&#160;{</div><div class="line"><a name="l00128"></a><span class="lineno">  128</span>&#160;    uchar16 temp = vload16(0, input_pixel);</div><div class="line"><a name="l00129"></a><span class="lineno">  129</span>&#160;    <span class="keywordflow">return</span> temp.s02468ace;</div><div class="line"><a name="l00130"></a><span class="lineno">  130</span>&#160;}</div><div class="line"><a name="l00131"></a><span class="lineno">  131</span>&#160;</div><div class="line"><a name="l00138"></a><span class="lineno">  138</span>&#160;<span class="keyword">inline</span> uchar8 extract_input_stride3(__global <span class="keyword">const</span> uchar *input_pixel)</div><div class="line"><a name="l00139"></a><span class="lineno">  139</span>&#160;{</div><div class="line"><a name="l00140"></a><span class="lineno">  140</span>&#160;    uchar16 temp1 = vload16(0, input_pixel);</div><div class="line"><a name="l00141"></a><span class="lineno">  141</span>&#160;    uchar16 temp2 = vload16(0, input_pixel + 12);</div><div class="line"><a name="l00142"></a><span class="lineno">  142</span>&#160;    <span class="keywordflow">return</span> (uchar8)(temp1.s0369, temp2.s0369);</div><div class="line"><a name="l00143"></a><span class="lineno">  143</span>&#160;}</div><div class="line"><a name="l00144"></a><span class="lineno">  144</span>&#160;</div><div class="line"><a name="l00145"></a><span class="lineno">  145</span>&#160;<span class="preprocessor">#else </span><span class="comment">/* KERNEL_SIZE not equals 1, 3 or 5 */</span><span class="preprocessor"></span></div><div class="line"><a name="l00146"></a><span class="lineno">  146</span>&#160;<span class="preprocessor">#error &quot;Only kernel sizes 1, 3 and 5 are supported&quot;</span></div><div class="line"><a name="l00147"></a><span class="lineno">  147</span>&#160;<span class="preprocessor">#endif </span><span class="comment">/* KERNEL_SIZE */</span><span class="preprocessor"></span></div><div class="line"><a name="l00148"></a><span class="lineno">  148</span>&#160;</div><div class="line"><a name="l00190"></a><span class="lineno">  190</span>&#160;__kernel <span class="keywordtype">void</span> direct_convolution_1x1_3x3_5x5_quantized(</div><div class="line"><a name="l00191"></a><span class="lineno">  191</span>&#160;    <a class="code" href="helpers_8h.xhtml#a6b83038822d1ae7ab619b684ed3b7fc0">TENSOR3D_DECLARATION</a>(<a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#a6743f0a130e8311e6f5b1a23df102472">src</a>),</div><div class="line"><a name="l00192"></a><span class="lineno">  192</span>&#160;    <a class="code" href="helpers_8h.xhtml#a6b83038822d1ae7ab619b684ed3b7fc0">TENSOR3D_DECLARATION</a>(<a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#adbf67dcee294e673cf796f1ed8aeb6a4">dst</a>),</div><div class="line"><a name="l00193"></a><span class="lineno">  193</span>&#160;    <a class="code" href="helpers_8h.xhtml#a6b83038822d1ae7ab619b684ed3b7fc0">TENSOR3D_DECLARATION</a>(weights),</div><div class="line"><a name="l00194"></a><span class="lineno">  194</span>&#160;#ifdef HAS_BIAS</div><div class="line"><a name="l00195"></a><span class="lineno">  195</span>&#160;    <a class="code" href="helpers_8h.xhtml#a40a6eb9f2a7712f08d6bb8ff6c9e6ca7">VECTOR_DECLARATION</a>(biases),</div><div class="line"><a name="l00196"></a><span class="lineno">  196</span>&#160;#endif <span class="comment">/* defined(HAS_BIAS) */</span></div><div class="line"><a name="l00197"></a><span class="lineno">  197</span>&#160;    <span class="keywordtype">unsigned</span> <span class="keywordtype">int</span> weights_stride_w,</div><div class="line"><a name="l00198"></a><span class="lineno">  198</span>&#160;    <span class="keywordtype">int</span>          input_offset,</div><div class="line"><a name="l00199"></a><span class="lineno">  199</span>&#160;    <span class="keywordtype">int</span>          weight_offset,</div><div class="line"><a name="l00200"></a><span class="lineno">  200</span>&#160;    <span class="keywordtype">int</span>          output_offset,</div><div class="line"><a name="l00201"></a><span class="lineno">  201</span>&#160;    <span class="keywordtype">int</span>          output_multiplier,</div><div class="line"><a name="l00202"></a><span class="lineno">  202</span>&#160;    <span class="keywordtype">int</span>          output_shift)</div><div class="line"><a name="l00203"></a><span class="lineno">  203</span>&#160;{</div><div class="line"><a name="l00204"></a><span class="lineno">  204</span>&#160;    <a class="code" href="struct_image.xhtml">Image</a>    <a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#a6743f0a130e8311e6f5b1a23df102472">src</a>     = <a class="code" href="helpers_8h.xhtml#aebe814363556c244be043b13e7969197">CONVERT_TO_IMAGE_STRUCT</a>(src);</div><div class="line"><a name="l00205"></a><span class="lineno">  205</span>&#160;    <a class="code" href="struct_tensor3_d.xhtml">Tensor3D</a> weights = <a class="code" href="helpers_8h.xhtml#a79e8e562daa6599317d2d1cd86ef1bf2">CONVERT_TO_TENSOR3D_STRUCT_NO_STEP</a>(weights);</div><div class="line"><a name="l00206"></a><span class="lineno">  206</span>&#160;    <a class="code" href="struct_tensor3_d.xhtml">Tensor3D</a> <a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#adbf67dcee294e673cf796f1ed8aeb6a4">dst</a>     = <a class="code" href="helpers_8h.xhtml#a31c8c760f08fb1a331b16b7c204321dc">CONVERT_TO_TENSOR3D_STRUCT</a>(dst);</div><div class="line"><a name="l00207"></a><span class="lineno">  207</span>&#160;</div><div class="line"><a name="l00208"></a><span class="lineno">  208</span>&#160;    int8 pixels0 = 0;</div><div class="line"><a name="l00209"></a><span class="lineno">  209</span>&#160;</div><div class="line"><a name="l00210"></a><span class="lineno">  210</span>&#160;    __global uchar *weights_addr = (__global uchar *)<a class="code" href="helpers_8h.xhtml#a2101b2fe0193ce227ae4e0945e321d85">tensor3D_offset</a>(&amp;weights, 0, 0, 0);</div><div class="line"><a name="l00211"></a><span class="lineno">  211</span>&#160;    __global uchar *src_addr     = (__global uchar *)<a class="code" href="helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>(&amp;src, 0, 0);</div><div class="line"><a name="l00212"></a><span class="lineno">  212</span>&#160;</div><div class="line"><a name="l00213"></a><span class="lineno">  213</span>&#160;    <span class="keyword">const</span> <span class="keywordtype">int</span> kernel_index = get_global_id(2);</div><div class="line"><a name="l00214"></a><span class="lineno">  214</span>&#160;    weights_addr += kernel_index * weights_stride_w;</div><div class="line"><a name="l00215"></a><span class="lineno">  215</span>&#160;</div><div class="line"><a name="l00216"></a><span class="lineno">  216</span>&#160;    <span class="keywordflow">for</span>(<span class="keyword">volatile</span> <span class="keywordtype">int</span> d = 0; d &lt; WEIGHTS_DEPTH; ++d)</div><div class="line"><a name="l00217"></a><span class="lineno">  217</span>&#160;    {</div><div class="line"><a name="l00218"></a><span class="lineno">  218</span>&#160;<span class="preprocessor">#if KERNEL_SIZE == 5</span></div><div class="line"><a name="l00219"></a><span class="lineno">  219</span>&#160;        CONVOLUTION1x5(pixels0, (__global uchar *)src_addr, (__global uchar *)weights_addr);</div><div class="line"><a name="l00220"></a><span class="lineno">  220</span>&#160;        CONVOLUTION1x5(pixels0, (__global uchar *)(src_addr + 1 * src_stride_y), (__global uchar *)(weights_addr + 1 * weights_stride_y));</div><div class="line"><a name="l00221"></a><span class="lineno">  221</span>&#160;        CONVOLUTION1x5(pixels0, (__global uchar *)(src_addr + 2 * src_stride_y), (__global uchar *)(weights_addr + 2 * weights_stride_y));</div><div class="line"><a name="l00222"></a><span class="lineno">  222</span>&#160;        CONVOLUTION1x5(pixels0, (__global uchar *)(src_addr + 3 * src_stride_y), (__global uchar *)(weights_addr + 3 * weights_stride_y));</div><div class="line"><a name="l00223"></a><span class="lineno">  223</span>&#160;        CONVOLUTION1x5(pixels0, (__global uchar *)(src_addr + 4 * src_stride_y), (__global uchar *)(weights_addr + 4 * weights_stride_y));</div><div class="line"><a name="l00224"></a><span class="lineno">  224</span>&#160;<span class="preprocessor">#elif KERNEL_SIZE == 3</span></div><div class="line"><a name="l00225"></a><span class="lineno">  225</span>&#160;        CONVOLUTION1x3(pixels0, (__global uchar *)(src_addr + 0 * src_stride_y), (__global uchar *)(weights_addr + 0 * weights_stride_y));</div><div class="line"><a name="l00226"></a><span class="lineno">  226</span>&#160;        CONVOLUTION1x3(pixels0, (__global uchar *)(src_addr + 1 * src_stride_y), (__global uchar *)(weights_addr + 1 * weights_stride_y));</div><div class="line"><a name="l00227"></a><span class="lineno">  227</span>&#160;        CONVOLUTION1x3(pixels0, (__global uchar *)(src_addr + 2 * src_stride_y), (__global uchar *)(weights_addr + 2 * weights_stride_y));</div><div class="line"><a name="l00228"></a><span class="lineno">  228</span>&#160;<span class="preprocessor">#elif KERNEL_SIZE == 1</span></div><div class="line"><a name="l00229"></a><span class="lineno">  229</span>&#160;        <span class="keywordtype">int</span> weight       = convert_int(*(__global uchar *)weights_addr);</div><div class="line"><a name="l00230"></a><span class="lineno">  230</span>&#160;        int8 input_pixel = convert_int8(INPUT_PIXEL((__global uchar *)src_addr));</div><div class="line"><a name="l00231"></a><span class="lineno">  231</span>&#160;        pixels0 += (input_pixel + input_offset) * ((int8)weight + weight_offset);</div><div class="line"><a name="l00232"></a><span class="lineno">  232</span>&#160;<span class="preprocessor">#endif </span><span class="comment">/* (KERNEL_SIZE == 1) || (KERNEL_SIZE == 3) || (KERNEL_SIZE == 5) */</span><span class="preprocessor"></span></div><div class="line"><a name="l00233"></a><span class="lineno">  233</span>&#160;</div><div class="line"><a name="l00234"></a><span class="lineno">  234</span>&#160;        src_addr += src_stride_z;</div><div class="line"><a name="l00235"></a><span class="lineno">  235</span>&#160;        weights_addr += weights_stride_z;</div><div class="line"><a name="l00236"></a><span class="lineno">  236</span>&#160;    }</div><div class="line"><a name="l00237"></a><span class="lineno">  237</span>&#160;</div><div class="line"><a name="l00238"></a><span class="lineno">  238</span>&#160;<span class="preprocessor">#ifdef HAS_BIAS</span></div><div class="line"><a name="l00239"></a><span class="lineno">  239</span>&#160;    <a class="code" href="struct_vector.xhtml">Vector</a>        biases    = <a class="code" href="helpers_8h.xhtml#a64d779f80eeb923e0ab2313433f7b40b">CONVERT_TO_VECTOR_STRUCT_NO_STEP</a>(biases);</div><div class="line"><a name="l00240"></a><span class="lineno">  240</span>&#160;    __global <span class="keywordtype">int</span> *bias_addr = ((__global <span class="keywordtype">int</span> *)(<a class="code" href="helpers_8h.xhtml#a7e4940407322d6f0ccb8b6b86b856019">vector_offset</a>(&amp;biases, kernel_index)));</div><div class="line"><a name="l00241"></a><span class="lineno">  241</span>&#160;    pixels0 += (int8)(*bias_addr);</div><div class="line"><a name="l00242"></a><span class="lineno">  242</span>&#160;<span class="preprocessor">#endif </span><span class="comment">/* defined(HAS_BIAS) */</span><span class="preprocessor"></span></div><div class="line"><a name="l00243"></a><span class="lineno">  243</span>&#160;</div><div class="line"><a name="l00244"></a><span class="lineno">  244</span>&#160;    pixels0 = <a class="code" href="helpers__asymm_8h.xhtml#a86de0ffca367bfcf27a8ae5dd1fdef2d">ASYMM_MULT_BY_QUANT_MULTIPLIER_LESS_THAN_ONE</a>(pixels0, output_multiplier, output_shift, 8);</div><div class="line"><a name="l00245"></a><span class="lineno">  245</span>&#160;    pixels0 = pixels0 + output_offset;</div><div class="line"><a name="l00246"></a><span class="lineno">  246</span>&#160;</div><div class="line"><a name="l00247"></a><span class="lineno">  247</span>&#160;    vstore8(convert_uchar8_sat(pixels0), 0, (__global uchar *)dst.<a class="code" href="struct_tensor3_d.xhtml#acf52c23cbd7424606c10a606524e3e32">ptr</a>);</div><div class="line"><a name="l00248"></a><span class="lineno">  248</span>&#160;}</div><div class="line"><a name="l00249"></a><span class="lineno">  249</span>&#160;<span class="preprocessor">#endif // defined(DATA_TYPE) &amp;&amp; defined(STRIDE_X) &amp;&amp; defined(WEIGHTS_DEPTH)</span></div><div class="line"><a name="l00250"></a><span class="lineno">  250</span>&#160;</div><div class="line"><a name="l00278"></a><span class="lineno"><a class="line" href="direct__convolution__1x1__3x3__5x5__quantized_8cl.xhtml#a176ca8947cee6643fab960588c64e10d">  278</a></span>&#160;__kernel <span class="keywordtype">void</span> <a class="code" href="direct__convolution__1x1__3x3__5x5__quantized_8cl.xhtml#a176ca8947cee6643fab960588c64e10d">output_stage_quantized</a>(</div><div class="line"><a name="l00279"></a><span class="lineno">  279</span>&#160;    <a class="code" href="helpers_8h.xhtml#a6b83038822d1ae7ab619b684ed3b7fc0">TENSOR3D_DECLARATION</a>(src),</div><div class="line"><a name="l00280"></a><span class="lineno">  280</span>&#160;    <a class="code" href="helpers_8h.xhtml#a6b83038822d1ae7ab619b684ed3b7fc0">TENSOR3D_DECLARATION</a>(dst),</div><div class="line"><a name="l00281"></a><span class="lineno">  281</span>&#160;#<span class="keywordflow">if</span> defined(HAS_BIAS)</div><div class="line"><a name="l00282"></a><span class="lineno">  282</span>&#160;    <a class="code" href="helpers_8h.xhtml#a40a6eb9f2a7712f08d6bb8ff6c9e6ca7">VECTOR_DECLARATION</a>(bias),</div><div class="line"><a name="l00283"></a><span class="lineno">  283</span>&#160;#endif <span class="comment">//defined(HAS_BIAS)</span></div><div class="line"><a name="l00284"></a><span class="lineno">  284</span>&#160;    <span class="keywordtype">int</span> output_offset,</div><div class="line"><a name="l00285"></a><span class="lineno">  285</span>&#160;    <span class="keywordtype">int</span> output_multiplier,</div><div class="line"><a name="l00286"></a><span class="lineno">  286</span>&#160;    <span class="keywordtype">int</span> output_shift)</div><div class="line"><a name="l00287"></a><span class="lineno">  287</span>&#160;{</div><div class="line"><a name="l00288"></a><span class="lineno">  288</span>&#160;    <a class="code" href="struct_image.xhtml">Image</a> src = <a class="code" href="helpers_8h.xhtml#a541f8db866a0fa93ee67d58ea31a7d0c">CONVERT_TENSOR3D_TO_IMAGE_STRUCT</a>(src);</div><div class="line"><a name="l00289"></a><span class="lineno">  289</span>&#160;    <a class="code" href="struct_image.xhtml">Image</a> dst = <a class="code" href="helpers_8h.xhtml#a541f8db866a0fa93ee67d58ea31a7d0c">CONVERT_TENSOR3D_TO_IMAGE_STRUCT</a>(dst);</div><div class="line"><a name="l00290"></a><span class="lineno">  290</span>&#160;<span class="preprocessor">#if defined(HAS_BIAS)</span></div><div class="line"><a name="l00291"></a><span class="lineno">  291</span>&#160;    <a class="code" href="struct_vector.xhtml">Vector</a> bias = <a class="code" href="helpers_8h.xhtml#a64d779f80eeb923e0ab2313433f7b40b">CONVERT_TO_VECTOR_STRUCT_NO_STEP</a>(bias);</div><div class="line"><a name="l00292"></a><span class="lineno">  292</span>&#160;<span class="preprocessor">#endif //defined(HAS_BIAS)</span></div><div class="line"><a name="l00293"></a><span class="lineno">  293</span>&#160;</div><div class="line"><a name="l00294"></a><span class="lineno">  294</span>&#160;    <span class="comment">// Load input</span></div><div class="line"><a name="l00295"></a><span class="lineno">  295</span>&#160;    int16 vals = vload16(0, (__global <span class="keywordtype">int</span> *)(src.<a class="code" href="struct_image.xhtml#acf52c23cbd7424606c10a606524e3e32">ptr</a>));</div><div class="line"><a name="l00296"></a><span class="lineno">  296</span>&#160;</div><div class="line"><a name="l00297"></a><span class="lineno">  297</span>&#160;<span class="preprocessor">#if defined(HAS_BIAS)</span></div><div class="line"><a name="l00298"></a><span class="lineno">  298</span>&#160;    <span class="comment">// Load and add bias</span></div><div class="line"><a name="l00299"></a><span class="lineno">  299</span>&#160;    <span class="keywordtype">int</span> bias_value = *((__global <span class="keywordtype">int</span> *)(<a class="code" href="helpers_8h.xhtml#a7e4940407322d6f0ccb8b6b86b856019">vector_offset</a>(&amp;bias, get_global_id(2))));</div><div class="line"><a name="l00300"></a><span class="lineno">  300</span>&#160;    vals += (int16)(bias_value);</div><div class="line"><a name="l00301"></a><span class="lineno">  301</span>&#160;<span class="preprocessor">#endif //defined(HAS_BIAS)</span></div><div class="line"><a name="l00302"></a><span class="lineno">  302</span>&#160;</div><div class="line"><a name="l00303"></a><span class="lineno">  303</span>&#160;    vals = <a class="code" href="helpers__asymm_8h.xhtml#a86de0ffca367bfcf27a8ae5dd1fdef2d">ASYMM_MULT_BY_QUANT_MULTIPLIER_LESS_THAN_ONE</a>(vals, output_multiplier, output_shift, 16);</div><div class="line"><a name="l00304"></a><span class="lineno">  304</span>&#160;    vals = vals + output_offset;</div><div class="line"><a name="l00305"></a><span class="lineno">  305</span>&#160;</div><div class="line"><a name="l00306"></a><span class="lineno">  306</span>&#160;    <span class="comment">// Store result in dst</span></div><div class="line"><a name="l00307"></a><span class="lineno">  307</span>&#160;    vstore16(convert_uchar16_sat(vals), 0, (__global uchar *)dst.<a class="code" href="struct_image.xhtml#acf52c23cbd7424606c10a606524e3e32">ptr</a>);</div><div class="line"><a name="l00308"></a><span class="lineno">  308</span>&#160;}</div><div class="ttc" id="struct_vector_xhtml"><div class="ttname"><a href="struct_vector.xhtml">Vector</a></div><div class="ttdoc">Structure to hold Vector information. </div><div class="ttdef"><b>Definition:</b> <a href="helpers_8h_source.xhtml#l00132">helpers.h:132</a></div></div>
 121 <div class="ttc" id="helpers_8h_xhtml_a31c8c760f08fb1a331b16b7c204321dc"><div class="ttname"><a href="helpers_8h.xhtml#a31c8c760f08fb1a331b16b7c204321dc">CONVERT_TO_TENSOR3D_STRUCT</a></div><div class="ttdeci">#define CONVERT_TO_TENSOR3D_STRUCT(name)</div><div class="ttdef"><b>Definition:</b> <a href="helpers_8h_source.xhtml#l00117">helpers.h:117</a></div></div>
 122 <div class="ttc" id="direct__convolution__1x1__3x3__5x5__quantized_8cl_xhtml_a176ca8947cee6643fab960588c64e10d"><div class="ttname"><a href="direct__convolution__1x1__3x3__5x5__quantized_8cl.xhtml#a176ca8947cee6643fab960588c64e10d">output_stage_quantized</a></div><div class="ttdeci">__kernel void output_stage_quantized(__global uchar *src_ptr, uint src_stride_x, uint src_step_x, uint src_stride_y, uint src_step_y, uint src_stride_z, uint src_step_z, uint src_offset_first_element_in_bytes, __global uchar *dst_ptr, uint dst_stride_x, uint dst_step_x, uint dst_stride_y, uint dst_step_y, uint dst_stride_z, uint dst_step_z, uint dst_offset_first_element_in_bytes, __global uchar *bias_ptr, uint bias_stride_x, uint bias_step_x, uint bias_offset_first_element_in_bytes, int output_offset, int output_multiplier, int output_shift)</div><div class="ttdoc">This function computes the output stage of a depthwise convolution. </div><div class="ttdef"><b>Definition:</b> <a href="direct__convolution__1x1__3x3__5x5__quantized_8cl_source.xhtml#l00278">direct_convolution_1x1_3x3_5x5_quantized.cl:278</a></div></div>
 123 <div class="ttc" id="struct_tensor3_d_xhtml"><div class="ttname"><a href="struct_tensor3_d.xhtml">Tensor3D</a></div><div class="ttdoc">Structure to hold 3D tensor information. </div><div class="ttdef"><b>Definition:</b> <a href="helpers_8h_source.xhtml#l00149">helpers.h:149</a></div></div>
 124 <div class="ttc" id="helpers_8h_xhtml_a009469e4d9b8fce3b6d5e97d2077827d"><div class="ttname"><a href="helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a></div><div class="ttdeci">__global uchar * offset(const Image *img, int x, int y)</div><div class="ttdoc">Get the pointer position of a Image. </div><div class="ttdef"><b>Definition:</b> <a href="helpers_8h_source.xhtml#l00301">helpers.h:301</a></div></div>
 125 <div class="ttc" id="helpers_8h_xhtml_aebe814363556c244be043b13e7969197"><div class="ttname"><a href="helpers_8h.xhtml#aebe814363556c244be043b13e7969197">CONVERT_TO_IMAGE_STRUCT</a></div><div class="ttdeci">#define CONVERT_TO_IMAGE_STRUCT(name)</div><div class="ttdef"><b>Definition:</b> <a href="helpers_8h_source.xhtml#l00102">helpers.h:102</a></div></div>
 126 <div class="ttc" id="helpers_8h_xhtml_a2101b2fe0193ce227ae4e0945e321d85"><div class="ttname"><a href="helpers_8h.xhtml#a2101b2fe0193ce227ae4e0945e321d85">tensor3D_offset</a></div><div class="ttdeci">__global const uchar * tensor3D_offset(const Tensor3D *tensor, int x, int y, int z)</div><div class="ttdoc">Get the pointer position of a Tensor3D. </div><div class="ttdef"><b>Definition:</b> <a href="helpers_8h_source.xhtml#l00313">helpers.h:313</a></div></div>
 127 <div class="ttc" id="helpers_8h_xhtml_a40a6eb9f2a7712f08d6bb8ff6c9e6ca7"><div class="ttname"><a href="helpers_8h.xhtml#a40a6eb9f2a7712f08d6bb8ff6c9e6ca7">VECTOR_DECLARATION</a></div><div class="ttdeci">#define VECTOR_DECLARATION(name)</div><div class="ttdef"><b>Definition:</b> <a href="helpers_8h_source.xhtml#l00060">helpers.h:60</a></div></div>
 128 <div class="ttc" id="namespacearm__compute_1_1test_1_1validation_xhtml_adbf67dcee294e673cf796f1ed8aeb6a4"><div class="ttname"><a href="namespacearm__compute_1_1test_1_1validation.xhtml#adbf67dcee294e673cf796f1ed8aeb6a4">arm_compute::test::validation::dst</a></div><div class="ttdeci">CLTensor dst</div><div class="ttdef"><b>Definition:</b> <a href="_c_l_2_convolution_8cpp_source.xhtml#l00137">Convolution.cpp:137</a></div></div>
 129 <div class="ttc" id="struct_image_xhtml"><div class="ttname"><a href="struct_image.xhtml">Image</a></div><div class="ttdoc">Structure to hold Image information. </div><div class="ttdef"><b>Definition:</b> <a href="helpers_8h_source.xhtml#l00140">helpers.h:140</a></div></div>
 130 <div class="ttc" id="helpers_8h_xhtml_a6b83038822d1ae7ab619b684ed3b7fc0"><div class="ttname"><a href="helpers_8h.xhtml#a6b83038822d1ae7ab619b684ed3b7fc0">TENSOR3D_DECLARATION</a></div><div class="ttdeci">#define TENSOR3D_DECLARATION(name)</div><div class="ttdef"><b>Definition:</b> <a href="helpers_8h_source.xhtml#l00074">helpers.h:74</a></div></div>
 131 <div class="ttc" id="struct_image_xhtml_acf52c23cbd7424606c10a606524e3e32"><div class="ttname"><a href="struct_image.xhtml#acf52c23cbd7424606c10a606524e3e32">Image::ptr</a></div><div class="ttdeci">__global uchar * ptr</div><div class="ttdoc">Pointer to the starting postion of the buffer. </div><div class="ttdef"><b>Definition:</b> <a href="helpers_8h_source.xhtml#l00142">helpers.h:142</a></div></div>
 132 <div class="ttc" id="helpers__asymm_8h_xhtml"><div class="ttname"><a href="helpers__asymm_8h.xhtml">helpers_asymm.h</a></div></div>
 133 <div class="ttc" id="helpers_8h_xhtml_a79e8e562daa6599317d2d1cd86ef1bf2"><div class="ttname"><a href="helpers_8h.xhtml#a79e8e562daa6599317d2d1cd86ef1bf2">CONVERT_TO_TENSOR3D_STRUCT_NO_STEP</a></div><div class="ttdeci">#define CONVERT_TO_TENSOR3D_STRUCT_NO_STEP(name)</div><div class="ttdef"><b>Definition:</b> <a href="helpers_8h_source.xhtml#l00121">helpers.h:121</a></div></div>
 134 <div class="ttc" id="helpers_8h_xhtml_a64d779f80eeb923e0ab2313433f7b40b"><div class="ttname"><a href="helpers_8h.xhtml#a64d779f80eeb923e0ab2313433f7b40b">CONVERT_TO_VECTOR_STRUCT_NO_STEP</a></div><div class="ttdeci">#define CONVERT_TO_VECTOR_STRUCT_NO_STEP(name)</div><div class="ttdef"><b>Definition:</b> <a href="helpers_8h_source.xhtml#l00099">helpers.h:99</a></div></div>
 135 <div class="ttc" id="helpers_8h_xhtml_a7e4940407322d6f0ccb8b6b86b856019"><div class="ttname"><a href="helpers_8h.xhtml#a7e4940407322d6f0ccb8b6b86b856019">vector_offset</a></div><div class="ttdeci">__global const uchar * vector_offset(const Vector *vec, int x)</div><div class="ttdoc">Get the pointer position of a Vector. </div><div class="ttdef"><b>Definition:</b> <a href="helpers_8h_source.xhtml#l00290">helpers.h:290</a></div></div>
 136 <div class="ttc" id="helpers__asymm_8h_xhtml_a86de0ffca367bfcf27a8ae5dd1fdef2d"><div class="ttname"><a href="helpers__asymm_8h.xhtml#a86de0ffca367bfcf27a8ae5dd1fdef2d">ASYMM_MULT_BY_QUANT_MULTIPLIER_LESS_THAN_ONE</a></div><div class="ttdeci">#define ASYMM_MULT_BY_QUANT_MULTIPLIER_LESS_THAN_ONE(x, quantized_multiplier, right_shift, size)</div><div class="ttdef"><b>Definition:</b> <a href="helpers__asymm_8h_source.xhtml#l00296">helpers_asymm.h:296</a></div></div>
 137 <div class="ttc" id="helpers_8h_xhtml_a541f8db866a0fa93ee67d58ea31a7d0c"><div class="ttname"><a href="helpers_8h.xhtml#a541f8db866a0fa93ee67d58ea31a7d0c">CONVERT_TENSOR3D_TO_IMAGE_STRUCT</a></div><div class="ttdeci">#define CONVERT_TENSOR3D_TO_IMAGE_STRUCT(name)</div><div class="ttdef"><b>Definition:</b> <a href="helpers_8h_source.xhtml#l00114">helpers.h:114</a></div></div>
 138 <div class="ttc" id="struct_tensor3_d_xhtml_acf52c23cbd7424606c10a606524e3e32"><div class="ttname"><a href="struct_tensor3_d.xhtml#acf52c23cbd7424606c10a606524e3e32">Tensor3D::ptr</a></div><div class="ttdeci">__global uchar * ptr</div><div class="ttdoc">Pointer to the starting postion of the buffer. </div><div class="ttdef"><b>Definition:</b> <a href="helpers_8h_source.xhtml#l00151">helpers.h:151</a></div></div>
 139 <div class="ttc" id="namespacearm__compute_1_1test_1_1validation_xhtml_a6743f0a130e8311e6f5b1a23df102472"><div class="ttname"><a href="namespacearm__compute_1_1test_1_1validation.xhtml#a6743f0a130e8311e6f5b1a23df102472">arm_compute::test::validation::src</a></div><div class="ttdeci">convolution configure &amp; src</div><div class="ttdef"><b>Definition:</b> <a href="_c_l_2_convolution_8cpp_source.xhtml#l00147">Convolution.cpp:147</a></div></div>
 140 </div><!-- fragment --></div><!-- contents -->
 141 </div><!-- doc-content -->
 142 <!-- start footer part -->
 143 <div id="nav-path" class="navpath"><!-- id is needed for treeview function! -->
 144   <ul>
 145     <li class="navelem"><a class="el" href="dir_68267d1309a1af8e8297ef4c3efbcdba.xhtml">src</a></li><li class="navelem"><a class="el" href="dir_aebb8dcc11953d78e620bbef0b9e2183.xhtml">core</a></li><li class="navelem"><a class="el" href="dir_8c278f79c760e5c5fbd911f9870614c1.xhtml">CL</a></li><li class="navelem"><a class="el" href="dir_25885286e9dad4fa105b7b25a8031bbf.xhtml">cl_kernels</a></li><li class="navelem"><a class="el" href="direct__convolution__1x1__3x3__5x5__quantized_8cl.xhtml">direct_convolution_1x1_3x3_5x5_quantized.cl</a></li>
 146     <li class="footer">Generated on Thu Feb 22 2018 15:45:19 for Compute Library by
 147     <a href="http://www.doxygen.org/index.html">
 148     <img class="footer" src="doxygen.png" alt="doxygen"/></a> 1.8.11 </li>
 149   </ul>
 150 </div>
 151 </body>
 152 </html>