Google Git
Sign in
android / platform / external / ComputeLibrary / 975dfe175 / . / documentation / gemmlowp_8cl_source.xhtml
blob: c84bdcfa74a05eded6dde70f5814853b675b5b75 [file] [log] [blame]
<!-- HTML header for doxygen 1.8.15-->
<!-- Remember to use version doxygen 1.8.15 +-->
<!DOCTYPE html PUBLIC "-//W3C//DTD XHTML 1.0 Transitional//EN" "http://www.w3.org/TR/xhtml1/DTD/xhtml1-transitional.dtd">
<html xmlns="http://www.w3.org/1999/xhtml">
<head>
<meta http-equiv="Content-Type" content="text/xhtml;charset=UTF-8"/>
<meta http-equiv="X-UA-Compatible" content="IE=9"/>
<meta name="generator" content="Doxygen 1.8.15"/>
<meta name="robots" content="NOINDEX, NOFOLLOW" /> <!-- Prevent indexing by search engines -->
<title>Compute Library: src/core/CL/cl_kernels/gemmlowp.cl Source File</title>
<link href="tabs.css" rel="stylesheet" type="text/css"/>
<script type="text/javascript" src="jquery.js"></script>
<script type="text/javascript" src="dynsections.js"></script>
<link href="navtree.css" rel="stylesheet" type="text/css"/>
<script type="text/javascript" src="resize.js"></script>
<script type="text/javascript" src="navtreedata.js"></script>
<script type="text/javascript" src="navtree.js"></script>
<script type="text/javascript">
/* @license magnet:?xt=urn:btih:cf05388f2679ee054f2beb29a391d25f4e673ac3&amp;dn=gpl-2.0.txt GPL-v2 */
$(document).ready(initResizable);
/* @license-end */</script>
<link href="search/search.css" rel="stylesheet" type="text/css"/>
<script type="text/javascript" src="search/searchdata.js"></script>
<script type="text/javascript" src="search/search.js"></script>
<script type="text/x-mathjax-config">
MathJax.Hub.Config({
extensions: ["tex2jax.js"],
jax: ["input/TeX","output/HTML-CSS"],
});
</script><script type="text/javascript" async="async" src="http://cdn.mathjax.org/mathjax/latest/MathJax.js"></script>
<link href="doxygen.css" rel="stylesheet" type="text/css" />
<link href="stylesheet.css" rel="stylesheet" type="text/css"/>
</head>
<body>
<div id="top"><!-- do not remove this div, it is closed by doxygen! -->
<div id="titlearea">
<table cellspacing="0" cellpadding="0">
<tbody>
<tr style="height: 56px;">
<img alt="Compute Library" src="https://raw.githubusercontent.com/ARM-software/ComputeLibrary/gh-pages/ACL_logo.png" style="max-width: 100%;margin-top: 15px;margin-left: 10px"/>
<td style="padding-left: 0.5em;">
<div id="projectname">
&#160;<span id="projectnumber">19.08</span>
</div>
</td>
</tr>
</tbody>
</table>
</div>
<!-- end header part -->
<!-- Generated by Doxygen 1.8.15 -->
<script type="text/javascript">
/* @license magnet:?xt=urn:btih:cf05388f2679ee054f2beb29a391d25f4e673ac3&amp;dn=gpl-2.0.txt GPL-v2 */
var searchBox = new SearchBox("searchBox", "search",false,'Search');
/* @license-end */
</script>
<script type="text/javascript" src="menudata.js"></script>
<script type="text/javascript" src="menu.js"></script>
<script type="text/javascript">
/* @license magnet:?xt=urn:btih:cf05388f2679ee054f2beb29a391d25f4e673ac3&amp;dn=gpl-2.0.txt GPL-v2 */
$(function() {
initMenu('',true,false,'search.php','Search');
$(document).ready(function() { init_search(); });
});
/* @license-end */</script>
<div id="main-nav"></div>
</div><!-- top -->
<div id="side-nav" class="ui-resizable side-nav-resizable">
<div id="nav-tree">
<div id="nav-tree-contents">
<div id="nav-sync" class="sync"></div>
</div>
</div>
<div id="splitbar" style="-moz-user-select:none;"
class="ui-resizable-handle">
</div>
</div>
<script type="text/javascript">
/* @license magnet:?xt=urn:btih:cf05388f2679ee054f2beb29a391d25f4e673ac3&amp;dn=gpl-2.0.txt GPL-v2 */
$(document).ready(function(){initNavTree('gemmlowp_8cl_source.xhtml','');});
/* @license-end */
</script>
<div id="doc-content">
<!-- window showing the filter options -->
<div id="MSearchSelectWindow"
onmouseover="return searchBox.OnSearchSelectShow()"
onmouseout="return searchBox.OnSearchSelectHide()"
onkeydown="return searchBox.OnSearchSelectKey(event)">
</div>
<!-- iframe showing the search results (closed by default) -->
<div id="MSearchResultsWindow">
<iframe src="javascript:void(0)" frameborder="0"
name="MSearchResults" id="MSearchResults">
</iframe>
</div>
<div class="header">
<div class="headertitle">
<div class="title">gemmlowp.cl</div> </div>
</div><!--header-->
<div class="contents">
<a href="gemmlowp_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) 2017-2019 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="gemm__helpers_8h.xhtml">gemm_helpers.h</a>&quot;</span></div><div class="line"><a name="l00025"></a><span class="lineno"> 25</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="l00026"></a><span class="lineno"> 26</span>&#160;<span class="preprocessor">#include &quot;<a class="code" href="repeat_8h.xhtml">repeat.h</a>&quot;</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(ARM_COMPUTE_OPENCL_DOT8_ENABLED) &amp;&amp; defined(cl_arm_integer_dot_product_int8)</span></div><div class="line"><a name="l00029"></a><span class="lineno"> 29</span>&#160;<span class="preprocessor">#if defined(ARM_COMPUTE_OPENCL_DOT8_ACC_ENABLED) &amp;&amp; defined(cl_arm_integer_dot_product_accumulate_int8)</span></div><div class="line"><a name="l00030"></a><span class="lineno"> 30</span>&#160;<span class="preprocessor">#define ARM_DOT(x, y, val) val = arm_dot_acc((x), (y), (val));</span></div><div class="line"><a name="l00031"></a><span class="lineno"> 31</span>&#160;<span class="preprocessor">#else // defined(ARM_COMPUTE_OPENCL_DOT8_ACC_ENABLED) &amp;&amp; defined(cl_arm_integer_dot_product_accumulate_int8)</span></div><div class="line"><a name="l00032"></a><span class="lineno"> 32</span>&#160;<span class="preprocessor">#define ARM_DOT(x, y, val) val += arm_dot((x), (y));</span></div><div class="line"><a name="l00033"></a><span class="lineno"> 33</span>&#160;<span class="preprocessor">#endif // defined(ARM_COMPUTE_OPENCL_DOT8_ACC_ENABLED) &amp;&amp; defined(cl_arm_integer_dot_product_accumulate_int8)</span></div><div class="line"><a name="l00034"></a><span class="lineno"> 34</span>&#160;<span class="preprocessor">#endif // defined(ARM_COMPUTE_OPENCL_DOT8_ENABLED) &amp;&amp; defined(cl_arm_integer_dot_product_int8)</span></div><div class="line"><a name="l00035"></a><span class="lineno"> 35</span>&#160;</div><div class="line"><a name="l00036"></a><span class="lineno"> 36</span>&#160;<span class="preprocessor">#if defined(ARM_COMPUTE_OPENCL_DOT8_ENABLED) &amp;&amp; defined(cl_arm_integer_dot_product_int8)</span></div><div class="line"><a name="l00037"></a><span class="lineno"> 37</span>&#160;</div><div class="line"><a name="l00039"></a><span class="lineno"> 39</span>&#160;<span class="preprocessor">#define ARM_DOT1(a, b, c) \</span></div><div class="line"><a name="l00040"></a><span class="lineno"> 40</span>&#160;<span class="preprocessor"> ({ \</span></div><div class="line"><a name="l00041"></a><span class="lineno"> 41</span>&#160;<span class="preprocessor"> ARM_DOT((uchar4)(a, (uchar3)0), (uchar4)(b, (uchar3)0), c); \</span></div><div class="line"><a name="l00042"></a><span class="lineno"> 42</span>&#160;<span class="preprocessor"> })</span></div><div class="line"><a name="l00043"></a><span class="lineno"> 43</span>&#160;<span class="preprocessor">#define ARM_DOT2(a, b, c) \</span></div><div class="line"><a name="l00044"></a><span class="lineno"> 44</span>&#160;<span class="preprocessor"> ({ \</span></div><div class="line"><a name="l00045"></a><span class="lineno"> 45</span>&#160;<span class="preprocessor"> ARM_DOT((uchar4)(a, (uchar2)0), (uchar4)(b, (uchar2)0), c); \</span></div><div class="line"><a name="l00046"></a><span class="lineno"> 46</span>&#160;<span class="preprocessor"> })</span></div><div class="line"><a name="l00047"></a><span class="lineno"> 47</span>&#160;<span class="preprocessor">#define ARM_DOT3(a, b, c) \</span></div><div class="line"><a name="l00048"></a><span class="lineno"> 48</span>&#160;<span class="preprocessor"> ({ \</span></div><div class="line"><a name="l00049"></a><span class="lineno"> 49</span>&#160;<span class="preprocessor"> ARM_DOT((uchar4)(a, (uchar)0), (uchar4)(b, (uchar)0), c); \</span></div><div class="line"><a name="l00050"></a><span class="lineno"> 50</span>&#160;<span class="preprocessor"> })</span></div><div class="line"><a name="l00051"></a><span class="lineno"> 51</span>&#160;<span class="preprocessor">#define ARM_DOT4(a, b, c) \</span></div><div class="line"><a name="l00052"></a><span class="lineno"> 52</span>&#160;<span class="preprocessor"> ({ \</span></div><div class="line"><a name="l00053"></a><span class="lineno"> 53</span>&#160;<span class="preprocessor"> ARM_DOT(a, b, c); \</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">#define ARM_DOT8(a, b, c) \</span></div><div class="line"><a name="l00056"></a><span class="lineno"> 56</span>&#160;<span class="preprocessor"> ({ \</span></div><div class="line"><a name="l00057"></a><span class="lineno"> 57</span>&#160;<span class="preprocessor"> ARM_DOT4((a.lo), (b.lo), c); \</span></div><div class="line"><a name="l00058"></a><span class="lineno"> 58</span>&#160;<span class="preprocessor"> ARM_DOT4((a.hi), (b.hi), c); \</span></div><div class="line"><a name="l00059"></a><span class="lineno"> 59</span>&#160;<span class="preprocessor"> })</span></div><div class="line"><a name="l00060"></a><span class="lineno"> 60</span>&#160;<span class="preprocessor">#define ARM_DOT16(a, b, c) \</span></div><div class="line"><a name="l00061"></a><span class="lineno"> 61</span>&#160;<span class="preprocessor"> ({ \</span></div><div class="line"><a name="l00062"></a><span class="lineno"> 62</span>&#160;<span class="preprocessor"> ARM_DOT8((a.lo), (b.lo), c); \</span></div><div class="line"><a name="l00063"></a><span class="lineno"> 63</span>&#160;<span class="preprocessor"> ARM_DOT8((a.hi), (b.hi), c); \</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">#else // defined(ARM_COMPUTE_OPENCL_DOT8_ENABLED) &amp;&amp; defined(cl_arm_integer_dot_product_int8)</span></div><div class="line"><a name="l00067"></a><span class="lineno"> 67</span>&#160;</div><div class="line"><a name="l00069"></a><span class="lineno"><a class="line" href="gemmlowp_8cl.xhtml#af6f3e915f4b9b0e76dad9ea458a17ba6"> 69</a></span>&#160;<span class="preprocessor">#define ARM_DOT1(a, b, c) \</span></div><div class="line"><a name="l00070"></a><span class="lineno"> 70</span>&#160;<span class="preprocessor"> ({ \</span></div><div class="line"><a name="l00071"></a><span class="lineno"> 71</span>&#160;<span class="preprocessor"> c += (uint)a * b; \</span></div><div class="line"><a name="l00072"></a><span class="lineno"> 72</span>&#160;<span class="preprocessor"> })</span></div><div class="line"><a name="l00073"></a><span class="lineno"><a class="line" href="gemmlowp_8cl.xhtml#ae34a7418dab77ded3f299ad8e01dc545"> 73</a></span>&#160;<span class="preprocessor">#define ARM_DOT2(a, b, c) \</span></div><div class="line"><a name="l00074"></a><span class="lineno"> 74</span>&#160;<span class="preprocessor"> ({ \</span></div><div class="line"><a name="l00075"></a><span class="lineno"> 75</span>&#160;<span class="preprocessor"> c += (uint)a.s0 * b.s0; \</span></div><div class="line"><a name="l00076"></a><span class="lineno"> 76</span>&#160;<span class="preprocessor"> c += (uint)a.s1 * b.s1; \</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"><a class="line" href="gemmlowp_8cl.xhtml#aa1fc470aae7d425cb48b4eddd923c1f1"> 78</a></span>&#160;<span class="preprocessor">#define ARM_DOT3(a, b, c) \</span></div><div class="line"><a name="l00079"></a><span class="lineno"> 79</span>&#160;<span class="preprocessor"> ({ \</span></div><div class="line"><a name="l00080"></a><span class="lineno"> 80</span>&#160;<span class="preprocessor"> ARM_DOT2(a, b, c); \</span></div><div class="line"><a name="l00081"></a><span class="lineno"> 81</span>&#160;<span class="preprocessor"> c += (uint)a.s2 * b.s2; \</span></div><div class="line"><a name="l00082"></a><span class="lineno"> 82</span>&#160;<span class="preprocessor"> })</span></div><div class="line"><a name="l00083"></a><span class="lineno"><a class="line" href="gemmlowp_8cl.xhtml#a691532f1c550b718a64219c0bf8d1aea"> 83</a></span>&#160;<span class="preprocessor">#define ARM_DOT4(a, b, c) \</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;<span class="preprocessor"> ARM_DOT3(a, b, c); \</span></div><div class="line"><a name="l00086"></a><span class="lineno"> 86</span>&#160;<span class="preprocessor"> c += (uint)a.s3 * b.s3; \</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"><a class="line" href="gemmlowp_8cl.xhtml#aa6675b3d030e11613b7af2f591a3215f"> 88</a></span>&#160;<span class="preprocessor">#define ARM_DOT8(a, b, c) \</span></div><div class="line"><a name="l00089"></a><span class="lineno"> 89</span>&#160;<span class="preprocessor"> ({ \</span></div><div class="line"><a name="l00090"></a><span class="lineno"> 90</span>&#160;<span class="preprocessor"> ARM_DOT4((a.lo), (b.lo), c); \</span></div><div class="line"><a name="l00091"></a><span class="lineno"> 91</span>&#160;<span class="preprocessor"> ARM_DOT4((a.hi), (b.hi), c); \</span></div><div class="line"><a name="l00092"></a><span class="lineno"> 92</span>&#160;<span class="preprocessor"> })</span></div><div class="line"><a name="l00093"></a><span class="lineno"><a class="line" href="gemmlowp_8cl.xhtml#a82a34a6b94dcdb93e446287e1a3e630d"> 93</a></span>&#160;<span class="preprocessor">#define ARM_DOT16(a, b, c) \</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;<span class="preprocessor"> ARM_DOT8((a.lo), (b.lo), c); \</span></div><div class="line"><a name="l00096"></a><span class="lineno"> 96</span>&#160;<span class="preprocessor"> ARM_DOT8((a.hi), (b.hi), c); \</span></div><div class="line"><a name="l00097"></a><span class="lineno"> 97</span>&#160;<span class="preprocessor"> })</span></div><div class="line"><a name="l00098"></a><span class="lineno"> 98</span>&#160;<span class="preprocessor">#endif // defined(ARM_COMPUTE_OPENCL_DOT8_ENABLED) &amp;&amp; defined(cl_arm_integer_dot_product_int8)</span></div><div class="line"><a name="l00099"></a><span class="lineno"> 99</span>&#160;</div><div class="line"><a name="l00101"></a><span class="lineno"><a class="line" href="gemmlowp_8cl.xhtml#ae1635754535ac499cdad8b067e90c435"> 101</a></span>&#160;<span class="preprocessor">#define ARM_DOT_K0X2(k0, a, b, c) \</span></div><div class="line"><a name="l00102"></a><span class="lineno"> 102</span>&#160;<span class="preprocessor"> ({ \</span></div><div class="line"><a name="l00103"></a><span class="lineno"> 103</span>&#160;<span class="preprocessor"> ARM_DOT_K0(k0, (a), (b##0), (c.s0)); \</span></div><div class="line"><a name="l00104"></a><span class="lineno"> 104</span>&#160;<span class="preprocessor"> ARM_DOT_K0(k0, (a), (b##1), (c.s1)); \</span></div><div class="line"><a name="l00105"></a><span class="lineno"> 105</span>&#160;<span class="preprocessor"> })</span></div><div class="line"><a name="l00106"></a><span class="lineno"><a class="line" href="gemmlowp_8cl.xhtml#a204079a2adf17e607e9cfcad60b89602"> 106</a></span>&#160;<span class="preprocessor">#define ARM_DOT_K0X3(k0, a, b, c) \</span></div><div class="line"><a name="l00107"></a><span class="lineno"> 107</span>&#160;<span class="preprocessor"> ({ \</span></div><div class="line"><a name="l00108"></a><span class="lineno"> 108</span>&#160;<span class="preprocessor"> ARM_DOT_K0X2(k0, a, b, c); \</span></div><div class="line"><a name="l00109"></a><span class="lineno"> 109</span>&#160;<span class="preprocessor"> ARM_DOT_K0(k0, (a), (b##2), (c.s2)); \</span></div><div class="line"><a name="l00110"></a><span class="lineno"> 110</span>&#160;<span class="preprocessor"> })</span></div><div class="line"><a name="l00111"></a><span class="lineno"><a class="line" href="gemmlowp_8cl.xhtml#a95f17ea569b0d658d161981c656ce028"> 111</a></span>&#160;<span class="preprocessor">#define ARM_DOT_K0X4(k0, a, b, c) \</span></div><div class="line"><a name="l00112"></a><span class="lineno"> 112</span>&#160;<span class="preprocessor"> ({ \</span></div><div class="line"><a name="l00113"></a><span class="lineno"> 113</span>&#160;<span class="preprocessor"> ARM_DOT_K0X3(k0, a, b, c); \</span></div><div class="line"><a name="l00114"></a><span class="lineno"> 114</span>&#160;<span class="preprocessor"> ARM_DOT_K0(k0, (a), (b##3), (c.s3)); \</span></div><div class="line"><a name="l00115"></a><span class="lineno"> 115</span>&#160;<span class="preprocessor"> })</span></div><div class="line"><a name="l00116"></a><span class="lineno"><a class="line" href="gemmlowp_8cl.xhtml#a9fe47ceb84fabdb9ecc8b96c3104e8c7"> 116</a></span>&#160;<span class="preprocessor">#define ARM_DOT_K0X8(k0, a, b, c) \</span></div><div class="line"><a name="l00117"></a><span class="lineno"> 117</span>&#160;<span class="preprocessor"> ({ \</span></div><div class="line"><a name="l00118"></a><span class="lineno"> 118</span>&#160;<span class="preprocessor"> ARM_DOT_K0X4(k0, a, b, c); \</span></div><div class="line"><a name="l00119"></a><span class="lineno"> 119</span>&#160;<span class="preprocessor"> ARM_DOT_K0(k0, (a), (b##4), (c.s4)); \</span></div><div class="line"><a name="l00120"></a><span class="lineno"> 120</span>&#160;<span class="preprocessor"> ARM_DOT_K0(k0, (a), (b##5), (c.s5)); \</span></div><div class="line"><a name="l00121"></a><span class="lineno"> 121</span>&#160;<span class="preprocessor"> ARM_DOT_K0(k0, (a), (b##6), (c.s6)); \</span></div><div class="line"><a name="l00122"></a><span class="lineno"> 122</span>&#160;<span class="preprocessor"> ARM_DOT_K0(k0, (a), (b##7), (c.s7)); \</span></div><div class="line"><a name="l00123"></a><span class="lineno"> 123</span>&#160;<span class="preprocessor"> })</span></div><div class="line"><a name="l00124"></a><span class="lineno"><a class="line" href="gemmlowp_8cl.xhtml#a9a4dd8e50aea16bfbdd001ae160c52f1"> 124</a></span>&#160;<span class="preprocessor">#define ARM_DOT_K0X16(k0, a, b, c) \</span></div><div class="line"><a name="l00125"></a><span class="lineno"> 125</span>&#160;<span class="preprocessor"> ({ \</span></div><div class="line"><a name="l00126"></a><span class="lineno"> 126</span>&#160;<span class="preprocessor"> ARM_DOT_K0X8(k0, a, b, c); \</span></div><div class="line"><a name="l00127"></a><span class="lineno"> 127</span>&#160;<span class="preprocessor"> ARM_DOT_K0(k0, (a), (b##8), (c.s8)); \</span></div><div class="line"><a name="l00128"></a><span class="lineno"> 128</span>&#160;<span class="preprocessor"> ARM_DOT_K0(k0, (a), (b##9), (c.s9)); \</span></div><div class="line"><a name="l00129"></a><span class="lineno"> 129</span>&#160;<span class="preprocessor"> ARM_DOT_K0(k0, (a), (b##A), (c.sA)); \</span></div><div class="line"><a name="l00130"></a><span class="lineno"> 130</span>&#160;<span class="preprocessor"> ARM_DOT_K0(k0, (a), (b##B), (c.sB)); \</span></div><div class="line"><a name="l00131"></a><span class="lineno"> 131</span>&#160;<span class="preprocessor"> ARM_DOT_K0(k0, (a), (b##C), (c.sC)); \</span></div><div class="line"><a name="l00132"></a><span class="lineno"> 132</span>&#160;<span class="preprocessor"> ARM_DOT_K0(k0, (a), (b##D), (c.sD)); \</span></div><div class="line"><a name="l00133"></a><span class="lineno"> 133</span>&#160;<span class="preprocessor"> ARM_DOT_K0(k0, (a), (b##E), (c.sE)); \</span></div><div class="line"><a name="l00134"></a><span class="lineno"> 134</span>&#160;<span class="preprocessor"> ARM_DOT_K0(k0, (a), (b##F), (c.sF)); \</span></div><div class="line"><a name="l00135"></a><span class="lineno"> 135</span>&#160;<span class="preprocessor"> })</span></div><div class="line"><a name="l00136"></a><span class="lineno"> 136</span>&#160;</div><div class="line"><a name="l00138"></a><span class="lineno"><a class="line" href="gemmlowp_8cl.xhtml#a6a30e0c9210948b09a983dff1b3db289"> 138</a></span>&#160;<span class="preprocessor">#define ARM_MM_K0XN0X1(n0, k0, a, b, c) \</span></div><div class="line"><a name="l00139"></a><span class="lineno"> 139</span>&#160;<span class="preprocessor"> ({ \</span></div><div class="line"><a name="l00140"></a><span class="lineno"> 140</span>&#160;<span class="preprocessor"> ARM_DOT_K0XN0(n0, k0, (a##0), b, (c##0)); \</span></div><div class="line"><a name="l00141"></a><span class="lineno"> 141</span>&#160;<span class="preprocessor"> })</span></div><div class="line"><a name="l00142"></a><span class="lineno"><a class="line" href="gemmlowp_8cl.xhtml#ad2c6dbc81440272f6ef8f19c7a5e7843"> 142</a></span>&#160;<span class="preprocessor">#define ARM_MM_K0XN0X2(n0, k0, a, b, c) \</span></div><div class="line"><a name="l00143"></a><span class="lineno"> 143</span>&#160;<span class="preprocessor"> ({ \</span></div><div class="line"><a name="l00144"></a><span class="lineno"> 144</span>&#160;<span class="preprocessor"> ARM_MM_K0XN0X1(n0, k0, a, b, c); \</span></div><div class="line"><a name="l00145"></a><span class="lineno"> 145</span>&#160;<span class="preprocessor"> ARM_DOT_K0XN0(n0, k0, (a##1), b, (c##1)); \</span></div><div class="line"><a name="l00146"></a><span class="lineno"> 146</span>&#160;<span class="preprocessor"> })</span></div><div class="line"><a name="l00147"></a><span class="lineno"><a class="line" href="gemmlowp_8cl.xhtml#a82ee8878109026b86a3296e786248d15"> 147</a></span>&#160;<span class="preprocessor">#define ARM_MM_K0XN0X3(n0, k0, a, b, c) \</span></div><div class="line"><a name="l00148"></a><span class="lineno"> 148</span>&#160;<span class="preprocessor"> ({ \</span></div><div class="line"><a name="l00149"></a><span class="lineno"> 149</span>&#160;<span class="preprocessor"> ARM_MM_K0XN0X2(n0, k0, a, b, c); \</span></div><div class="line"><a name="l00150"></a><span class="lineno"> 150</span>&#160;<span class="preprocessor"> ARM_DOT_K0XN0(n0, k0, (a##2), b, (c##2)); \</span></div><div class="line"><a name="l00151"></a><span class="lineno"> 151</span>&#160;<span class="preprocessor"> })</span></div><div class="line"><a name="l00152"></a><span class="lineno"><a class="line" href="gemmlowp_8cl.xhtml#a7422ff84359302ff418d39bfa212a4cc"> 152</a></span>&#160;<span class="preprocessor">#define ARM_MM_K0XN0X4(n0, k0, a, b, c) \</span></div><div class="line"><a name="l00153"></a><span class="lineno"> 153</span>&#160;<span class="preprocessor"> ({ \</span></div><div class="line"><a name="l00154"></a><span class="lineno"> 154</span>&#160;<span class="preprocessor"> ARM_MM_K0XN0X3(n0, k0, a, b, c); \</span></div><div class="line"><a name="l00155"></a><span class="lineno"> 155</span>&#160;<span class="preprocessor"> ARM_DOT_K0XN0(n0, k0, (a##3), b, (c##3)); \</span></div><div class="line"><a name="l00156"></a><span class="lineno"> 156</span>&#160;<span class="preprocessor"> })</span></div><div class="line"><a name="l00157"></a><span class="lineno"><a class="line" href="gemmlowp_8cl.xhtml#a9ab09032778390f87d7050d58a0645e9"> 157</a></span>&#160;<span class="preprocessor">#define ARM_MM_K0XN0X5(n0, k0, a, b, c) \</span></div><div class="line"><a name="l00158"></a><span class="lineno"> 158</span>&#160;<span class="preprocessor"> ({ \</span></div><div class="line"><a name="l00159"></a><span class="lineno"> 159</span>&#160;<span class="preprocessor"> ARM_MM_K0XN0X4(n0, k0, a, b, c); \</span></div><div class="line"><a name="l00160"></a><span class="lineno"> 160</span>&#160;<span class="preprocessor"> ARM_DOT_K0XN0(n0, k0, (a##4), b, (c##4)); \</span></div><div class="line"><a name="l00161"></a><span class="lineno"> 161</span>&#160;<span class="preprocessor"> })</span></div><div class="line"><a name="l00162"></a><span class="lineno"><a class="line" href="gemmlowp_8cl.xhtml#a74a18e1f44fcfd47b8ec54dca07435e7"> 162</a></span>&#160;<span class="preprocessor">#define ARM_MM_K0XN0X6(n0, k0, a, b, c) \</span></div><div class="line"><a name="l00163"></a><span class="lineno"> 163</span>&#160;<span class="preprocessor"> ({ \</span></div><div class="line"><a name="l00164"></a><span class="lineno"> 164</span>&#160;<span class="preprocessor"> ARM_MM_K0XN0X5(n0, k0, a, b, c); \</span></div><div class="line"><a name="l00165"></a><span class="lineno"> 165</span>&#160;<span class="preprocessor"> ARM_DOT_K0XN0(n0, k0, (a##5), b, (c##5)); \</span></div><div class="line"><a name="l00166"></a><span class="lineno"> 166</span>&#160;<span class="preprocessor"> })</span></div><div class="line"><a name="l00167"></a><span class="lineno"><a class="line" href="gemmlowp_8cl.xhtml#a9233e80ae6cba402274f8921ffb22078"> 167</a></span>&#160;<span class="preprocessor">#define ARM_MM_K0XN0X7(n0, k0, a, b, c) \</span></div><div class="line"><a name="l00168"></a><span class="lineno"> 168</span>&#160;<span class="preprocessor"> ({ \</span></div><div class="line"><a name="l00169"></a><span class="lineno"> 169</span>&#160;<span class="preprocessor"> ARM_MM_K0XN0X6(n0, k0, a, b, c); \</span></div><div class="line"><a name="l00170"></a><span class="lineno"> 170</span>&#160;<span class="preprocessor"> ARM_DOT_K0XN0(n0, k0, (a##6), b, (c##6)); \</span></div><div class="line"><a name="l00171"></a><span class="lineno"> 171</span>&#160;<span class="preprocessor"> })</span></div><div class="line"><a name="l00172"></a><span class="lineno"><a class="line" href="gemmlowp_8cl.xhtml#a5cc688208a6d222018ca05eeabb461a2"> 172</a></span>&#160;<span class="preprocessor">#define ARM_MM_K0XN0X8(n0, k0, a, b, c) \</span></div><div class="line"><a name="l00173"></a><span class="lineno"> 173</span>&#160;<span class="preprocessor"> ({ \</span></div><div class="line"><a name="l00174"></a><span class="lineno"> 174</span>&#160;<span class="preprocessor"> ARM_MM_K0XN0X7(n0, k0, a, b, c); \</span></div><div class="line"><a name="l00175"></a><span class="lineno"> 175</span>&#160;<span class="preprocessor"> ARM_DOT_K0XN0(n0, k0, (a##7), b, (c##7)); \</span></div><div class="line"><a name="l00176"></a><span class="lineno"> 176</span>&#160;<span class="preprocessor"> })</span></div><div class="line"><a name="l00177"></a><span class="lineno"> 177</span>&#160;</div><div class="line"><a name="l00178"></a><span class="lineno"><a class="line" href="gemmlowp_8cl.xhtml#a66f8007bc273c19637859c71e89b7ef3"> 178</a></span>&#160;<span class="preprocessor">#define ARM_DOT_K0(k0, a, b, c) \</span></div><div class="line"><a name="l00179"></a><span class="lineno"> 179</span>&#160;<span class="preprocessor"> ({ \</span></div><div class="line"><a name="l00180"></a><span class="lineno"> 180</span>&#160;<span class="preprocessor"> CONCAT(ARM_DOT, k0) \</span></div><div class="line"><a name="l00181"></a><span class="lineno"> 181</span>&#160;<span class="preprocessor"> ((a), (b), (c)); \</span></div><div class="line"><a name="l00182"></a><span class="lineno"> 182</span>&#160;<span class="preprocessor"> })</span></div><div class="line"><a name="l00183"></a><span class="lineno"> 183</span>&#160;</div><div class="line"><a name="l00184"></a><span class="lineno"><a class="line" href="gemmlowp_8cl.xhtml#ae991dacacb5b9b0b176538400a7e8269"> 184</a></span>&#160;<span class="preprocessor">#define ARM_DOT_K0XN0(n0, k0, a, b, c) \</span></div><div class="line"><a name="l00185"></a><span class="lineno"> 185</span>&#160;<span class="preprocessor"> ({ \</span></div><div class="line"><a name="l00186"></a><span class="lineno"> 186</span>&#160;<span class="preprocessor"> CONCAT(ARM_DOT_K0X, n0) \</span></div><div class="line"><a name="l00187"></a><span class="lineno"> 187</span>&#160;<span class="preprocessor"> (k0, (a), b, (c)); \</span></div><div class="line"><a name="l00188"></a><span class="lineno"> 188</span>&#160;<span class="preprocessor"> })</span></div><div class="line"><a name="l00189"></a><span class="lineno"> 189</span>&#160;</div><div class="line"><a name="l00190"></a><span class="lineno"><a class="line" href="gemmlowp_8cl.xhtml#af110f6648cbcc21216a66fb3342632b0"> 190</a></span>&#160;<span class="preprocessor">#define ARM_MM_K0XN0XM0(m0, n0, k0, a, b, c) \</span></div><div class="line"><a name="l00191"></a><span class="lineno"> 191</span>&#160;<span class="preprocessor"> ({ \</span></div><div class="line"><a name="l00192"></a><span class="lineno"> 192</span>&#160;<span class="preprocessor"> CONCAT(ARM_MM_K0XN0X, m0) \</span></div><div class="line"><a name="l00193"></a><span class="lineno"> 193</span>&#160;<span class="preprocessor"> (n0, k0, a, b, c); \</span></div><div class="line"><a name="l00194"></a><span class="lineno"> 194</span>&#160;<span class="preprocessor"> })</span></div><div class="line"><a name="l00195"></a><span class="lineno"> 195</span>&#160;</div><div class="line"><a name="l00196"></a><span class="lineno"> 196</span>&#160;<span class="preprocessor">#if defined(NUM_ELEMS_PROCESSED_PER_THREAD_X) &amp;&amp; defined(NUM_ELEMS_PROCESSED_PER_THREAD_Y) &amp;&amp; defined(COLS_A)</span></div><div class="line"><a name="l00197"></a><span class="lineno"> 197</span>&#160;<span class="preprocessor">#define VECTOR_UCHAR VEC_DATA_TYPE(uchar, NUM_ELEMS_PROCESSED_PER_THREAD_X)</span></div><div class="line"><a name="l00198"></a><span class="lineno"> 198</span>&#160;<span class="preprocessor">#define VECTOR_UINT VEC_DATA_TYPE(uint, NUM_ELEMS_PROCESSED_PER_THREAD_X)</span></div><div class="line"><a name="l00199"></a><span class="lineno"> 199</span>&#160;<span class="preprocessor">#define VECTOR_INT VEC_DATA_TYPE(int, NUM_ELEMS_PROCESSED_PER_THREAD_X)</span></div><div class="line"><a name="l00200"></a><span class="lineno"> 200</span>&#160;</div><div class="line"><a name="l00235"></a><span class="lineno"> 235</span>&#160;__kernel <span class="keywordtype">void</span> gemmlowp_mm_midgard(<a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a22f42fcf2077d951271df83b55c1a71a">IMAGE_DECLARATION</a>(src0),</div><div class="line"><a name="l00236"></a><span class="lineno"> 236</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a22f42fcf2077d951271df83b55c1a71a">IMAGE_DECLARATION</a>(src1),</div><div class="line"><a name="l00237"></a><span class="lineno"> 237</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a22f42fcf2077d951271df83b55c1a71a">IMAGE_DECLARATION</a>(<a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#adbf67dcee294e673cf796f1ed8aeb6a4">dst</a>),</div><div class="line"><a name="l00238"></a><span class="lineno"> 238</span>&#160; uint src0_stride_z,</div><div class="line"><a name="l00239"></a><span class="lineno"> 239</span>&#160; uint src1_stride_z,</div><div class="line"><a name="l00240"></a><span class="lineno"> 240</span>&#160; uint dst_stride_z</div><div class="line"><a name="l00241"></a><span class="lineno"> 241</span>&#160;#<span class="keywordflow">if</span> defined(REINTERPRET_INPUT_AS_3D)</div><div class="line"><a name="l00242"></a><span class="lineno"> 242</span>&#160; ,</div><div class="line"><a name="l00243"></a><span class="lineno"> 243</span>&#160; uint src_cross_plane_pad</div><div class="line"><a name="l00244"></a><span class="lineno"> 244</span>&#160;#endif <span class="comment">// REINTERPRET_INPUT_AS_3D</span></div><div class="line"><a name="l00245"></a><span class="lineno"> 245</span>&#160;#<span class="keywordflow">if</span> defined(REINTERPRET_OUTPUT_AS_3D)</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; uint dst_cross_plane_pad</div><div class="line"><a name="l00248"></a><span class="lineno"> 248</span>&#160;#endif <span class="comment">// REINTERPRET_OUTPUT_AS_3D</span></div><div class="line"><a name="l00249"></a><span class="lineno"> 249</span>&#160; )</div><div class="line"><a name="l00250"></a><span class="lineno"> 250</span>&#160;{</div><div class="line"><a name="l00251"></a><span class="lineno"> 251</span>&#160; <span class="keywordtype">int</span> idx = get_global_id(0) * NUM_ELEMS_PROCESSED_PER_THREAD_X;</div><div class="line"><a name="l00252"></a><span class="lineno"> 252</span>&#160;</div><div class="line"><a name="l00253"></a><span class="lineno"> 253</span>&#160; <span class="comment">// Compute starting address for matrix A and Matrix B</span></div><div class="line"><a name="l00254"></a><span class="lineno"> 254</span>&#160; int2 src_addr = ((int2)(src0_offset_first_element_in_bytes, src1_offset_first_element_in_bytes));</div><div class="line"><a name="l00255"></a><span class="lineno"> 255</span>&#160;</div><div class="line"><a name="l00256"></a><span class="lineno"> 256</span>&#160; <span class="comment">// Update address for the matrix A</span></div><div class="line"><a name="l00257"></a><span class="lineno"> 257</span>&#160; src_addr.s0 += get_global_id(1) * src0_stride_y * NUM_ELEMS_PROCESSED_PER_THREAD_Y;</div><div class="line"><a name="l00258"></a><span class="lineno"> 258</span>&#160;</div><div class="line"><a name="l00259"></a><span class="lineno"> 259</span>&#160; <span class="comment">// Update address for the matrix B</span></div><div class="line"><a name="l00260"></a><span class="lineno"> 260</span>&#160; src_addr.s1 += idx;</div><div class="line"><a name="l00261"></a><span class="lineno"> 261</span>&#160;</div><div class="line"><a name="l00262"></a><span class="lineno"> 262</span>&#160;<span class="preprocessor">#if defined(REINTERPRET_INPUT_AS_3D)</span></div><div class="line"><a name="l00263"></a><span class="lineno"> 263</span>&#160; <span class="comment">// Since we load a 2D input tile from a 3D tensor, we need to check when the plane changes across the z dimension</span></div><div class="line"><a name="l00264"></a><span class="lineno"> 264</span>&#160; <span class="comment">// in order to take into account the presence of possible cross plane paddings</span></div><div class="line"><a name="l00265"></a><span class="lineno"> 265</span>&#160; <span class="comment">//</span></div><div class="line"><a name="l00266"></a><span class="lineno"> 266</span>&#160; <span class="comment">// | |</span></div><div class="line"><a name="l00267"></a><span class="lineno"> 267</span>&#160; <span class="comment">// | plane0 |</span></div><div class="line"><a name="l00268"></a><span class="lineno"> 268</span>&#160; <span class="comment">// | |</span></div><div class="line"><a name="l00269"></a><span class="lineno"> 269</span>&#160; <span class="comment">// |__________________|</span></div><div class="line"><a name="l00270"></a><span class="lineno"> 270</span>&#160; <span class="comment">// |******************|</span></div><div class="line"><a name="l00271"></a><span class="lineno"> 271</span>&#160; <span class="comment">// | cross_plane_pad |</span></div><div class="line"><a name="l00272"></a><span class="lineno"> 272</span>&#160; <span class="comment">// |******************|</span></div><div class="line"><a name="l00273"></a><span class="lineno"> 273</span>&#160; <span class="comment">// | |</span></div><div class="line"><a name="l00274"></a><span class="lineno"> 274</span>&#160; <span class="comment">// | plane1 |</span></div><div class="line"><a name="l00275"></a><span class="lineno"> 275</span>&#160; <span class="comment">// | |</span></div><div class="line"><a name="l00276"></a><span class="lineno"> 276</span>&#160; <span class="comment">// |__________________|</span></div><div class="line"><a name="l00277"></a><span class="lineno"> 277</span>&#160;</div><div class="line"><a name="l00278"></a><span class="lineno"> 278</span>&#160; <span class="comment">// The plane (zin) is calculated dividing M (get_global_id(1) * NUM_ELEMS_PROCESSED_PER_THREAD_Y) by HEIGHT_GEMM3D</span></div><div class="line"><a name="l00279"></a><span class="lineno"> 279</span>&#160; uint4 zin = ((uint4)(0, 1, 2, 3) + (uint4)(get_global_id(1) * NUM_ELEMS_PROCESSED_PER_THREAD_Y)) / (uint4)HEIGHT_GEMM3D;</div><div class="line"><a name="l00280"></a><span class="lineno"> 280</span>&#160; zin = min(DEPTH_GEMM3D - 1, zin);</div><div class="line"><a name="l00281"></a><span class="lineno"> 281</span>&#160;</div><div class="line"><a name="l00282"></a><span class="lineno"> 282</span>&#160; <span class="comment">// Add offset due to the cross plane paddings</span></div><div class="line"><a name="l00283"></a><span class="lineno"> 283</span>&#160; zin *= (src_cross_plane_pad * src0_stride_y);</div><div class="line"><a name="l00284"></a><span class="lineno"> 284</span>&#160;</div><div class="line"><a name="l00285"></a><span class="lineno"> 285</span>&#160; <span class="comment">// Add offset for batched GEMM. The batches will be in the fourth dimension and for this reason we</span></div><div class="line"><a name="l00286"></a><span class="lineno"> 286</span>&#160; <span class="comment">// multiply src0_stride_z by DEPTH_GEMM3D</span></div><div class="line"><a name="l00287"></a><span class="lineno"> 287</span>&#160; src_addr.s0 += get_global_id(2) * src0_stride_z * DEPTH_GEMM3D;</div><div class="line"><a name="l00288"></a><span class="lineno"> 288</span>&#160;</div><div class="line"><a name="l00289"></a><span class="lineno"> 289</span>&#160;<span class="preprocessor">#else // defined(REINTERPRET_INPUT_AS_3D)</span></div><div class="line"><a name="l00290"></a><span class="lineno"> 290</span>&#160;</div><div class="line"><a name="l00291"></a><span class="lineno"> 291</span>&#160; <span class="comment">// Add offset for batched GEMM</span></div><div class="line"><a name="l00292"></a><span class="lineno"> 292</span>&#160; src_addr.s0 += get_global_id(2) * src0_stride_z;</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="preprocessor">#endif // defined(REINTERPRET_INPUT_AS_3D)</span></div><div class="line"><a name="l00295"></a><span class="lineno"> 295</span>&#160;</div><div class="line"><a name="l00296"></a><span class="lineno"> 296</span>&#160;<span class="preprocessor">#if defined(MATRIX_B_DEPTH)</span></div><div class="line"><a name="l00297"></a><span class="lineno"> 297</span>&#160; <span class="comment">// Do not slide matrix B if the matrix B has 3 dimensions and matrix A more than 3</span></div><div class="line"><a name="l00298"></a><span class="lineno"> 298</span>&#160; src_addr.s1 += (get_global_id(2) % MATRIX_B_DEPTH) * src1_stride_z;</div><div class="line"><a name="l00299"></a><span class="lineno"> 299</span>&#160;<span class="preprocessor">#else // defined(MATRIX_B_DEPTH)</span></div><div class="line"><a name="l00300"></a><span class="lineno"> 300</span>&#160; src_addr.s1 += get_global_id(2) * src1_stride_z;</div><div class="line"><a name="l00301"></a><span class="lineno"> 301</span>&#160;<span class="preprocessor">#endif // defined(MATRIX_B_DEPTH)</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; <span class="keywordtype">int</span> end_row_vec_a = src_addr.s0 + COLS_A;</div><div class="line"><a name="l00304"></a><span class="lineno"> 304</span>&#160;</div><div class="line"><a name="l00305"></a><span class="lineno"> 305</span>&#160; VECTOR_UINT acc0 = 0;</div><div class="line"><a name="l00306"></a><span class="lineno"> 306</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 1</span></div><div class="line"><a name="l00307"></a><span class="lineno"> 307</span>&#160; VECTOR_UINT acc1 = 0;</div><div class="line"><a name="l00308"></a><span class="lineno"> 308</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 1</span></div><div class="line"><a name="l00309"></a><span class="lineno"> 309</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 2</span></div><div class="line"><a name="l00310"></a><span class="lineno"> 310</span>&#160; VECTOR_UINT acc2 = 0;</div><div class="line"><a name="l00311"></a><span class="lineno"> 311</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 2</span></div><div class="line"><a name="l00312"></a><span class="lineno"> 312</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 3</span></div><div class="line"><a name="l00313"></a><span class="lineno"> 313</span>&#160; VECTOR_UINT acc3 = 0;</div><div class="line"><a name="l00314"></a><span class="lineno"> 314</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 3</span></div><div class="line"><a name="l00315"></a><span class="lineno"> 315</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 4</span></div><div class="line"><a name="l00316"></a><span class="lineno"> 316</span>&#160; VECTOR_UINT acc4 = 0;</div><div class="line"><a name="l00317"></a><span class="lineno"> 317</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 4</span></div><div class="line"><a name="l00318"></a><span class="lineno"> 318</span>&#160;</div><div class="line"><a name="l00319"></a><span class="lineno"> 319</span>&#160; <span class="keywordflow">for</span>(; src_addr.s0 &lt;= (end_row_vec_a - 2); src_addr += (int2)(2, 2 * src1_stride_y))</div><div class="line"><a name="l00320"></a><span class="lineno"> 320</span>&#160; {</div><div class="line"><a name="l00321"></a><span class="lineno"> 321</span>&#160; <span class="comment">// Load values from matrix A</span></div><div class="line"><a name="l00322"></a><span class="lineno"> 322</span>&#160; uchar2 a0 = vload2(0, src0_ptr + src_addr.s0 + 0 * src0_stride_y);</div><div class="line"><a name="l00323"></a><span class="lineno"> 323</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 1</span></div><div class="line"><a name="l00324"></a><span class="lineno"> 324</span>&#160; uchar2 a1 = vload2(0, src0_ptr + src_addr.s0 + 1 * src0_stride_y);</div><div class="line"><a name="l00325"></a><span class="lineno"> 325</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 1</span></div><div class="line"><a name="l00326"></a><span class="lineno"> 326</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 2</span></div><div class="line"><a name="l00327"></a><span class="lineno"> 327</span>&#160; uchar2 a2 = vload2(0, src0_ptr + src_addr.s0 + 2 * src0_stride_y);</div><div class="line"><a name="l00328"></a><span class="lineno"> 328</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 2</span></div><div class="line"><a name="l00329"></a><span class="lineno"> 329</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 3</span></div><div class="line"><a name="l00330"></a><span class="lineno"> 330</span>&#160; uchar2 a3 = vload2(0, src0_ptr + src_addr.s0 + 3 * src0_stride_y);</div><div class="line"><a name="l00331"></a><span class="lineno"> 331</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 3</span></div><div class="line"><a name="l00332"></a><span class="lineno"> 332</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 4</span></div><div class="line"><a name="l00333"></a><span class="lineno"> 333</span>&#160; uchar2 a4 = vload2(0, src0_ptr + src_addr.s0 + 4 * src0_stride_y);</div><div class="line"><a name="l00334"></a><span class="lineno"> 334</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 4</span></div><div class="line"><a name="l00335"></a><span class="lineno"> 335</span>&#160; <span class="comment">// Load values from matrix B</span></div><div class="line"><a name="l00336"></a><span class="lineno"> 336</span>&#160; VECTOR_UCHAR b0 = <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a287e2fc366c312b468382c95bb90f91f">VLOAD</a>(NUM_ELEMS_PROCESSED_PER_THREAD_X)(0, src1_ptr + src_addr.s1);</div><div class="line"><a name="l00337"></a><span class="lineno"> 337</span>&#160; VECTOR_UCHAR b1 = <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a287e2fc366c312b468382c95bb90f91f">VLOAD</a>(NUM_ELEMS_PROCESSED_PER_THREAD_X)(0, src1_ptr + src_addr.s1 + src1_stride_y);</div><div class="line"><a name="l00338"></a><span class="lineno"> 338</span>&#160;</div><div class="line"><a name="l00339"></a><span class="lineno"> 339</span>&#160; <span class="comment">// Accumulate</span></div><div class="line"><a name="l00340"></a><span class="lineno"> 340</span>&#160; acc0 += <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#aa8d95ba04fc73845abc6045952cae5be">CONVERT</a>(b0, VECTOR_UINT) * (VECTOR_UINT)a0.s0;</div><div class="line"><a name="l00341"></a><span class="lineno"> 341</span>&#160; acc0 += <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#aa8d95ba04fc73845abc6045952cae5be">CONVERT</a>(b1, VECTOR_UINT) * (VECTOR_UINT)a0.s1;</div><div class="line"><a name="l00342"></a><span class="lineno"> 342</span>&#160;#<span class="keywordflow">if</span> NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 1</div><div class="line"><a name="l00343"></a><span class="lineno"> 343</span>&#160; acc1 += <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#aa8d95ba04fc73845abc6045952cae5be">CONVERT</a>(b0, VECTOR_UINT) * (VECTOR_UINT)a1.s0;</div><div class="line"><a name="l00344"></a><span class="lineno"> 344</span>&#160; acc1 += <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#aa8d95ba04fc73845abc6045952cae5be">CONVERT</a>(b1, VECTOR_UINT) * (VECTOR_UINT)a1.s1;</div><div class="line"><a name="l00345"></a><span class="lineno"> 345</span>&#160;#endif <span class="comment">// NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 1</span></div><div class="line"><a name="l00346"></a><span class="lineno"> 346</span>&#160;#<span class="keywordflow">if</span> NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 2</div><div class="line"><a name="l00347"></a><span class="lineno"> 347</span>&#160; acc2 += <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#aa8d95ba04fc73845abc6045952cae5be">CONVERT</a>(b0, VECTOR_UINT) * (VECTOR_UINT)a2.s0;</div><div class="line"><a name="l00348"></a><span class="lineno"> 348</span>&#160; acc2 += <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#aa8d95ba04fc73845abc6045952cae5be">CONVERT</a>(b1, VECTOR_UINT) * (VECTOR_UINT)a2.s1;</div><div class="line"><a name="l00349"></a><span class="lineno"> 349</span>&#160;#endif <span class="comment">// NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 2</span></div><div class="line"><a name="l00350"></a><span class="lineno"> 350</span>&#160;#<span class="keywordflow">if</span> NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 3</div><div class="line"><a name="l00351"></a><span class="lineno"> 351</span>&#160; acc3 += <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#aa8d95ba04fc73845abc6045952cae5be">CONVERT</a>(b0, VECTOR_UINT) * (VECTOR_UINT)a3.s0;</div><div class="line"><a name="l00352"></a><span class="lineno"> 352</span>&#160; acc3 += <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#aa8d95ba04fc73845abc6045952cae5be">CONVERT</a>(b1, VECTOR_UINT) * (VECTOR_UINT)a3.s1;</div><div class="line"><a name="l00353"></a><span class="lineno"> 353</span>&#160;#endif <span class="comment">// NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 3</span></div><div class="line"><a name="l00354"></a><span class="lineno"> 354</span>&#160;#<span class="keywordflow">if</span> NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 4</div><div class="line"><a name="l00355"></a><span class="lineno"> 355</span>&#160; acc4 += <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#aa8d95ba04fc73845abc6045952cae5be">CONVERT</a>(b0, VECTOR_UINT) * (VECTOR_UINT)a4.s0;</div><div class="line"><a name="l00356"></a><span class="lineno"> 356</span>&#160; acc4 += <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#aa8d95ba04fc73845abc6045952cae5be">CONVERT</a>(b1, VECTOR_UINT) * (VECTOR_UINT)a4.s1;</div><div class="line"><a name="l00357"></a><span class="lineno"> 357</span>&#160;#endif <span class="comment">// NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 4</span></div><div class="line"><a name="l00358"></a><span class="lineno"> 358</span>&#160; }</div><div class="line"><a name="l00359"></a><span class="lineno"> 359</span>&#160;</div><div class="line"><a name="l00360"></a><span class="lineno"> 360</span>&#160; <span class="keywordflow">for</span>(; src_addr.s0 &lt; end_row_vec_a; src_addr += (int2)(1, src1_stride_y))</div><div class="line"><a name="l00361"></a><span class="lineno"> 361</span>&#160; {</div><div class="line"><a name="l00362"></a><span class="lineno"> 362</span>&#160; <span class="comment">// Load values from matrix A</span></div><div class="line"><a name="l00363"></a><span class="lineno"> 363</span>&#160; uchar a0 = *(src0_ptr + src_addr.s0 + 0 * src0_stride_y);</div><div class="line"><a name="l00364"></a><span class="lineno"> 364</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 1</span></div><div class="line"><a name="l00365"></a><span class="lineno"> 365</span>&#160; uchar a1 = *(src0_ptr + src_addr.s0 + 1 * src0_stride_y);</div><div class="line"><a name="l00366"></a><span class="lineno"> 366</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 1</span></div><div class="line"><a name="l00367"></a><span class="lineno"> 367</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 2</span></div><div class="line"><a name="l00368"></a><span class="lineno"> 368</span>&#160; uchar a2 = *(src0_ptr + src_addr.s0 + 2 * src0_stride_y);</div><div class="line"><a name="l00369"></a><span class="lineno"> 369</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 2</span></div><div class="line"><a name="l00370"></a><span class="lineno"> 370</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 3</span></div><div class="line"><a name="l00371"></a><span class="lineno"> 371</span>&#160; uchar a3 = *(src0_ptr + src_addr.s0 + 3 * src0_stride_y);</div><div class="line"><a name="l00372"></a><span class="lineno"> 372</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 3</span></div><div class="line"><a name="l00373"></a><span class="lineno"> 373</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 4</span></div><div class="line"><a name="l00374"></a><span class="lineno"> 374</span>&#160; uchar a4 = *(src0_ptr + src_addr.s0 + 4 * src0_stride_y);</div><div class="line"><a name="l00375"></a><span class="lineno"> 375</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 4</span></div><div class="line"><a name="l00376"></a><span class="lineno"> 376</span>&#160; <span class="comment">// Load values from matrix B</span></div><div class="line"><a name="l00377"></a><span class="lineno"> 377</span>&#160; VECTOR_UCHAR b0 = <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a287e2fc366c312b468382c95bb90f91f">VLOAD</a>(NUM_ELEMS_PROCESSED_PER_THREAD_X)(0, src1_ptr + src_addr.s1);</div><div class="line"><a name="l00378"></a><span class="lineno"> 378</span>&#160;</div><div class="line"><a name="l00379"></a><span class="lineno"> 379</span>&#160; <span class="comment">// Accumulate</span></div><div class="line"><a name="l00380"></a><span class="lineno"> 380</span>&#160; acc0 += <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#aa8d95ba04fc73845abc6045952cae5be">CONVERT</a>(b0, VECTOR_UINT) * (VECTOR_UINT)a0;</div><div class="line"><a name="l00381"></a><span class="lineno"> 381</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 1</span></div><div class="line"><a name="l00382"></a><span class="lineno"> 382</span>&#160; acc1 += <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#aa8d95ba04fc73845abc6045952cae5be">CONVERT</a>(b0, VECTOR_UINT) * (VECTOR_UINT)a1;</div><div class="line"><a name="l00383"></a><span class="lineno"> 383</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 1</span></div><div class="line"><a name="l00384"></a><span class="lineno"> 384</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 2</span></div><div class="line"><a name="l00385"></a><span class="lineno"> 385</span>&#160; acc2 += <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#aa8d95ba04fc73845abc6045952cae5be">CONVERT</a>(b0, VECTOR_UINT) * (VECTOR_UINT)a2;</div><div class="line"><a name="l00386"></a><span class="lineno"> 386</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 2</span></div><div class="line"><a name="l00387"></a><span class="lineno"> 387</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 3</span></div><div class="line"><a name="l00388"></a><span class="lineno"> 388</span>&#160; acc3 += <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#aa8d95ba04fc73845abc6045952cae5be">CONVERT</a>(b0, VECTOR_UINT) * (VECTOR_UINT)a3;</div><div class="line"><a name="l00389"></a><span class="lineno"> 389</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 3</span></div><div class="line"><a name="l00390"></a><span class="lineno"> 390</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 4</span></div><div class="line"><a name="l00391"></a><span class="lineno"> 391</span>&#160; acc4 += <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#aa8d95ba04fc73845abc6045952cae5be">CONVERT</a>(b0, VECTOR_UINT) * (VECTOR_UINT)a4;</div><div class="line"><a name="l00392"></a><span class="lineno"> 392</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 4</span></div><div class="line"><a name="l00393"></a><span class="lineno"> 393</span>&#160; }</div><div class="line"><a name="l00394"></a><span class="lineno"> 394</span>&#160;</div><div class="line"><a name="l00395"></a><span class="lineno"> 395</span>&#160; <span class="keyword">const</span> <span class="keywordtype">int</span> z = get_global_id(2);</div><div class="line"><a name="l00396"></a><span class="lineno"> 396</span>&#160;</div><div class="line"><a name="l00397"></a><span class="lineno"> 397</span>&#160; <span class="comment">// Compute destination address</span></div><div class="line"><a name="l00398"></a><span class="lineno"> 398</span>&#160; <a class="code" href="struct_image.xhtml">Image</a> <a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#adbf67dcee294e673cf796f1ed8aeb6a4">dst</a> = <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#aebe814363556c244be043b13e7969197">CONVERT_TO_IMAGE_STRUCT</a>(<a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#adbf67dcee294e673cf796f1ed8aeb6a4">dst</a>);</div><div class="line"><a name="l00399"></a><span class="lineno"> 399</span>&#160;</div><div class="line"><a name="l00400"></a><span class="lineno"> 400</span>&#160;<span class="preprocessor">#if defined(REINTERPRET_OUTPUT_AS_3D)</span></div><div class="line"><a name="l00401"></a><span class="lineno"> 401</span>&#160; <span class="comment">// Since we store a 2D output tile in a 3D tensor, we need to check when the plane changes across the z dimension</span></div><div class="line"><a name="l00402"></a><span class="lineno"> 402</span>&#160; <span class="comment">// in order to take into account the presence of possible cross plane paddings</span></div><div class="line"><a name="l00403"></a><span class="lineno"> 403</span>&#160; <span class="comment">//</span></div><div class="line"><a name="l00404"></a><span class="lineno"> 404</span>&#160; <span class="comment">// | |</span></div><div class="line"><a name="l00405"></a><span class="lineno"> 405</span>&#160; <span class="comment">// | plane0 |</span></div><div class="line"><a name="l00406"></a><span class="lineno"> 406</span>&#160; <span class="comment">// | |</span></div><div class="line"><a name="l00407"></a><span class="lineno"> 407</span>&#160; <span class="comment">// |__________________|</span></div><div class="line"><a name="l00408"></a><span class="lineno"> 408</span>&#160; <span class="comment">// |******************|</span></div><div class="line"><a name="l00409"></a><span class="lineno"> 409</span>&#160; <span class="comment">// | cross_plane_pad |</span></div><div class="line"><a name="l00410"></a><span class="lineno"> 410</span>&#160; <span class="comment">// |******************|</span></div><div class="line"><a name="l00411"></a><span class="lineno"> 411</span>&#160; <span class="comment">// | |</span></div><div class="line"><a name="l00412"></a><span class="lineno"> 412</span>&#160; <span class="comment">// | plane1 |</span></div><div class="line"><a name="l00413"></a><span class="lineno"> 413</span>&#160; <span class="comment">// | |</span></div><div class="line"><a name="l00414"></a><span class="lineno"> 414</span>&#160; <span class="comment">// |__________________|</span></div><div class="line"><a name="l00415"></a><span class="lineno"> 415</span>&#160;</div><div class="line"><a name="l00416"></a><span class="lineno"> 416</span>&#160; <span class="comment">// The plane (zout) is calculated dividing M (get_global_id(1) * NUM_ELEMS_PROCESSED_PER_THREAD_Y) by HEIGHT_GEMM3D</span></div><div class="line"><a name="l00417"></a><span class="lineno"> 417</span>&#160; uint8 zout = ((uint8)(0, 1, 2, 3, 4, 5, 6, 7) + (uint8)(get_global_id(1) * NUM_ELEMS_PROCESSED_PER_THREAD_Y)) / (uint8)HEIGHT_GEMM3D;</div><div class="line"><a name="l00418"></a><span class="lineno"> 418</span>&#160; zout = min(DEPTH_GEMM3D - 1, zout);</div><div class="line"><a name="l00419"></a><span class="lineno"> 419</span>&#160;</div><div class="line"><a name="l00420"></a><span class="lineno"> 420</span>&#160; <span class="comment">// Add offset due to the cross plane paddings</span></div><div class="line"><a name="l00421"></a><span class="lineno"> 421</span>&#160; zout *= (dst_cross_plane_pad * dst_stride_y);</div><div class="line"><a name="l00422"></a><span class="lineno"> 422</span>&#160;</div><div class="line"><a name="l00423"></a><span class="lineno"> 423</span>&#160; <span class="comment">// Add offset for batched GEMM. The batches will be in the fourth dimension and for this reason we</span></div><div class="line"><a name="l00424"></a><span class="lineno"> 424</span>&#160; <span class="comment">// multiply dst_stride_z by DEPTH_GEMM3D</span></div><div class="line"><a name="l00425"></a><span class="lineno"> 425</span>&#160; <a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#adbf67dcee294e673cf796f1ed8aeb6a4">dst</a>.ptr += z * dst_stride_z * DEPTH_GEMM3D;</div><div class="line"><a name="l00426"></a><span class="lineno"> 426</span>&#160;</div><div class="line"><a name="l00427"></a><span class="lineno"> 427</span>&#160; <span class="comment">// Store the result</span></div><div class="line"><a name="l00428"></a><span class="lineno"> 428</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#acb282042d1edeeaa3cc979a206f78b54">VSTORE</a>(NUM_ELEMS_PROCESSED_PER_THREAD_X)</div><div class="line"><a name="l00429"></a><span class="lineno"> 429</span>&#160; (<a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#aa8d95ba04fc73845abc6045952cae5be">CONVERT</a>(acc0, VECTOR_INT), 0, (__global <span class="keywordtype">int</span> *)(<a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#adbf67dcee294e673cf796f1ed8aeb6a4">dst</a>.ptr + 0 * dst_stride_y + zout.s0));</div><div class="line"><a name="l00430"></a><span class="lineno"> 430</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 1</span></div><div class="line"><a name="l00431"></a><span class="lineno"> 431</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#acb282042d1edeeaa3cc979a206f78b54">VSTORE</a>(NUM_ELEMS_PROCESSED_PER_THREAD_X)</div><div class="line"><a name="l00432"></a><span class="lineno"> 432</span>&#160; (<a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#aa8d95ba04fc73845abc6045952cae5be">CONVERT</a>(acc1, VECTOR_INT), 0, (__global <span class="keywordtype">int</span> *)(<a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#adbf67dcee294e673cf796f1ed8aeb6a4">dst</a>.ptr + 1 * dst_stride_y + zout.s1));</div><div class="line"><a name="l00433"></a><span class="lineno"> 433</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 1</span></div><div class="line"><a name="l00434"></a><span class="lineno"> 434</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 2</span></div><div class="line"><a name="l00435"></a><span class="lineno"> 435</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#acb282042d1edeeaa3cc979a206f78b54">VSTORE</a>(NUM_ELEMS_PROCESSED_PER_THREAD_X)</div><div class="line"><a name="l00436"></a><span class="lineno"> 436</span>&#160; (<a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#aa8d95ba04fc73845abc6045952cae5be">CONVERT</a>(acc2, VECTOR_INT), 0, (__global <span class="keywordtype">int</span> *)(<a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#adbf67dcee294e673cf796f1ed8aeb6a4">dst</a>.ptr + 2 * dst_stride_y + zout.s2));</div><div class="line"><a name="l00437"></a><span class="lineno"> 437</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 2</span></div><div class="line"><a name="l00438"></a><span class="lineno"> 438</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 3</span></div><div class="line"><a name="l00439"></a><span class="lineno"> 439</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#acb282042d1edeeaa3cc979a206f78b54">VSTORE</a>(NUM_ELEMS_PROCESSED_PER_THREAD_X)</div><div class="line"><a name="l00440"></a><span class="lineno"> 440</span>&#160; (<a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#aa8d95ba04fc73845abc6045952cae5be">CONVERT</a>(acc3, VECTOR_INT), 0, (__global <span class="keywordtype">int</span> *)(<a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#adbf67dcee294e673cf796f1ed8aeb6a4">dst</a>.ptr + 3 * dst_stride_y + zout.s3));</div><div class="line"><a name="l00441"></a><span class="lineno"> 441</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 3</span></div><div class="line"><a name="l00442"></a><span class="lineno"> 442</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 4</span></div><div class="line"><a name="l00443"></a><span class="lineno"> 443</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#acb282042d1edeeaa3cc979a206f78b54">VSTORE</a>(NUM_ELEMS_PROCESSED_PER_THREAD_X)</div><div class="line"><a name="l00444"></a><span class="lineno"> 444</span>&#160; (<a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#aa8d95ba04fc73845abc6045952cae5be">CONVERT</a>(acc4, VECTOR_INT), 0, (__global <span class="keywordtype">int</span> *)(<a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#adbf67dcee294e673cf796f1ed8aeb6a4">dst</a>.ptr + 4 * dst_stride_y + zout.s4));</div><div class="line"><a name="l00445"></a><span class="lineno"> 445</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 4</span></div><div class="line"><a name="l00446"></a><span class="lineno"> 446</span>&#160;</div><div class="line"><a name="l00447"></a><span class="lineno"> 447</span>&#160;<span class="preprocessor">#else // defined(REINTERPRET_OUTPUT_AS_3D)</span></div><div class="line"><a name="l00448"></a><span class="lineno"> 448</span>&#160; <span class="comment">// Add offset for batched GEMM</span></div><div class="line"><a name="l00449"></a><span class="lineno"> 449</span>&#160; <a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#adbf67dcee294e673cf796f1ed8aeb6a4">dst</a>.ptr += z * dst_stride_z;</div><div class="line"><a name="l00450"></a><span class="lineno"> 450</span>&#160;</div><div class="line"><a name="l00451"></a><span class="lineno"> 451</span>&#160; <span class="comment">// Store the result</span></div><div class="line"><a name="l00452"></a><span class="lineno"> 452</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#acb282042d1edeeaa3cc979a206f78b54">VSTORE</a>(NUM_ELEMS_PROCESSED_PER_THREAD_X)</div><div class="line"><a name="l00453"></a><span class="lineno"> 453</span>&#160; (<a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#aa8d95ba04fc73845abc6045952cae5be">CONVERT</a>(acc0, VECTOR_INT), 0, (__global <span class="keywordtype">int</span> *)(<a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#adbf67dcee294e673cf796f1ed8aeb6a4">dst</a>.ptr + 0 * dst_stride_y));</div><div class="line"><a name="l00454"></a><span class="lineno"> 454</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 1</span></div><div class="line"><a name="l00455"></a><span class="lineno"> 455</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#acb282042d1edeeaa3cc979a206f78b54">VSTORE</a>(NUM_ELEMS_PROCESSED_PER_THREAD_X)</div><div class="line"><a name="l00456"></a><span class="lineno"> 456</span>&#160; (<a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#aa8d95ba04fc73845abc6045952cae5be">CONVERT</a>(acc1, VECTOR_INT), 0, (__global <span class="keywordtype">int</span> *)(<a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#adbf67dcee294e673cf796f1ed8aeb6a4">dst</a>.ptr + 1 * dst_stride_y));</div><div class="line"><a name="l00457"></a><span class="lineno"> 457</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 1</span></div><div class="line"><a name="l00458"></a><span class="lineno"> 458</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 2</span></div><div class="line"><a name="l00459"></a><span class="lineno"> 459</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#acb282042d1edeeaa3cc979a206f78b54">VSTORE</a>(NUM_ELEMS_PROCESSED_PER_THREAD_X)</div><div class="line"><a name="l00460"></a><span class="lineno"> 460</span>&#160; (<a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#aa8d95ba04fc73845abc6045952cae5be">CONVERT</a>(acc2, VECTOR_INT), 0, (__global <span class="keywordtype">int</span> *)(<a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#adbf67dcee294e673cf796f1ed8aeb6a4">dst</a>.ptr + 2 * dst_stride_y));</div><div class="line"><a name="l00461"></a><span class="lineno"> 461</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 2</span></div><div class="line"><a name="l00462"></a><span class="lineno"> 462</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 3</span></div><div class="line"><a name="l00463"></a><span class="lineno"> 463</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#acb282042d1edeeaa3cc979a206f78b54">VSTORE</a>(NUM_ELEMS_PROCESSED_PER_THREAD_X)</div><div class="line"><a name="l00464"></a><span class="lineno"> 464</span>&#160; (<a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#aa8d95ba04fc73845abc6045952cae5be">CONVERT</a>(acc3, VECTOR_INT), 0, (__global <span class="keywordtype">int</span> *)(<a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#adbf67dcee294e673cf796f1ed8aeb6a4">dst</a>.ptr + 3 * dst_stride_y));</div><div class="line"><a name="l00465"></a><span class="lineno"> 465</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 3</span></div><div class="line"><a name="l00466"></a><span class="lineno"> 466</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 4</span></div><div class="line"><a name="l00467"></a><span class="lineno"> 467</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#acb282042d1edeeaa3cc979a206f78b54">VSTORE</a>(NUM_ELEMS_PROCESSED_PER_THREAD_X)</div><div class="line"><a name="l00468"></a><span class="lineno"> 468</span>&#160; (<a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#aa8d95ba04fc73845abc6045952cae5be">CONVERT</a>(acc4, VECTOR_INT), 0, (__global <span class="keywordtype">int</span> *)(<a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#adbf67dcee294e673cf796f1ed8aeb6a4">dst</a>.ptr + 4 * dst_stride_y));</div><div class="line"><a name="l00469"></a><span class="lineno"> 469</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 4</span></div><div class="line"><a name="l00470"></a><span class="lineno"> 470</span>&#160;<span class="preprocessor">#endif // defined(REINTERPRET_OUTPUT_AS_3D)</span></div><div class="line"><a name="l00471"></a><span class="lineno"> 471</span>&#160;}</div><div class="line"><a name="l00472"></a><span class="lineno"> 472</span>&#160;<span class="preprocessor">#endif // defined(NUM_ELEMS_PROCESSED_PER_THREAD_X) &amp;&amp; defined(NUM_ELEMS_PROCESSED_PER_THREAD_Y) &amp;&amp; defined(COLS_A)</span></div><div class="line"><a name="l00473"></a><span class="lineno"> 473</span>&#160;</div><div class="line"><a name="l00474"></a><span class="lineno"> 474</span>&#160;<span class="preprocessor">#if defined(M0) &amp;&amp; defined(N0) &amp;&amp; defined(K0) &amp;&amp; defined(V0) &amp;&amp; defined(H0) &amp;&amp; defined(M) &amp;&amp; defined(N)</span></div><div class="line"><a name="l00475"></a><span class="lineno"> 475</span>&#160;</div><div class="line"><a name="l00523"></a><span class="lineno"> 523</span>&#160;__kernel <span class="keywordtype">void</span> gemmlowp_mm_reshaped_lhs_nt_rhs_t(<a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a22f42fcf2077d951271df83b55c1a71a">IMAGE_DECLARATION</a>(lhs),</div><div class="line"><a name="l00524"></a><span class="lineno"> 524</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a22f42fcf2077d951271df83b55c1a71a">IMAGE_DECLARATION</a>(rhs),</div><div class="line"><a name="l00525"></a><span class="lineno"> 525</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a22f42fcf2077d951271df83b55c1a71a">IMAGE_DECLARATION</a>(<a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#adbf67dcee294e673cf796f1ed8aeb6a4">dst</a>),</div><div class="line"><a name="l00526"></a><span class="lineno"> 526</span>&#160; uint k,</div><div class="line"><a name="l00527"></a><span class="lineno"> 527</span>&#160; uint lhs_stride_z,</div><div class="line"><a name="l00528"></a><span class="lineno"> 528</span>&#160; uint rhs_stride_z,</div><div class="line"><a name="l00529"></a><span class="lineno"> 529</span>&#160; uint dst_stride_z</div><div class="line"><a name="l00530"></a><span class="lineno"> 530</span>&#160;#<span class="keywordflow">if</span> defined(REINTERPRET_OUTPUT_AS_3D)</div><div class="line"><a name="l00531"></a><span class="lineno"> 531</span>&#160; ,</div><div class="line"><a name="l00532"></a><span class="lineno"> 532</span>&#160; uint dst_cross_plane_pad</div><div class="line"><a name="l00533"></a><span class="lineno"> 533</span>&#160;#endif <span class="comment">// REINTERPRET_OUTPUT_AS_3D</span></div><div class="line"><a name="l00534"></a><span class="lineno"> 534</span>&#160; )</div><div class="line"><a name="l00535"></a><span class="lineno"> 535</span>&#160;{</div><div class="line"><a name="l00536"></a><span class="lineno"> 536</span>&#160; <span class="comment">// Block size</span></div><div class="line"><a name="l00537"></a><span class="lineno"> 537</span>&#160;<span class="preprocessor">#define LHS_BLOCK_SIZE ((K0) * (M0))</span></div><div class="line"><a name="l00538"></a><span class="lineno"> 538</span>&#160;</div><div class="line"><a name="l00539"></a><span class="lineno"> 539</span>&#160;<span class="preprocessor">#if defined(LHS_INTERLEAVE)</span></div><div class="line"><a name="l00540"></a><span class="lineno"> 540</span>&#160;<span class="preprocessor">#define LHS_OFFSET_X (K0)</span></div><div class="line"><a name="l00541"></a><span class="lineno"> 541</span>&#160;<span class="preprocessor">#define LHS_STEP_X ((K0) * (V0))</span></div><div class="line"><a name="l00542"></a><span class="lineno"> 542</span>&#160;<span class="preprocessor">#define LHS_STEP_LOOP (1)</span></div><div class="line"><a name="l00543"></a><span class="lineno"> 543</span>&#160;<span class="preprocessor">#else // defined(INTERLEAVE)</span></div><div class="line"><a name="l00544"></a><span class="lineno"> 544</span>&#160;<span class="preprocessor">#define LHS_OFFSET_X (LHS_BLOCK_SIZE)</span></div><div class="line"><a name="l00545"></a><span class="lineno"> 545</span>&#160;<span class="preprocessor">#define LHS_STEP_X (K0)</span></div><div class="line"><a name="l00546"></a><span class="lineno"> 546</span>&#160;<span class="preprocessor">#define LHS_STEP_LOOP (V0)</span></div><div class="line"><a name="l00547"></a><span class="lineno"> 547</span>&#160;<span class="preprocessor">#endif // defined(INTERLEAVE)</span></div><div class="line"><a name="l00548"></a><span class="lineno"> 548</span>&#160;</div><div class="line"><a name="l00549"></a><span class="lineno"> 549</span>&#160; <span class="comment">// Block size</span></div><div class="line"><a name="l00550"></a><span class="lineno"> 550</span>&#160;<span class="preprocessor">#define RHS_BLOCK_SIZE ((K0) * (N0))</span></div><div class="line"><a name="l00551"></a><span class="lineno"> 551</span>&#160;</div><div class="line"><a name="l00552"></a><span class="lineno"> 552</span>&#160; <span class="comment">// RHS offset and step X</span></div><div class="line"><a name="l00553"></a><span class="lineno"> 553</span>&#160;<span class="preprocessor">#if defined(RHS_INTERLEAVE)</span></div><div class="line"><a name="l00554"></a><span class="lineno"> 554</span>&#160;<span class="preprocessor">#define RHS_OFFSET_X (K0)</span></div><div class="line"><a name="l00555"></a><span class="lineno"> 555</span>&#160;<span class="preprocessor">#define RHS_STEP_X ((K0) * (H0))</span></div><div class="line"><a name="l00556"></a><span class="lineno"> 556</span>&#160;<span class="preprocessor">#define RHS_STEP_LOOP (1)</span></div><div class="line"><a name="l00557"></a><span class="lineno"> 557</span>&#160;<span class="preprocessor">#else // defined(RHS_INTERLEAVE)</span></div><div class="line"><a name="l00558"></a><span class="lineno"> 558</span>&#160;<span class="preprocessor">#define RHS_OFFSET_X (RHS_BLOCK_SIZE)</span></div><div class="line"><a name="l00559"></a><span class="lineno"> 559</span>&#160;<span class="preprocessor">#define RHS_STEP_X (K0)</span></div><div class="line"><a name="l00560"></a><span class="lineno"> 560</span>&#160;<span class="preprocessor">#define RHS_STEP_LOOP (H0)</span></div><div class="line"><a name="l00561"></a><span class="lineno"> 561</span>&#160;<span class="preprocessor">#endif // defined(RHS_INTERLEAVE)</span></div><div class="line"><a name="l00562"></a><span class="lineno"> 562</span>&#160;</div><div class="line"><a name="l00563"></a><span class="lineno"> 563</span>&#160; uint x = get_global_id(0);</div><div class="line"><a name="l00564"></a><span class="lineno"> 564</span>&#160; uint y = get_global_id(1);</div><div class="line"><a name="l00565"></a><span class="lineno"> 565</span>&#160; uint z = get_global_id(2);</div><div class="line"><a name="l00566"></a><span class="lineno"> 566</span>&#160;</div><div class="line"><a name="l00567"></a><span class="lineno"> 567</span>&#160;<span class="preprocessor">#if defined(DUMMY_WORK_ITEMS)</span></div><div class="line"><a name="l00568"></a><span class="lineno"> 568</span>&#160; <span class="keywordflow">if</span>((x * N0 &gt;= N) || (y * M0 &gt;= M))</div><div class="line"><a name="l00569"></a><span class="lineno"> 569</span>&#160; {</div><div class="line"><a name="l00570"></a><span class="lineno"> 570</span>&#160; <span class="keywordflow">return</span>;</div><div class="line"><a name="l00571"></a><span class="lineno"> 571</span>&#160; }</div><div class="line"><a name="l00572"></a><span class="lineno"> 572</span>&#160;<span class="preprocessor">#endif // defined(DUMMY_WORK_ITEMS)</span></div><div class="line"><a name="l00573"></a><span class="lineno"> 573</span>&#160;</div><div class="line"><a name="l00574"></a><span class="lineno"> 574</span>&#160; <span class="comment">// Compute LHS matrix address</span></div><div class="line"><a name="l00575"></a><span class="lineno"> 575</span>&#160; __global uchar *lhs_addr = lhs_ptr + lhs_offset_first_element_in_bytes + (y % V0) * (uint)LHS_OFFSET_X + (y / V0) * (uint)lhs_stride_y + (z * lhs_stride_z);</div><div class="line"><a name="l00576"></a><span class="lineno"> 576</span>&#160;</div><div class="line"><a name="l00577"></a><span class="lineno"> 577</span>&#160; <span class="comment">// Compute RHS matrix address</span></div><div class="line"><a name="l00578"></a><span class="lineno"> 578</span>&#160; __global uchar *rhs_addr = rhs_ptr + rhs_offset_first_element_in_bytes + (x % H0) * (uint)RHS_OFFSET_X + (x / (uint)H0) * rhs_stride_y;</div><div class="line"><a name="l00579"></a><span class="lineno"> 579</span>&#160;</div><div class="line"><a name="l00580"></a><span class="lineno"> 580</span>&#160;<span class="preprocessor">#if defined(MATRIX_B_DEPTH)</span></div><div class="line"><a name="l00581"></a><span class="lineno"> 581</span>&#160; <span class="comment">// Do not slide matrix B if the matrix B has 3 dimensions and matrix A more than 3</span></div><div class="line"><a name="l00582"></a><span class="lineno"> 582</span>&#160; rhs_addr += (z % MATRIX_B_DEPTH) * rhs_stride_z;</div><div class="line"><a name="l00583"></a><span class="lineno"> 583</span>&#160;<span class="preprocessor">#else // defined(MATRIX_B_DEPTH)</span></div><div class="line"><a name="l00584"></a><span class="lineno"> 584</span>&#160; rhs_addr += z * rhs_stride_z;</div><div class="line"><a name="l00585"></a><span class="lineno"> 585</span>&#160;<span class="preprocessor">#endif // defined(MATRIX_B_DEPTH)</span></div><div class="line"><a name="l00586"></a><span class="lineno"> 586</span>&#160;</div><div class="line"><a name="l00587"></a><span class="lineno"> 587</span>&#160; <a class="code" href="repeat_8h.xhtml#a4a61a8df72f9bc1b1dcc4cc4ec6ac3b9">REPEAT_VAR_INIT_TO_CONST</a>(8, uint, zlhs, 0); <span class="comment">//uint zout0=0,zout1=0,zout2=0,... zout7=0;</span></div><div class="line"><a name="l00588"></a><span class="lineno"> 588</span>&#160; <a class="code" href="repeat_8h.xhtml#a4a61a8df72f9bc1b1dcc4cc4ec6ac3b9">REPEAT_VAR_INIT_TO_CONST</a>(16, uint, zrhs, 0);</div><div class="line"><a name="l00589"></a><span class="lineno"> 589</span>&#160;</div><div class="line"><a name="l00590"></a><span class="lineno"> 590</span>&#160; <span class="comment">// Initialize the accumulators</span></div><div class="line"><a name="l00591"></a><span class="lineno"> 591</span>&#160; <a class="code" href="repeat_8h.xhtml#a4a61a8df72f9bc1b1dcc4cc4ec6ac3b9">REPEAT_VAR_INIT_TO_CONST</a>(M0, <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(uint, N0), c, 0); <span class="comment">//VEC_DATA_TYPE(uint, N0) c0=0,c1=0,c2=0,... c(M0-1)=0;</span></div><div class="line"><a name="l00592"></a><span class="lineno"> 592</span>&#160;</div><div class="line"><a name="l00593"></a><span class="lineno"> 593</span>&#160; <span class="keywordflow">for</span>(<span class="keywordtype">int</span> i = 0; i &lt; k; i += K0)</div><div class="line"><a name="l00594"></a><span class="lineno"> 594</span>&#160; {</div><div class="line"><a name="l00595"></a><span class="lineno"> 595</span>&#160; <span class="comment">// Load values from LHS matrix</span></div><div class="line"><a name="l00596"></a><span class="lineno"> 596</span>&#160; <a class="code" href="gemm__helpers_8h.xhtml#a9fe7ded27e552f48fa285d34fb00427d">LOAD_BLOCK</a>(M0, K0, uchar, a, lhs_addr, 0, LHS_STEP_X, zlhs);</div><div class="line"><a name="l00597"></a><span class="lineno"> 597</span>&#160;</div><div class="line"><a name="l00598"></a><span class="lineno"> 598</span>&#160; <span class="comment">// Load values from RHS matrix</span></div><div class="line"><a name="l00599"></a><span class="lineno"> 599</span>&#160; <a class="code" href="gemm__helpers_8h.xhtml#a9fe7ded27e552f48fa285d34fb00427d">LOAD_BLOCK</a>(N0, K0, uchar, <a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#aa76b4a6e74940dabc5b7fc6b2dab3545">b</a>, rhs_addr, 0, RHS_STEP_X, zrhs);</div><div class="line"><a name="l00600"></a><span class="lineno"> 600</span>&#160;</div><div class="line"><a name="l00601"></a><span class="lineno"> 601</span>&#160; <span class="comment">// Partial matrix multiplication M0,N0,K0</span></div><div class="line"><a name="l00602"></a><span class="lineno"> 602</span>&#160; <a class="code" href="gemmlowp_8cl.xhtml#af110f6648cbcc21216a66fb3342632b0">ARM_MM_K0XN0XM0</a>(M0, N0, K0, a, <a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#aa76b4a6e74940dabc5b7fc6b2dab3545">b</a>, c);</div><div class="line"><a name="l00603"></a><span class="lineno"> 603</span>&#160;</div><div class="line"><a name="l00604"></a><span class="lineno"> 604</span>&#160; <span class="comment">// Update address</span></div><div class="line"><a name="l00605"></a><span class="lineno"> 605</span>&#160; lhs_addr += (M0 * LHS_STEP_X * LHS_STEP_LOOP);</div><div class="line"><a name="l00606"></a><span class="lineno"> 606</span>&#160; rhs_addr += (N0 * RHS_STEP_X * RHS_STEP_LOOP);</div><div class="line"><a name="l00607"></a><span class="lineno"> 607</span>&#160; }</div><div class="line"><a name="l00608"></a><span class="lineno"> 608</span>&#160;</div><div class="line"><a name="l00609"></a><span class="lineno"> 609</span>&#160; __global uchar *dst_addr = dst_ptr + dst_offset_first_element_in_bytes + (x * (uint)N0 * <span class="keyword">sizeof</span>(<span class="keywordtype">int</span>)) + (y * (uint)M0 * dst_stride_y);</div><div class="line"><a name="l00610"></a><span class="lineno"> 610</span>&#160;</div><div class="line"><a name="l00611"></a><span class="lineno"> 611</span>&#160; <a class="code" href="repeat_8h.xhtml#a4a61a8df72f9bc1b1dcc4cc4ec6ac3b9">REPEAT_VAR_INIT_TO_CONST</a>(8, uint, zout, 0); <span class="comment">//uint zout0=0,zout1=0,zout2=0,... zout7=0;</span></div><div class="line"><a name="l00612"></a><span class="lineno"> 612</span>&#160;</div><div class="line"><a name="l00613"></a><span class="lineno"> 613</span>&#160;<span class="preprocessor">#if defined(REINTERPRET_OUTPUT_AS_3D)</span></div><div class="line"><a name="l00614"></a><span class="lineno"> 614</span>&#160; <span class="comment">// The plane (zout) is calculated dividing M (y * M0) by HEIGHT_GEMM3D</span></div><div class="line"><a name="l00615"></a><span class="lineno"> 615</span>&#160; <a class="code" href="gemm__helpers_8h.xhtml#ade58486d6b2d13d7941106a9d2ec86d3">CALCULATE_Z_OFFSET</a>(M0, uint, zout, y, HEIGHT_GEMM3D, DEPTH_GEMM3D, dst_cross_plane_pad, dst_stride_y);</div><div class="line"><a name="l00616"></a><span class="lineno"> 616</span>&#160;</div><div class="line"><a name="l00617"></a><span class="lineno"> 617</span>&#160; <span class="comment">// Add offset for batched GEMM. The batches will be in the fourth dimension and for this reason we</span></div><div class="line"><a name="l00618"></a><span class="lineno"> 618</span>&#160; <span class="comment">// multiply dst_stride_z by DEPTH_GEMM3D</span></div><div class="line"><a name="l00619"></a><span class="lineno"> 619</span>&#160; dst_addr += z * dst_stride_z * DEPTH_GEMM3D;</div><div class="line"><a name="l00620"></a><span class="lineno"> 620</span>&#160;</div><div class="line"><a name="l00621"></a><span class="lineno"> 621</span>&#160;<span class="preprocessor">#else // defined(REINTERPRET_OUTPUT_AS_3D)</span></div><div class="line"><a name="l00622"></a><span class="lineno"> 622</span>&#160;</div><div class="line"><a name="l00623"></a><span class="lineno"> 623</span>&#160; <span class="comment">// Add offset for batched GEMM</span></div><div class="line"><a name="l00624"></a><span class="lineno"> 624</span>&#160; dst_addr += z * dst_stride_z;</div><div class="line"><a name="l00625"></a><span class="lineno"> 625</span>&#160;</div><div class="line"><a name="l00626"></a><span class="lineno"> 626</span>&#160;<span class="preprocessor">#endif // defined(REINTERPRET_OUTPUT_AS_3D)</span></div><div class="line"><a name="l00627"></a><span class="lineno"> 627</span>&#160;</div><div class="line"><a name="l00628"></a><span class="lineno"> 628</span>&#160; <span class="comment">// Convert and store output block</span></div><div class="line"><a name="l00629"></a><span class="lineno"> 629</span>&#160; <a class="code" href="gemm__helpers_8h.xhtml#af5c0a8dfefbd611d38bb91212684ef7d">CONVERT_STORE_BLOCK</a>(M0, N0, <span class="keywordtype">int</span>, c, dst_addr, dst_stride_y, zout);</div><div class="line"><a name="l00630"></a><span class="lineno"> 630</span>&#160;</div><div class="line"><a name="l00631"></a><span class="lineno"> 631</span>&#160;<span class="preprocessor">#undef LHS_BLOCK_SIZE</span></div><div class="line"><a name="l00632"></a><span class="lineno"> 632</span>&#160;<span class="preprocessor">#undef LHS_OFFSET_X</span></div><div class="line"><a name="l00633"></a><span class="lineno"> 633</span>&#160;<span class="preprocessor">#undef LHS_STEP_X</span></div><div class="line"><a name="l00634"></a><span class="lineno"> 634</span>&#160;<span class="preprocessor">#undef RHS_BLOCK_SIZE</span></div><div class="line"><a name="l00635"></a><span class="lineno"> 635</span>&#160;<span class="preprocessor">#undef RHS_OFFSET_X</span></div><div class="line"><a name="l00636"></a><span class="lineno"> 636</span>&#160;<span class="preprocessor">#undef RHS_STEP_X</span></div><div class="line"><a name="l00637"></a><span class="lineno"> 637</span>&#160;}</div><div class="line"><a name="l00638"></a><span class="lineno"> 638</span>&#160;<span class="preprocessor">#endif // defined(M0) &amp;&amp; defined(N0) &amp;&amp; defined(K0) &amp;&amp; defined(V0) &amp;&amp; defined(H0) &amp;&amp; defined(K)</span></div><div class="line"><a name="l00639"></a><span class="lineno"> 639</span>&#160;</div><div class="line"><a name="l00640"></a><span class="lineno"> 640</span>&#160;<span class="preprocessor">#if defined(M0) &amp;&amp; defined(N0) &amp;&amp; defined(K0) &amp;&amp; defined(H0) &amp;&amp; defined(K)</span></div><div class="line"><a name="l00641"></a><span class="lineno"> 641</span>&#160;</div><div class="line"><a name="l00688"></a><span class="lineno"> 688</span>&#160;__kernel <span class="keywordtype">void</span> gemmlowp_mm_reshaped_only_rhs_t(<a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a22f42fcf2077d951271df83b55c1a71a">IMAGE_DECLARATION</a>(lhs),</div><div class="line"><a name="l00689"></a><span class="lineno"> 689</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a22f42fcf2077d951271df83b55c1a71a">IMAGE_DECLARATION</a>(rhs),</div><div class="line"><a name="l00690"></a><span class="lineno"> 690</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a22f42fcf2077d951271df83b55c1a71a">IMAGE_DECLARATION</a>(<a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#adbf67dcee294e673cf796f1ed8aeb6a4">dst</a>),</div><div class="line"><a name="l00691"></a><span class="lineno"> 691</span>&#160; uint lhs_stride_z,</div><div class="line"><a name="l00692"></a><span class="lineno"> 692</span>&#160; uint rhs_stride_z,</div><div class="line"><a name="l00693"></a><span class="lineno"> 693</span>&#160; uint dst_stride_z</div><div class="line"><a name="l00694"></a><span class="lineno"> 694</span>&#160;#<span class="keywordflow">if</span> defined(REINTERPRET_INPUT_AS_3D)</div><div class="line"><a name="l00695"></a><span class="lineno"> 695</span>&#160; ,</div><div class="line"><a name="l00696"></a><span class="lineno"> 696</span>&#160; uint lhs_cross_plane_pad</div><div class="line"><a name="l00697"></a><span class="lineno"> 697</span>&#160;#endif <span class="comment">// REINTERPRET_INPUT_AS_3D</span></div><div class="line"><a name="l00698"></a><span class="lineno"> 698</span>&#160;#<span class="keywordflow">if</span> defined(REINTERPRET_OUTPUT_AS_3D)</div><div class="line"><a name="l00699"></a><span class="lineno"> 699</span>&#160; ,</div><div class="line"><a name="l00700"></a><span class="lineno"> 700</span>&#160; uint dst_cross_plane_pad</div><div class="line"><a name="l00701"></a><span class="lineno"> 701</span>&#160;#endif <span class="comment">// REINTERPRET_OUTPUT_AS_3D</span></div><div class="line"><a name="l00702"></a><span class="lineno"> 702</span>&#160; )</div><div class="line"><a name="l00703"></a><span class="lineno"> 703</span>&#160;{</div><div class="line"><a name="l00704"></a><span class="lineno"> 704</span>&#160; <span class="comment">// Block size</span></div><div class="line"><a name="l00705"></a><span class="lineno"> 705</span>&#160;<span class="preprocessor">#define RHS_BLOCK_SIZE ((K0) * (N0))</span></div><div class="line"><a name="l00706"></a><span class="lineno"> 706</span>&#160;</div><div class="line"><a name="l00707"></a><span class="lineno"> 707</span>&#160; <span class="comment">// RHS offset and step X</span></div><div class="line"><a name="l00708"></a><span class="lineno"> 708</span>&#160;<span class="preprocessor">#if defined(RHS_INTERLEAVE)</span></div><div class="line"><a name="l00709"></a><span class="lineno"> 709</span>&#160;<span class="preprocessor">#define RHS_OFFSET_X (K0)</span></div><div class="line"><a name="l00710"></a><span class="lineno"> 710</span>&#160;<span class="preprocessor">#define RHS_STEP_X ((K0) * (H0))</span></div><div class="line"><a name="l00711"></a><span class="lineno"> 711</span>&#160;<span class="preprocessor">#define RHS_STEP_LOOP (1)</span></div><div class="line"><a name="l00712"></a><span class="lineno"> 712</span>&#160;<span class="preprocessor">#else // defined(RHS_INTERLEAVE)</span></div><div class="line"><a name="l00713"></a><span class="lineno"> 713</span>&#160;<span class="preprocessor">#define RHS_OFFSET_X (RHS_BLOCK_SIZE)</span></div><div class="line"><a name="l00714"></a><span class="lineno"> 714</span>&#160;<span class="preprocessor">#define RHS_STEP_X (K0)</span></div><div class="line"><a name="l00715"></a><span class="lineno"> 715</span>&#160;<span class="preprocessor">#define RHS_STEP_LOOP (H0)</span></div><div class="line"><a name="l00716"></a><span class="lineno"> 716</span>&#160;<span class="preprocessor">#endif // defined(RHS_INTERLEAVE)</span></div><div class="line"><a name="l00717"></a><span class="lineno"> 717</span>&#160;</div><div class="line"><a name="l00718"></a><span class="lineno"> 718</span>&#160; uint x = get_global_id(0);</div><div class="line"><a name="l00719"></a><span class="lineno"> 719</span>&#160; uint y = get_global_id(1);</div><div class="line"><a name="l00720"></a><span class="lineno"> 720</span>&#160; uint z = get_global_id(2);</div><div class="line"><a name="l00721"></a><span class="lineno"> 721</span>&#160;</div><div class="line"><a name="l00722"></a><span class="lineno"> 722</span>&#160;<span class="preprocessor">#if defined(DUMMY_WORK_ITEMS)</span></div><div class="line"><a name="l00723"></a><span class="lineno"> 723</span>&#160; <span class="keywordflow">if</span>((x * N0 &gt;= N) || (y * M0 &gt;= M))</div><div class="line"><a name="l00724"></a><span class="lineno"> 724</span>&#160; {</div><div class="line"><a name="l00725"></a><span class="lineno"> 725</span>&#160; <span class="keywordflow">return</span>;</div><div class="line"><a name="l00726"></a><span class="lineno"> 726</span>&#160; }</div><div class="line"><a name="l00727"></a><span class="lineno"> 727</span>&#160;<span class="preprocessor">#endif // defined(DUMMY_WORK_ITEMS)</span></div><div class="line"><a name="l00728"></a><span class="lineno"> 728</span>&#160;</div><div class="line"><a name="l00729"></a><span class="lineno"> 729</span>&#160; <span class="comment">// Compute LHS matrix address</span></div><div class="line"><a name="l00730"></a><span class="lineno"> 730</span>&#160; uint lhs_offset = lhs_offset_first_element_in_bytes + y * M0 * (uint)lhs_stride_y;</div><div class="line"><a name="l00731"></a><span class="lineno"> 731</span>&#160;</div><div class="line"><a name="l00732"></a><span class="lineno"> 732</span>&#160; <span class="comment">// Compute RHS matrix address</span></div><div class="line"><a name="l00733"></a><span class="lineno"> 733</span>&#160; uint rhs_offset = rhs_offset_first_element_in_bytes + (x % H0) * (uint)RHS_OFFSET_X + (x / (uint)H0) * rhs_stride_y;</div><div class="line"><a name="l00734"></a><span class="lineno"> 734</span>&#160;</div><div class="line"><a name="l00735"></a><span class="lineno"> 735</span>&#160;<span class="preprocessor">#if defined(MATRIX_B_DEPTH)</span></div><div class="line"><a name="l00736"></a><span class="lineno"> 736</span>&#160; <span class="comment">// Do not slide matrix B if the matrix B has 3 dimensions and matrix A more than 3</span></div><div class="line"><a name="l00737"></a><span class="lineno"> 737</span>&#160; rhs_offset += (z % MATRIX_B_DEPTH) * rhs_stride_z;</div><div class="line"><a name="l00738"></a><span class="lineno"> 738</span>&#160;<span class="preprocessor">#else // defined(MATRIX_B_DEPTH)</span></div><div class="line"><a name="l00739"></a><span class="lineno"> 739</span>&#160; rhs_offset += z * rhs_stride_z;</div><div class="line"><a name="l00740"></a><span class="lineno"> 740</span>&#160;<span class="preprocessor">#endif // defined(MATRIX_B_DEPTH)</span></div><div class="line"><a name="l00741"></a><span class="lineno"> 741</span>&#160;</div><div class="line"><a name="l00742"></a><span class="lineno"> 742</span>&#160; <a class="code" href="repeat_8h.xhtml#a4a61a8df72f9bc1b1dcc4cc4ec6ac3b9">REPEAT_VAR_INIT_TO_CONST</a>(8, uint, zlhs, 0); <span class="comment">//uint zout0=0,zout1=0,zout2=0,... zout7=0;</span></div><div class="line"><a name="l00743"></a><span class="lineno"> 743</span>&#160; <a class="code" href="repeat_8h.xhtml#a4a61a8df72f9bc1b1dcc4cc4ec6ac3b9">REPEAT_VAR_INIT_TO_CONST</a>(16, uint, zrhs, 0);</div><div class="line"><a name="l00744"></a><span class="lineno"> 744</span>&#160;</div><div class="line"><a name="l00745"></a><span class="lineno"> 745</span>&#160;<span class="preprocessor">#if defined(REINTERPRET_INPUT_AS_3D)</span></div><div class="line"><a name="l00746"></a><span class="lineno"> 746</span>&#160; <span class="comment">// The plane (zlhs) is calculated dividing M (y * M0) by HEIGHT_GEMM3D</span></div><div class="line"><a name="l00747"></a><span class="lineno"> 747</span>&#160; <a class="code" href="gemm__helpers_8h.xhtml#ade58486d6b2d13d7941106a9d2ec86d3">CALCULATE_Z_OFFSET</a>(M0, uint, zlhs, y, HEIGHT_GEMM3D, DEPTH_GEMM3D, lhs_cross_plane_pad, lhs_stride_y);</div><div class="line"><a name="l00748"></a><span class="lineno"> 748</span>&#160;</div><div class="line"><a name="l00749"></a><span class="lineno"> 749</span>&#160; <span class="comment">// Add offset for batched GEMM. The batches will be in the fourth dimension and for this reason we</span></div><div class="line"><a name="l00750"></a><span class="lineno"> 750</span>&#160; <span class="comment">// multiply lhs_stride_z by DEPTH_GEMM3D</span></div><div class="line"><a name="l00751"></a><span class="lineno"> 751</span>&#160; lhs_offset += z * lhs_stride_z * DEPTH_GEMM3D;</div><div class="line"><a name="l00752"></a><span class="lineno"> 752</span>&#160;</div><div class="line"><a name="l00753"></a><span class="lineno"> 753</span>&#160;<span class="preprocessor">#else // defined(REINTERPRET_INPUT_AS_3D)</span></div><div class="line"><a name="l00754"></a><span class="lineno"> 754</span>&#160;</div><div class="line"><a name="l00755"></a><span class="lineno"> 755</span>&#160; <span class="comment">// Add offset for batched GEMM</span></div><div class="line"><a name="l00756"></a><span class="lineno"> 756</span>&#160; lhs_offset += z * lhs_stride_z;</div><div class="line"><a name="l00757"></a><span class="lineno"> 757</span>&#160;</div><div class="line"><a name="l00758"></a><span class="lineno"> 758</span>&#160;<span class="preprocessor">#endif // defined(REINTERPRET_INPUT_AS_3D)</span></div><div class="line"><a name="l00759"></a><span class="lineno"> 759</span>&#160;</div><div class="line"><a name="l00760"></a><span class="lineno"> 760</span>&#160; <span class="comment">// Initialize the accumulators</span></div><div class="line"><a name="l00761"></a><span class="lineno"> 761</span>&#160; <a class="code" href="repeat_8h.xhtml#a4a61a8df72f9bc1b1dcc4cc4ec6ac3b9">REPEAT_VAR_INIT_TO_CONST</a>(M0, <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(uint, N0), c, 0); <span class="comment">//VEC_DATA_TYPE(uint, N0) c0=0,c1=0,c2=0,... c(N0-1)=0;</span></div><div class="line"><a name="l00762"></a><span class="lineno"> 762</span>&#160;</div><div class="line"><a name="l00763"></a><span class="lineno"> 763</span>&#160; <span class="keywordflow">for</span>(<span class="keywordtype">int</span> i = 0; i &lt; K; i += K0)</div><div class="line"><a name="l00764"></a><span class="lineno"> 764</span>&#160; {</div><div class="line"><a name="l00765"></a><span class="lineno"> 765</span>&#160; <span class="comment">// Load values from LHS matrix</span></div><div class="line"><a name="l00766"></a><span class="lineno"> 766</span>&#160; <a class="code" href="gemm__helpers_8h.xhtml#a9fe7ded27e552f48fa285d34fb00427d">LOAD_BLOCK</a>(M0, K0, uchar, a, lhs_ptr, lhs_offset, lhs_stride_y, zlhs);</div><div class="line"><a name="l00767"></a><span class="lineno"> 767</span>&#160;</div><div class="line"><a name="l00768"></a><span class="lineno"> 768</span>&#160; <span class="comment">// Load values from RHS matrix</span></div><div class="line"><a name="l00769"></a><span class="lineno"> 769</span>&#160; <a class="code" href="gemm__helpers_8h.xhtml#a9fe7ded27e552f48fa285d34fb00427d">LOAD_BLOCK</a>(N0, K0, uchar, <a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#aa76b4a6e74940dabc5b7fc6b2dab3545">b</a>, rhs_ptr, rhs_offset, RHS_STEP_X, zrhs);</div><div class="line"><a name="l00770"></a><span class="lineno"> 770</span>&#160;</div><div class="line"><a name="l00771"></a><span class="lineno"> 771</span>&#160; <span class="comment">// Partial matrix multiplication M0,N0,K0</span></div><div class="line"><a name="l00772"></a><span class="lineno"> 772</span>&#160; <a class="code" href="gemmlowp_8cl.xhtml#af110f6648cbcc21216a66fb3342632b0">ARM_MM_K0XN0XM0</a>(M0, N0, K0, a, <a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#aa76b4a6e74940dabc5b7fc6b2dab3545">b</a>, c);</div><div class="line"><a name="l00773"></a><span class="lineno"> 773</span>&#160;</div><div class="line"><a name="l00774"></a><span class="lineno"> 774</span>&#160; lhs_offset += K0;</div><div class="line"><a name="l00775"></a><span class="lineno"> 775</span>&#160; rhs_offset += N0 * RHS_STEP_X * RHS_STEP_LOOP;</div><div class="line"><a name="l00776"></a><span class="lineno"> 776</span>&#160; }</div><div class="line"><a name="l00777"></a><span class="lineno"> 777</span>&#160;</div><div class="line"><a name="l00778"></a><span class="lineno"> 778</span>&#160; __global uchar *dst_addr = dst_ptr + dst_offset_first_element_in_bytes + (x * (uint)N0) * <span class="keyword">sizeof</span>(int) + (y * (uint)M0 * dst_stride_y);</div><div class="line"><a name="l00779"></a><span class="lineno"> 779</span>&#160;</div><div class="line"><a name="l00780"></a><span class="lineno"> 780</span>&#160; <a class="code" href="repeat_8h.xhtml#a4a61a8df72f9bc1b1dcc4cc4ec6ac3b9">REPEAT_VAR_INIT_TO_CONST</a>(8, uint, zout, 0); <span class="comment">//uint zout0=0,zout1=0,zout2=0,... zout7=0;</span></div><div class="line"><a name="l00781"></a><span class="lineno"> 781</span>&#160;</div><div class="line"><a name="l00782"></a><span class="lineno"> 782</span>&#160;<span class="preprocessor">#if defined(REINTERPRET_OUTPUT_AS_3D)</span></div><div class="line"><a name="l00783"></a><span class="lineno"> 783</span>&#160; <span class="comment">// The plane (zout) is calculated dividing M (y * M0) by HEIGHT_GEMM3D</span></div><div class="line"><a name="l00784"></a><span class="lineno"> 784</span>&#160; <a class="code" href="gemm__helpers_8h.xhtml#ade58486d6b2d13d7941106a9d2ec86d3">CALCULATE_Z_OFFSET</a>(M0, uint, zout, y, HEIGHT_GEMM3D, DEPTH_GEMM3D, dst_cross_plane_pad, dst_stride_y);</div><div class="line"><a name="l00785"></a><span class="lineno"> 785</span>&#160;</div><div class="line"><a name="l00786"></a><span class="lineno"> 786</span>&#160; <span class="comment">// Add offset for batched GEMM. The batches will be in the fourth dimension and for this reason we</span></div><div class="line"><a name="l00787"></a><span class="lineno"> 787</span>&#160; <span class="comment">// multiply dst_stride_z by DEPTH_GEMM3D</span></div><div class="line"><a name="l00788"></a><span class="lineno"> 788</span>&#160; dst_addr += z * dst_stride_z * DEPTH_GEMM3D;</div><div class="line"><a name="l00789"></a><span class="lineno"> 789</span>&#160;</div><div class="line"><a name="l00790"></a><span class="lineno"> 790</span>&#160;<span class="preprocessor">#else // defined(REINTERPRET_OUTPUT_AS_3D)</span></div><div class="line"><a name="l00791"></a><span class="lineno"> 791</span>&#160;</div><div class="line"><a name="l00792"></a><span class="lineno"> 792</span>&#160; <span class="comment">// Add offset for batched GEMM</span></div><div class="line"><a name="l00793"></a><span class="lineno"> 793</span>&#160; dst_addr += z * dst_stride_z;</div><div class="line"><a name="l00794"></a><span class="lineno"> 794</span>&#160;</div><div class="line"><a name="l00795"></a><span class="lineno"> 795</span>&#160;<span class="preprocessor">#endif // defined(REINTERPRET_OUTPUT_AS_3D)</span></div><div class="line"><a name="l00796"></a><span class="lineno"> 796</span>&#160;</div><div class="line"><a name="l00797"></a><span class="lineno"> 797</span>&#160; <span class="comment">// Convert and store output block</span></div><div class="line"><a name="l00798"></a><span class="lineno"> 798</span>&#160; <a class="code" href="gemm__helpers_8h.xhtml#af5c0a8dfefbd611d38bb91212684ef7d">CONVERT_STORE_BLOCK</a>(M0, N0, <span class="keywordtype">int</span>, c, dst_addr, dst_stride_y, zout);</div><div class="line"><a name="l00799"></a><span class="lineno"> 799</span>&#160;</div><div class="line"><a name="l00800"></a><span class="lineno"> 800</span>&#160;<span class="preprocessor">#undef RHS_BLOCK_SIZE</span></div><div class="line"><a name="l00801"></a><span class="lineno"> 801</span>&#160;<span class="preprocessor">#undef RHS_OFFSET_X</span></div><div class="line"><a name="l00802"></a><span class="lineno"> 802</span>&#160;<span class="preprocessor">#undef RHS_STEP_X</span></div><div class="line"><a name="l00803"></a><span class="lineno"> 803</span>&#160;}</div><div class="line"><a name="l00804"></a><span class="lineno"> 804</span>&#160;<span class="preprocessor">#endif // defined(M0) &amp;&amp; defined(N0) &amp;&amp; defined(K0) &amp;&amp; defined(H0) &amp;&amp; defined(DATA_TYPE) &amp;&amp; defined(K)</span></div><div class="line"><a name="l00805"></a><span class="lineno"> 805</span>&#160;</div><div class="line"><a name="l00806"></a><span class="lineno"> 806</span>&#160;<span class="preprocessor">#if defined(M0) &amp;&amp; defined(N0) &amp;&amp; defined(K0) &amp;&amp; defined(K)</span></div><div class="line"><a name="l00807"></a><span class="lineno"> 807</span>&#160;</div><div class="line"><a name="l00852"></a><span class="lineno"> 852</span>&#160;__kernel <span class="keywordtype">void</span> gemmlowp_mm_native(<a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a22f42fcf2077d951271df83b55c1a71a">IMAGE_DECLARATION</a>(lhs),</div><div class="line"><a name="l00853"></a><span class="lineno"> 853</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a22f42fcf2077d951271df83b55c1a71a">IMAGE_DECLARATION</a>(rhs),</div><div class="line"><a name="l00854"></a><span class="lineno"> 854</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a22f42fcf2077d951271df83b55c1a71a">IMAGE_DECLARATION</a>(<a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#adbf67dcee294e673cf796f1ed8aeb6a4">dst</a>),</div><div class="line"><a name="l00855"></a><span class="lineno"> 855</span>&#160; uint lhs_stride_z,</div><div class="line"><a name="l00856"></a><span class="lineno"> 856</span>&#160; uint rhs_stride_z,</div><div class="line"><a name="l00857"></a><span class="lineno"> 857</span>&#160; uint dst_stride_z</div><div class="line"><a name="l00858"></a><span class="lineno"> 858</span>&#160;#<span class="keywordflow">if</span> defined(REINTERPRET_INPUT_AS_3D)</div><div class="line"><a name="l00859"></a><span class="lineno"> 859</span>&#160; ,</div><div class="line"><a name="l00860"></a><span class="lineno"> 860</span>&#160; uint lhs_cross_plane_pad</div><div class="line"><a name="l00861"></a><span class="lineno"> 861</span>&#160;#endif <span class="comment">// REINTERPRET_INPUT_AS_3D</span></div><div class="line"><a name="l00862"></a><span class="lineno"> 862</span>&#160;#<span class="keywordflow">if</span> defined(REINTERPRET_OUTPUT_AS_3D)</div><div class="line"><a name="l00863"></a><span class="lineno"> 863</span>&#160; ,</div><div class="line"><a name="l00864"></a><span class="lineno"> 864</span>&#160; uint dst_cross_plane_pad</div><div class="line"><a name="l00865"></a><span class="lineno"> 865</span>&#160;#endif <span class="comment">// REINTERPRET_OUTPUT_AS_3D</span></div><div class="line"><a name="l00866"></a><span class="lineno"> 866</span>&#160; )</div><div class="line"><a name="l00867"></a><span class="lineno"> 867</span>&#160;{</div><div class="line"><a name="l00868"></a><span class="lineno"> 868</span>&#160; uint x = get_global_id(0);</div><div class="line"><a name="l00869"></a><span class="lineno"> 869</span>&#160; uint y = get_global_id(1);</div><div class="line"><a name="l00870"></a><span class="lineno"> 870</span>&#160; uint z = get_global_id(2);</div><div class="line"><a name="l00871"></a><span class="lineno"> 871</span>&#160;</div><div class="line"><a name="l00872"></a><span class="lineno"> 872</span>&#160;<span class="preprocessor">#if defined(DUMMY_WORK_ITEMS)</span></div><div class="line"><a name="l00873"></a><span class="lineno"> 873</span>&#160; <span class="keywordflow">if</span>((x * N0 &gt;= N) || (y * M0 &gt;= M))</div><div class="line"><a name="l00874"></a><span class="lineno"> 874</span>&#160; {</div><div class="line"><a name="l00875"></a><span class="lineno"> 875</span>&#160; <span class="keywordflow">return</span>;</div><div class="line"><a name="l00876"></a><span class="lineno"> 876</span>&#160; }</div><div class="line"><a name="l00877"></a><span class="lineno"> 877</span>&#160;<span class="preprocessor">#endif // defined(DUMMY_WORK_ITEMS)</span></div><div class="line"><a name="l00878"></a><span class="lineno"> 878</span>&#160;</div><div class="line"><a name="l00879"></a><span class="lineno"> 879</span>&#160; <span class="comment">// Compute LHS matrix address</span></div><div class="line"><a name="l00880"></a><span class="lineno"> 880</span>&#160; uint lhs_offset = lhs_offset_first_element_in_bytes + y * M0 * (uint)lhs_stride_y;</div><div class="line"><a name="l00881"></a><span class="lineno"> 881</span>&#160;</div><div class="line"><a name="l00882"></a><span class="lineno"> 882</span>&#160; <span class="comment">// Compute RHS matrix address</span></div><div class="line"><a name="l00883"></a><span class="lineno"> 883</span>&#160; uint rhs_offset = rhs_offset_first_element_in_bytes + x * N0;</div><div class="line"><a name="l00884"></a><span class="lineno"> 884</span>&#160;</div><div class="line"><a name="l00885"></a><span class="lineno"> 885</span>&#160;<span class="preprocessor">#if defined(MATRIX_B_DEPTH)</span></div><div class="line"><a name="l00886"></a><span class="lineno"> 886</span>&#160; <span class="comment">// Do not slide matrix B if the matrix B has 3 dimensions and matrix A more than 3</span></div><div class="line"><a name="l00887"></a><span class="lineno"> 887</span>&#160; rhs_offset += (z % MATRIX_B_DEPTH) * rhs_stride_z;</div><div class="line"><a name="l00888"></a><span class="lineno"> 888</span>&#160;<span class="preprocessor">#else // defined(MATRIX_B_DEPTH)</span></div><div class="line"><a name="l00889"></a><span class="lineno"> 889</span>&#160; rhs_offset += z * rhs_stride_z;</div><div class="line"><a name="l00890"></a><span class="lineno"> 890</span>&#160;<span class="preprocessor">#endif // defined(MATRIX_B_DEPTH)</span></div><div class="line"><a name="l00891"></a><span class="lineno"> 891</span>&#160;</div><div class="line"><a name="l00892"></a><span class="lineno"> 892</span>&#160; <a class="code" href="repeat_8h.xhtml#a4a61a8df72f9bc1b1dcc4cc4ec6ac3b9">REPEAT_VAR_INIT_TO_CONST</a>(8, uint, zlhs, 0);</div><div class="line"><a name="l00893"></a><span class="lineno"> 893</span>&#160; <a class="code" href="repeat_8h.xhtml#a4a61a8df72f9bc1b1dcc4cc4ec6ac3b9">REPEAT_VAR_INIT_TO_CONST</a>(16, uint, zrhs, 0);</div><div class="line"><a name="l00894"></a><span class="lineno"> 894</span>&#160;</div><div class="line"><a name="l00895"></a><span class="lineno"> 895</span>&#160;<span class="preprocessor">#if defined(REINTERPRET_INPUT_AS_3D)</span></div><div class="line"><a name="l00896"></a><span class="lineno"> 896</span>&#160; <span class="comment">// The plane (zlhs) is calculated dividing M (y * M0) by HEIGHT_GEMM3D</span></div><div class="line"><a name="l00897"></a><span class="lineno"> 897</span>&#160; <a class="code" href="gemm__helpers_8h.xhtml#ade58486d6b2d13d7941106a9d2ec86d3">CALCULATE_Z_OFFSET</a>(M0, uint, zlhs, y, HEIGHT_GEMM3D, DEPTH_GEMM3D, lhs_cross_plane_pad, lhs_stride_y);</div><div class="line"><a name="l00898"></a><span class="lineno"> 898</span>&#160;</div><div class="line"><a name="l00899"></a><span class="lineno"> 899</span>&#160; <span class="comment">// Add offset for batched GEMM. The batches will be in the fourth dimension and for this reason we</span></div><div class="line"><a name="l00900"></a><span class="lineno"> 900</span>&#160; <span class="comment">// multiply lhs_stride_z by DEPTH_GEMM3D</span></div><div class="line"><a name="l00901"></a><span class="lineno"> 901</span>&#160; lhs_offset += z * lhs_stride_z * DEPTH_GEMM3D;</div><div class="line"><a name="l00902"></a><span class="lineno"> 902</span>&#160;</div><div class="line"><a name="l00903"></a><span class="lineno"> 903</span>&#160;<span class="preprocessor">#else // defined(REINTERPRET_INPUT_AS_3D)</span></div><div class="line"><a name="l00904"></a><span class="lineno"> 904</span>&#160;</div><div class="line"><a name="l00905"></a><span class="lineno"> 905</span>&#160; <span class="comment">// Add offset for batched GEMM</span></div><div class="line"><a name="l00906"></a><span class="lineno"> 906</span>&#160; lhs_offset += z * lhs_stride_z;</div><div class="line"><a name="l00907"></a><span class="lineno"> 907</span>&#160;</div><div class="line"><a name="l00908"></a><span class="lineno"> 908</span>&#160;<span class="preprocessor">#endif // defined(REINTERPRET_INPUT_AS_3D)</span></div><div class="line"><a name="l00909"></a><span class="lineno"> 909</span>&#160;</div><div class="line"><a name="l00910"></a><span class="lineno"> 910</span>&#160; <span class="comment">// Initialize the accumulators</span></div><div class="line"><a name="l00911"></a><span class="lineno"> 911</span>&#160; <a class="code" href="repeat_8h.xhtml#a4a61a8df72f9bc1b1dcc4cc4ec6ac3b9">REPEAT_VAR_INIT_TO_CONST</a>(M0, <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(uint, N0), c, 0); <span class="comment">//VEC_DATA_TYPE(uint, N0) c0=0,c1=0,c2=0,... c(M0-1)=0;</span></div><div class="line"><a name="l00912"></a><span class="lineno"> 912</span>&#160;</div><div class="line"><a name="l00913"></a><span class="lineno"> 913</span>&#160; <span class="keywordtype">int</span> i = 0;</div><div class="line"><a name="l00914"></a><span class="lineno"> 914</span>&#160;</div><div class="line"><a name="l00915"></a><span class="lineno"> 915</span>&#160; <span class="keywordflow">for</span>(; i &lt;= (K - K0); i += K0)</div><div class="line"><a name="l00916"></a><span class="lineno"> 916</span>&#160; {</div><div class="line"><a name="l00917"></a><span class="lineno"> 917</span>&#160; <span class="comment">// Load values from LHS matrix</span></div><div class="line"><a name="l00918"></a><span class="lineno"> 918</span>&#160; <a class="code" href="gemm__helpers_8h.xhtml#a9fe7ded27e552f48fa285d34fb00427d">LOAD_BLOCK</a>(M0, K0, uchar, a, lhs_ptr, lhs_offset, lhs_stride_y, zlhs);</div><div class="line"><a name="l00919"></a><span class="lineno"> 919</span>&#160;</div><div class="line"><a name="l00920"></a><span class="lineno"> 920</span>&#160; <span class="comment">// Load values from RHS matrix</span></div><div class="line"><a name="l00921"></a><span class="lineno"> 921</span>&#160; <a class="code" href="gemm__helpers_8h.xhtml#a9fe7ded27e552f48fa285d34fb00427d">LOAD_BLOCK</a>(K0, N0, uchar, <a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#aa76b4a6e74940dabc5b7fc6b2dab3545">b</a>, rhs_ptr, rhs_offset, rhs_stride_y, zrhs);</div><div class="line"><a name="l00922"></a><span class="lineno"> 922</span>&#160;</div><div class="line"><a name="l00923"></a><span class="lineno"> 923</span>&#160; <span class="comment">// Transpose the values from RHS matrix</span></div><div class="line"><a name="l00924"></a><span class="lineno"> 924</span>&#160; <a class="code" href="gemm__helpers_8h.xhtml#af83fba9e0a00fc38a71258f0052b9c24">TRANSPOSE_K0XN0</a>(K0, N0, b_t, <a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#aa76b4a6e74940dabc5b7fc6b2dab3545">b</a>);</div><div class="line"><a name="l00925"></a><span class="lineno"> 925</span>&#160;</div><div class="line"><a name="l00926"></a><span class="lineno"> 926</span>&#160; <span class="comment">// Partial matrix multiplication M0,N0,K0</span></div><div class="line"><a name="l00927"></a><span class="lineno"> 927</span>&#160; <a class="code" href="gemmlowp_8cl.xhtml#af110f6648cbcc21216a66fb3342632b0">ARM_MM_K0XN0XM0</a>(M0, N0, K0, a, b_t, c);</div><div class="line"><a name="l00928"></a><span class="lineno"> 928</span>&#160;</div><div class="line"><a name="l00929"></a><span class="lineno"> 929</span>&#160; <span class="comment">// Update the offset</span></div><div class="line"><a name="l00930"></a><span class="lineno"> 930</span>&#160; lhs_offset += K0;</div><div class="line"><a name="l00931"></a><span class="lineno"> 931</span>&#160; rhs_offset += K0 * rhs_stride_y;</div><div class="line"><a name="l00932"></a><span class="lineno"> 932</span>&#160; }</div><div class="line"><a name="l00933"></a><span class="lineno"> 933</span>&#160;</div><div class="line"><a name="l00934"></a><span class="lineno"> 934</span>&#160; <span class="comment">// Left-over for loop</span></div><div class="line"><a name="l00935"></a><span class="lineno"> 935</span>&#160; <span class="keywordflow">for</span>(; i &lt; K; ++i)</div><div class="line"><a name="l00936"></a><span class="lineno"> 936</span>&#160; {</div><div class="line"><a name="l00937"></a><span class="lineno"> 937</span>&#160; <span class="comment">// Load values from LHS matrix</span></div><div class="line"><a name="l00938"></a><span class="lineno"> 938</span>&#160; <a class="code" href="gemm__helpers_8h.xhtml#a9fe7ded27e552f48fa285d34fb00427d">LOAD_BLOCK</a>(M0, 1, uchar, a, lhs_ptr, lhs_offset, lhs_stride_y, zlhs);</div><div class="line"><a name="l00939"></a><span class="lineno"> 939</span>&#160;</div><div class="line"><a name="l00940"></a><span class="lineno"> 940</span>&#160; <span class="comment">// Load values from RHS matrix</span></div><div class="line"><a name="l00941"></a><span class="lineno"> 941</span>&#160; <a class="code" href="gemm__helpers_8h.xhtml#a9fe7ded27e552f48fa285d34fb00427d">LOAD_BLOCK</a>(1, N0, uchar, <a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#aa76b4a6e74940dabc5b7fc6b2dab3545">b</a>, rhs_ptr, rhs_offset, rhs_stride_y, zrhs);</div><div class="line"><a name="l00942"></a><span class="lineno"> 942</span>&#160;</div><div class="line"><a name="l00943"></a><span class="lineno"> 943</span>&#160; <span class="comment">// Transpose the values from RHS matrix</span></div><div class="line"><a name="l00944"></a><span class="lineno"> 944</span>&#160; <a class="code" href="gemm__helpers_8h.xhtml#af83fba9e0a00fc38a71258f0052b9c24">TRANSPOSE_K0XN0</a>(1, N0, b_t, <a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#aa76b4a6e74940dabc5b7fc6b2dab3545">b</a>);</div><div class="line"><a name="l00945"></a><span class="lineno"> 945</span>&#160;</div><div class="line"><a name="l00946"></a><span class="lineno"> 946</span>&#160; <span class="comment">// Partial matrix multiplication M0,N0,1</span></div><div class="line"><a name="l00947"></a><span class="lineno"> 947</span>&#160; <a class="code" href="gemmlowp_8cl.xhtml#af110f6648cbcc21216a66fb3342632b0">ARM_MM_K0XN0XM0</a>(M0, N0, 1, a, b_t, c);</div><div class="line"><a name="l00948"></a><span class="lineno"> 948</span>&#160;</div><div class="line"><a name="l00949"></a><span class="lineno"> 949</span>&#160; <span class="comment">// Update the offset</span></div><div class="line"><a name="l00950"></a><span class="lineno"> 950</span>&#160; lhs_offset += 1;</div><div class="line"><a name="l00951"></a><span class="lineno"> 951</span>&#160; rhs_offset += rhs_stride_y;</div><div class="line"><a name="l00952"></a><span class="lineno"> 952</span>&#160; }</div><div class="line"><a name="l00953"></a><span class="lineno"> 953</span>&#160;</div><div class="line"><a name="l00954"></a><span class="lineno"> 954</span>&#160; __global uchar *dst_addr = dst_ptr + dst_offset_first_element_in_bytes + (x * (uint)N0) * <span class="keyword">sizeof</span>(int) + (y * (uint)M0 * dst_stride_y);</div><div class="line"><a name="l00955"></a><span class="lineno"> 955</span>&#160;</div><div class="line"><a name="l00956"></a><span class="lineno"> 956</span>&#160; <a class="code" href="repeat_8h.xhtml#a4a61a8df72f9bc1b1dcc4cc4ec6ac3b9">REPEAT_VAR_INIT_TO_CONST</a>(M0, uint, zout, 0); <span class="comment">//uint zout0=0,zout1=0,zout2=0,... zout7=0;</span></div><div class="line"><a name="l00957"></a><span class="lineno"> 957</span>&#160;</div><div class="line"><a name="l00958"></a><span class="lineno"> 958</span>&#160;<span class="preprocessor">#if defined(REINTERPRET_OUTPUT_AS_3D)</span></div><div class="line"><a name="l00959"></a><span class="lineno"> 959</span>&#160; <span class="comment">// The plane (zout) is calculated dividing M (y * M0) by HEIGHT_GEMM3D</span></div><div class="line"><a name="l00960"></a><span class="lineno"> 960</span>&#160; <a class="code" href="gemm__helpers_8h.xhtml#ade58486d6b2d13d7941106a9d2ec86d3">CALCULATE_Z_OFFSET</a>(M0, uint, zout, y, HEIGHT_GEMM3D, DEPTH_GEMM3D, dst_cross_plane_pad, dst_stride_y);</div><div class="line"><a name="l00961"></a><span class="lineno"> 961</span>&#160;</div><div class="line"><a name="l00962"></a><span class="lineno"> 962</span>&#160; <span class="comment">// Add offset for batched GEMM. The batches will be in the fourth dimension and for this reason we</span></div><div class="line"><a name="l00963"></a><span class="lineno"> 963</span>&#160; <span class="comment">// multiply dst_stride_z by DEPTH_GEMM3D</span></div><div class="line"><a name="l00964"></a><span class="lineno"> 964</span>&#160; dst_addr += z * dst_stride_z * DEPTH_GEMM3D;</div><div class="line"><a name="l00965"></a><span class="lineno"> 965</span>&#160;</div><div class="line"><a name="l00966"></a><span class="lineno"> 966</span>&#160;<span class="preprocessor">#else // defined(REINTERPRET_OUTPUT_AS_3D)</span></div><div class="line"><a name="l00967"></a><span class="lineno"> 967</span>&#160;</div><div class="line"><a name="l00968"></a><span class="lineno"> 968</span>&#160; <span class="comment">// Add offset for batched GEMM</span></div><div class="line"><a name="l00969"></a><span class="lineno"> 969</span>&#160; dst_addr += z * dst_stride_z;</div><div class="line"><a name="l00970"></a><span class="lineno"> 970</span>&#160;</div><div class="line"><a name="l00971"></a><span class="lineno"> 971</span>&#160;<span class="preprocessor">#endif // defined(REINTERPRET_OUTPUT_AS_3D)</span></div><div class="line"><a name="l00972"></a><span class="lineno"> 972</span>&#160;</div><div class="line"><a name="l00973"></a><span class="lineno"> 973</span>&#160; <span class="comment">// Convert and store output block</span></div><div class="line"><a name="l00974"></a><span class="lineno"> 974</span>&#160; <a class="code" href="gemm__helpers_8h.xhtml#af5c0a8dfefbd611d38bb91212684ef7d">CONVERT_STORE_BLOCK</a>(M0, N0, <span class="keywordtype">int</span>, c, dst_addr, dst_stride_y, zout);</div><div class="line"><a name="l00975"></a><span class="lineno"> 975</span>&#160;}</div><div class="line"><a name="l00976"></a><span class="lineno"> 976</span>&#160;<span class="preprocessor">#endif // defined(M0) &amp;&amp; defined(N0) &amp;&amp; defined(K0) &amp;&amp; defined(K)</span></div><div class="line"><a name="l00977"></a><span class="lineno"> 977</span>&#160;</div><div class="line"><a name="l00978"></a><span class="lineno"> 978</span>&#160;<span class="preprocessor">#if defined(COLS_A)</span></div><div class="line"><a name="l00979"></a><span class="lineno"> 979</span>&#160;</div><div class="line"><a name="l01001"></a><span class="lineno"> 1001</span>&#160;__kernel <span class="keywordtype">void</span> gemmlowp_matrix_a_reduction(<a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a6b83038822d1ae7ab619b684ed3b7fc0">TENSOR3D_DECLARATION</a>(<a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#a989ab3e96426615bb98e04e0235088ca">src</a>),</div><div class="line"><a name="l01002"></a><span class="lineno"> 1002</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a22f42fcf2077d951271df83b55c1a71a">IMAGE_DECLARATION</a>(<a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#adbf67dcee294e673cf796f1ed8aeb6a4">dst</a>))</div><div class="line"><a name="l01003"></a><span class="lineno"> 1003</span>&#160;{</div><div class="line"><a name="l01004"></a><span class="lineno"> 1004</span>&#160; <span class="comment">// Compute source and destination addresses</span></div><div class="line"><a name="l01005"></a><span class="lineno"> 1005</span>&#160; <a class="code" href="struct_tensor3_d.xhtml">Tensor3D</a> <a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#a989ab3e96426615bb98e04e0235088ca">src</a> = <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a31c8c760f08fb1a331b16b7c204321dc">CONVERT_TO_TENSOR3D_STRUCT</a>(<a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#a989ab3e96426615bb98e04e0235088ca">src</a>);</div><div class="line"><a name="l01006"></a><span class="lineno"> 1006</span>&#160; <a class="code" href="struct_image.xhtml">Image</a> <a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#adbf67dcee294e673cf796f1ed8aeb6a4">dst</a> = <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#aebe814363556c244be043b13e7969197">CONVERT_TO_IMAGE_STRUCT</a>(<a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#adbf67dcee294e673cf796f1ed8aeb6a4">dst</a>);</div><div class="line"><a name="l01007"></a><span class="lineno"> 1007</span>&#160;</div><div class="line"><a name="l01008"></a><span class="lineno"> 1008</span>&#160; uint4 sum_row_u32 = (uint4)0;</div><div class="line"><a name="l01009"></a><span class="lineno"> 1009</span>&#160; uint sum_row = 0;</div><div class="line"><a name="l01010"></a><span class="lineno"> 1010</span>&#160;</div><div class="line"><a name="l01011"></a><span class="lineno"> 1011</span>&#160; __global <span class="keyword">const</span> uchar *matrix_a = (__global <span class="keyword">const</span> uchar *)(<a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#a989ab3e96426615bb98e04e0235088ca">src</a>.ptr + get_global_id(0) * src_stride_y + get_global_id(1) * src_stride_z);</div><div class="line"><a name="l01012"></a><span class="lineno"> 1012</span>&#160;</div><div class="line"><a name="l01013"></a><span class="lineno"> 1013</span>&#160; <span class="keywordtype">int</span> i = 0;</div><div class="line"><a name="l01014"></a><span class="lineno"> 1014</span>&#160;</div><div class="line"><a name="l01015"></a><span class="lineno"> 1015</span>&#160; <span class="comment">// This for loop performs 16 accumulations</span></div><div class="line"><a name="l01016"></a><span class="lineno"> 1016</span>&#160; <span class="keywordflow">for</span>(; i &lt;= ((int)COLS_A - 16); i += 16)</div><div class="line"><a name="l01017"></a><span class="lineno"> 1017</span>&#160; {</div><div class="line"><a name="l01018"></a><span class="lineno"> 1018</span>&#160; <span class="keyword">const</span> uchar16 a0_u8 = vload16(0, matrix_a + i);</div><div class="line"><a name="l01019"></a><span class="lineno"> 1019</span>&#160;</div><div class="line"><a name="l01020"></a><span class="lineno"> 1020</span>&#160; sum_row_u32 += convert_uint4(a0_u8.s0123) + convert_uint4(a0_u8.s4567) + convert_uint4(a0_u8.s89AB) + convert_uint4(a0_u8.sCDEF);</div><div class="line"><a name="l01021"></a><span class="lineno"> 1021</span>&#160; }</div><div class="line"><a name="l01022"></a><span class="lineno"> 1022</span>&#160;</div><div class="line"><a name="l01023"></a><span class="lineno"> 1023</span>&#160; <span class="comment">// This for loop performs the leftover accumulations</span></div><div class="line"><a name="l01024"></a><span class="lineno"> 1024</span>&#160; <span class="keywordflow">for</span>(; i &lt; COLS_A; ++i)</div><div class="line"><a name="l01025"></a><span class="lineno"> 1025</span>&#160; {</div><div class="line"><a name="l01026"></a><span class="lineno"> 1026</span>&#160; sum_row += matrix_a[i];</div><div class="line"><a name="l01027"></a><span class="lineno"> 1027</span>&#160; }</div><div class="line"><a name="l01028"></a><span class="lineno"> 1028</span>&#160;</div><div class="line"><a name="l01029"></a><span class="lineno"> 1029</span>&#160; sum_row += sum_row_u32.s0 + sum_row_u32.s1 + sum_row_u32.s2 + sum_row_u32.s3;</div><div class="line"><a name="l01030"></a><span class="lineno"> 1030</span>&#160;</div><div class="line"><a name="l01031"></a><span class="lineno"> 1031</span>&#160; *((__global <span class="keywordtype">int</span> *)<a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#adbf67dcee294e673cf796f1ed8aeb6a4">dst</a>.ptr) = (int)sum_row;</div><div class="line"><a name="l01032"></a><span class="lineno"> 1032</span>&#160;}</div><div class="line"><a name="l01033"></a><span class="lineno"> 1033</span>&#160;</div><div class="line"><a name="l01034"></a><span class="lineno"> 1034</span>&#160;<span class="preprocessor">#if defined(ARM_COMPUTE_OPENCL_DOT8_ENABLED) &amp;&amp; defined(cl_arm_integer_dot_product_int8)</span></div><div class="line"><a name="l01035"></a><span class="lineno"> 1035</span>&#160;</div><div class="line"><a name="l01057"></a><span class="lineno"> 1057</span>&#160;__kernel <span class="keywordtype">void</span> gemmlowp_matrix_a_reduction_dot8(<a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a6b83038822d1ae7ab619b684ed3b7fc0">TENSOR3D_DECLARATION</a>(<a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#a989ab3e96426615bb98e04e0235088ca">src</a>),</div><div class="line"><a name="l01058"></a><span class="lineno"> 1058</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a22f42fcf2077d951271df83b55c1a71a">IMAGE_DECLARATION</a>(<a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#adbf67dcee294e673cf796f1ed8aeb6a4">dst</a>))</div><div class="line"><a name="l01059"></a><span class="lineno"> 1059</span>&#160;{</div><div class="line"><a name="l01060"></a><span class="lineno"> 1060</span>&#160; <span class="comment">// Compute source and destination addresses</span></div><div class="line"><a name="l01061"></a><span class="lineno"> 1061</span>&#160; <a class="code" href="struct_tensor3_d.xhtml">Tensor3D</a> <a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#a989ab3e96426615bb98e04e0235088ca">src</a> = <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a31c8c760f08fb1a331b16b7c204321dc">CONVERT_TO_TENSOR3D_STRUCT</a>(<a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#a989ab3e96426615bb98e04e0235088ca">src</a>);</div><div class="line"><a name="l01062"></a><span class="lineno"> 1062</span>&#160; <a class="code" href="struct_image.xhtml">Image</a> <a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#adbf67dcee294e673cf796f1ed8aeb6a4">dst</a> = <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#aebe814363556c244be043b13e7969197">CONVERT_TO_IMAGE_STRUCT</a>(<a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#adbf67dcee294e673cf796f1ed8aeb6a4">dst</a>);</div><div class="line"><a name="l01063"></a><span class="lineno"> 1063</span>&#160;</div><div class="line"><a name="l01064"></a><span class="lineno"> 1064</span>&#160; uint sum_row = 0;</div><div class="line"><a name="l01065"></a><span class="lineno"> 1065</span>&#160;</div><div class="line"><a name="l01066"></a><span class="lineno"> 1066</span>&#160; __global <span class="keyword">const</span> uchar *matrix_a = (__global <span class="keyword">const</span> uchar *)(<a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#a989ab3e96426615bb98e04e0235088ca">src</a>.ptr + get_global_id(0) * src_stride_y + get_global_id(1) * src_stride_z);</div><div class="line"><a name="l01067"></a><span class="lineno"> 1067</span>&#160;</div><div class="line"><a name="l01068"></a><span class="lineno"> 1068</span>&#160; <span class="keywordtype">int</span> i = 0;</div><div class="line"><a name="l01069"></a><span class="lineno"> 1069</span>&#160;</div><div class="line"><a name="l01070"></a><span class="lineno"> 1070</span>&#160; <span class="comment">// This for loop performs 16 accumulations</span></div><div class="line"><a name="l01071"></a><span class="lineno"> 1071</span>&#160; <span class="keywordflow">for</span>(; i &lt;= ((int)COLS_A - 32); i += 32)</div><div class="line"><a name="l01072"></a><span class="lineno"> 1072</span>&#160; {</div><div class="line"><a name="l01073"></a><span class="lineno"> 1073</span>&#160; uchar16 a0_u8 = vload16(0, matrix_a + i);</div><div class="line"><a name="l01074"></a><span class="lineno"> 1074</span>&#160;</div><div class="line"><a name="l01075"></a><span class="lineno"> 1075</span>&#160; sum_row += arm_dot(a0_u8.s0123, (uchar4)(1));</div><div class="line"><a name="l01076"></a><span class="lineno"> 1076</span>&#160; sum_row += arm_dot(a0_u8.s4567, (uchar4)(1));</div><div class="line"><a name="l01077"></a><span class="lineno"> 1077</span>&#160; sum_row += arm_dot(a0_u8.s89AB, (uchar4)(1));</div><div class="line"><a name="l01078"></a><span class="lineno"> 1078</span>&#160; sum_row += arm_dot(a0_u8.sCDEF, (uchar4)(1));</div><div class="line"><a name="l01079"></a><span class="lineno"> 1079</span>&#160;</div><div class="line"><a name="l01080"></a><span class="lineno"> 1080</span>&#160; a0_u8 = vload16(1, matrix_a + i);</div><div class="line"><a name="l01081"></a><span class="lineno"> 1081</span>&#160;</div><div class="line"><a name="l01082"></a><span class="lineno"> 1082</span>&#160; sum_row += arm_dot(a0_u8.s0123, (uchar4)(1));</div><div class="line"><a name="l01083"></a><span class="lineno"> 1083</span>&#160; sum_row += arm_dot(a0_u8.s4567, (uchar4)(1));</div><div class="line"><a name="l01084"></a><span class="lineno"> 1084</span>&#160; sum_row += arm_dot(a0_u8.s89AB, (uchar4)(1));</div><div class="line"><a name="l01085"></a><span class="lineno"> 1085</span>&#160; sum_row += arm_dot(a0_u8.sCDEF, (uchar4)(1));</div><div class="line"><a name="l01086"></a><span class="lineno"> 1086</span>&#160; }</div><div class="line"><a name="l01087"></a><span class="lineno"> 1087</span>&#160;</div><div class="line"><a name="l01088"></a><span class="lineno"> 1088</span>&#160; <span class="comment">// This for loop performs the leftover accumulations</span></div><div class="line"><a name="l01089"></a><span class="lineno"> 1089</span>&#160; <span class="keywordflow">for</span>(; i &lt; COLS_A; ++i)</div><div class="line"><a name="l01090"></a><span class="lineno"> 1090</span>&#160; {</div><div class="line"><a name="l01091"></a><span class="lineno"> 1091</span>&#160; sum_row += matrix_a[i];</div><div class="line"><a name="l01092"></a><span class="lineno"> 1092</span>&#160; }</div><div class="line"><a name="l01093"></a><span class="lineno"> 1093</span>&#160;</div><div class="line"><a name="l01094"></a><span class="lineno"> 1094</span>&#160; *((__global <span class="keywordtype">int</span> *)<a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#adbf67dcee294e673cf796f1ed8aeb6a4">dst</a>.ptr) = (int)sum_row;</div><div class="line"><a name="l01095"></a><span class="lineno"> 1095</span>&#160;}</div><div class="line"><a name="l01096"></a><span class="lineno"> 1096</span>&#160;<span class="preprocessor">#endif // defined(ARM_COMPUTE_OPENCL_DOT8_ENABLED) &amp;&amp; defined(cl_arm_integer_dot_product_int8)</span></div><div class="line"><a name="l01097"></a><span class="lineno"> 1097</span>&#160;<span class="preprocessor">#endif // defined(COLS_A)</span></div><div class="line"><a name="l01098"></a><span class="lineno"> 1098</span>&#160;</div><div class="line"><a name="l01099"></a><span class="lineno"> 1099</span>&#160;<span class="preprocessor">#if defined(COLS_B) &amp;&amp; defined(ROWS_B)</span></div><div class="line"><a name="l01100"></a><span class="lineno"> 1100</span>&#160;</div><div class="line"><a name="l01122"></a><span class="lineno"> 1122</span>&#160;__kernel <span class="keywordtype">void</span> gemmlowp_matrix_b_reduction(<a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a6b83038822d1ae7ab619b684ed3b7fc0">TENSOR3D_DECLARATION</a>(<a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#a989ab3e96426615bb98e04e0235088ca">src</a>),</div><div class="line"><a name="l01123"></a><span class="lineno"> 1123</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a22f42fcf2077d951271df83b55c1a71a">IMAGE_DECLARATION</a>(<a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#adbf67dcee294e673cf796f1ed8aeb6a4">dst</a>))</div><div class="line"><a name="l01124"></a><span class="lineno"> 1124</span>&#160;{</div><div class="line"><a name="l01125"></a><span class="lineno"> 1125</span>&#160; <span class="comment">// Compute source and destination addresses</span></div><div class="line"><a name="l01126"></a><span class="lineno"> 1126</span>&#160; <a class="code" href="struct_tensor3_d.xhtml">Tensor3D</a> <a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#a989ab3e96426615bb98e04e0235088ca">src</a> = <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a31c8c760f08fb1a331b16b7c204321dc">CONVERT_TO_TENSOR3D_STRUCT</a>(<a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#a989ab3e96426615bb98e04e0235088ca">src</a>);</div><div class="line"><a name="l01127"></a><span class="lineno"> 1127</span>&#160; <a class="code" href="struct_image.xhtml">Image</a> <a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#adbf67dcee294e673cf796f1ed8aeb6a4">dst</a> = <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#aebe814363556c244be043b13e7969197">CONVERT_TO_IMAGE_STRUCT</a>(<a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#adbf67dcee294e673cf796f1ed8aeb6a4">dst</a>);</div><div class="line"><a name="l01128"></a><span class="lineno"> 1128</span>&#160;</div><div class="line"><a name="l01129"></a><span class="lineno"> 1129</span>&#160; uint16 sum_col_u32 = (uint16)0;</div><div class="line"><a name="l01130"></a><span class="lineno"> 1130</span>&#160;</div><div class="line"><a name="l01131"></a><span class="lineno"> 1131</span>&#160; __global <span class="keyword">const</span> uchar *matrix_b = (__global <span class="keyword">const</span> uchar *)(<a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#a989ab3e96426615bb98e04e0235088ca">src</a>.ptr + get_global_id(1) * src_stride_z);</div><div class="line"><a name="l01132"></a><span class="lineno"> 1132</span>&#160;</div><div class="line"><a name="l01133"></a><span class="lineno"> 1133</span>&#160; <span class="keywordtype">int</span> i = 0;</div><div class="line"><a name="l01134"></a><span class="lineno"> 1134</span>&#160; <span class="comment">// This for loop performs 4 accumulations</span></div><div class="line"><a name="l01135"></a><span class="lineno"> 1135</span>&#160; <span class="keywordflow">for</span>(; i &lt;= ((int)ROWS_B - 4); i += 4)</div><div class="line"><a name="l01136"></a><span class="lineno"> 1136</span>&#160; {</div><div class="line"><a name="l01137"></a><span class="lineno"> 1137</span>&#160; <span class="keyword">const</span> uchar16 b0_u8 = vload16(0, matrix_b + 0 * src_stride_y);</div><div class="line"><a name="l01138"></a><span class="lineno"> 1138</span>&#160; <span class="keyword">const</span> uchar16 b1_u8 = vload16(0, matrix_b + 1 * src_stride_y);</div><div class="line"><a name="l01139"></a><span class="lineno"> 1139</span>&#160; <span class="keyword">const</span> uchar16 b2_u8 = vload16(0, matrix_b + 2 * src_stride_y);</div><div class="line"><a name="l01140"></a><span class="lineno"> 1140</span>&#160; <span class="keyword">const</span> uchar16 b3_u8 = vload16(0, matrix_b + 3 * src_stride_y);</div><div class="line"><a name="l01141"></a><span class="lineno"> 1141</span>&#160;</div><div class="line"><a name="l01142"></a><span class="lineno"> 1142</span>&#160; sum_col_u32 += convert_uint16(b0_u8) + convert_uint16(b1_u8) + convert_uint16(b2_u8) + convert_uint16(b3_u8);</div><div class="line"><a name="l01143"></a><span class="lineno"> 1143</span>&#160;</div><div class="line"><a name="l01144"></a><span class="lineno"> 1144</span>&#160; matrix_b += 4 * src_stride_y;</div><div class="line"><a name="l01145"></a><span class="lineno"> 1145</span>&#160; }</div><div class="line"><a name="l01146"></a><span class="lineno"> 1146</span>&#160;</div><div class="line"><a name="l01147"></a><span class="lineno"> 1147</span>&#160; <span class="comment">// This for loop perfoms the leftover accumulations</span></div><div class="line"><a name="l01148"></a><span class="lineno"> 1148</span>&#160; <span class="keywordflow">for</span>(; i &lt; (int)ROWS_B; ++i)</div><div class="line"><a name="l01149"></a><span class="lineno"> 1149</span>&#160; {</div><div class="line"><a name="l01150"></a><span class="lineno"> 1150</span>&#160; <span class="keyword">const</span> uchar16 b0_u8 = vload16(0, matrix_b);</div><div class="line"><a name="l01151"></a><span class="lineno"> 1151</span>&#160;</div><div class="line"><a name="l01152"></a><span class="lineno"> 1152</span>&#160; sum_col_u32 += convert_uint16(b0_u8);</div><div class="line"><a name="l01153"></a><span class="lineno"> 1153</span>&#160;</div><div class="line"><a name="l01154"></a><span class="lineno"> 1154</span>&#160; matrix_b += src_stride_y;</div><div class="line"><a name="l01155"></a><span class="lineno"> 1155</span>&#160; }</div><div class="line"><a name="l01156"></a><span class="lineno"> 1156</span>&#160;</div><div class="line"><a name="l01157"></a><span class="lineno"> 1157</span>&#160; vstore16(convert_int16(sum_col_u32), 0, (__global <span class="keywordtype">int</span> *)<a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#adbf67dcee294e673cf796f1ed8aeb6a4">dst</a>.ptr);</div><div class="line"><a name="l01158"></a><span class="lineno"> 1158</span>&#160;}</div><div class="line"><a name="l01159"></a><span class="lineno"> 1159</span>&#160;<span class="preprocessor">#endif // defined(COLS_B) &amp;&amp; defined(ROWS_B)</span></div><div class="line"><a name="l01160"></a><span class="lineno"> 1160</span>&#160;</div><div class="line"><a name="l01161"></a><span class="lineno"> 1161</span>&#160;<span class="preprocessor">#if defined(K_OFFSET)</span></div><div class="line"><a name="l01162"></a><span class="lineno"> 1162</span>&#160;</div><div class="line"><a name="l01163"></a><span class="lineno"> 1163</span>&#160;<span class="comment">/* Helper function used to calculate the offset contribution after @ref CLGEMMLowpMatrixMultiplyKernel.</span></div><div class="line"><a name="l01164"></a><span class="lineno"> 1164</span>&#160;<span class="comment"> *</span></div><div class="line"><a name="l01165"></a><span class="lineno"> 1165</span>&#160;<span class="comment"> * This kernel takes a final int32 accumulator value (the output of @CLGEMMLowpMatrixMultiplyKernel),</span></div><div class="line"><a name="l01166"></a><span class="lineno"> 1166</span>&#160;<span class="comment"> * and calculates the offset contribution of matrix A and matrix B.</span></div><div class="line"><a name="l01167"></a><span class="lineno"> 1167</span>&#160;<span class="comment"> *</span></div><div class="line"><a name="l01168"></a><span class="lineno"> 1168</span>&#160;<span class="comment"> * @attention The k_offset = a_offset * b_offset * k (where k is the number of matrix A columns) needs to be passed at compile time using -DK_OFFSET (i.e. -DK_OFFSET=1200)</span></div><div class="line"><a name="l01169"></a><span class="lineno"> 1169</span>&#160;<span class="comment"> * @note In case the offset contribution due to a_offset is required, a_offset needs to be passed at compile time using -DA_OFFSET (i.e. -DA_OFFSET=1)</span></div><div class="line"><a name="l01170"></a><span class="lineno"> 1170</span>&#160;<span class="comment"> * @note In case the offset contribution due to b_offset is required, b_offset needs to be passed at compile time using -DB_OFFSET (i.e. -DB_OFFSET=6)</span></div><div class="line"><a name="l01171"></a><span class="lineno"> 1171</span>&#160;<span class="comment"> * @note In case sum_col has batches, -DSUM_COL_HAS_BATCHES must be passed at compile time. Usually if gemmlowp is used to accelerate convolution layer, sum_col will not have batches</span></div><div class="line"><a name="l01172"></a><span class="lineno"> 1172</span>&#160;<span class="comment"> *</span></div><div class="line"><a name="l01173"></a><span class="lineno"> 1173</span>&#160;<span class="comment"> * @param[in] x get_global_id(0) * 4</span></div><div class="line"><a name="l01174"></a><span class="lineno"> 1174</span>&#160;<span class="comment"> * @param[in] y get_global_id(1)</span></div><div class="line"><a name="l01175"></a><span class="lineno"> 1175</span>&#160;<span class="comment"> * @param[in] z get_global_id(2)</span></div><div class="line"><a name="l01176"></a><span class="lineno"> 1176</span>&#160;<span class="comment"> * @param[in] sum_col_ptr (Optional) Pointer to the source tensor. Supported data type: same as @p mm_result_ptr</span></div><div class="line"><a name="l01177"></a><span class="lineno"> 1177</span>&#160;<span class="comment"> * @param[in] sum_col_stride_x (Optional) Stride of the source tensor in X dimension (in bytes)</span></div><div class="line"><a name="l01178"></a><span class="lineno"> 1178</span>&#160;<span class="comment"> * @param[in] sum_col_step_x (Optional) sum_col_stride_x * number of elements along X processed per workitem(in bytes)</span></div><div class="line"><a name="l01179"></a><span class="lineno"> 1179</span>&#160;<span class="comment"> * @param[in] sum_col_stride_y (Optional) Stride of the source tensor in Y dimension (in bytes)</span></div><div class="line"><a name="l01180"></a><span class="lineno"> 1180</span>&#160;<span class="comment"> * @param[in] sum_col_step_y (Optional) sum_col_stride_y * number of elements along Y processed per workitem(in bytes)</span></div><div class="line"><a name="l01181"></a><span class="lineno"> 1181</span>&#160;<span class="comment"> * @param[in] sum_col_offset_first_element_in_bytes (Optional) The offset of the first element in the source tensor</span></div><div class="line"><a name="l01182"></a><span class="lineno"> 1182</span>&#160;<span class="comment"> * @param[in] sum_row_ptr (Optional) Pointer to the source tensor. Supported data type: same as @p mm_result_ptr</span></div><div class="line"><a name="l01183"></a><span class="lineno"> 1183</span>&#160;<span class="comment"> * @param[in] sum_row_stride_x (Optional) Stride of the source tensor in X dimension (in bytes)</span></div><div class="line"><a name="l01184"></a><span class="lineno"> 1184</span>&#160;<span class="comment"> * @param[in] sum_row_step_x (Optional) sum_row_stride_x * number of elements along X processed per workitem(in bytes)</span></div><div class="line"><a name="l01185"></a><span class="lineno"> 1185</span>&#160;<span class="comment"> * @param[in] sum_row_stride_y (Optional) Stride of the source tensor in Y dimension (in bytes)</span></div><div class="line"><a name="l01186"></a><span class="lineno"> 1186</span>&#160;<span class="comment"> * @param[in] sum_row_step_y (Optional) sum_row_stride_y * number of elements along Y processed per workitem(in bytes)</span></div><div class="line"><a name="l01187"></a><span class="lineno"> 1187</span>&#160;<span class="comment"> * @param[in] sum_row_offset_first_element_in_bytes (Optional) The offset of the first element in the source tensor</span></div><div class="line"><a name="l01188"></a><span class="lineno"> 1188</span>&#160;<span class="comment"> * @param[in] biases_ptr (Optional) Pointer to the biases tensor. Supported data type: same as @p src_ptr</span></div><div class="line"><a name="l01189"></a><span class="lineno"> 1189</span>&#160;<span class="comment"> * @param[in] biases_stride_x (Optional) Stride of the biases tensor in X dimension (in bytes)</span></div><div class="line"><a name="l01190"></a><span class="lineno"> 1190</span>&#160;<span class="comment"> * @param[in] biases_step_x (Optional) biases_stride_x * number of elements along X processed per workitem(in bytes)</span></div><div class="line"><a name="l01191"></a><span class="lineno"> 1191</span>&#160;<span class="comment"> * @param[in] biases_offset_first_element_in_bytes (Optional) The offset of the first element in the biases tensor</span></div><div class="line"><a name="l01192"></a><span class="lineno"> 1192</span>&#160;<span class="comment"> */</span></div><div class="line"><a name="l01193"></a><span class="lineno"> 1193</span>&#160;<span class="keyword">inline</span> int4 offset_contribution(</div><div class="line"><a name="l01194"></a><span class="lineno"> 1194</span>&#160; <span class="keywordtype">int</span> x,</div><div class="line"><a name="l01195"></a><span class="lineno"> 1195</span>&#160; <span class="keywordtype">int</span> y,</div><div class="line"><a name="l01196"></a><span class="lineno"> 1196</span>&#160; <span class="keywordtype">int</span> z</div><div class="line"><a name="l01197"></a><span class="lineno"> 1197</span>&#160;#<span class="keywordflow">if</span> defined(A_OFFSET)</div><div class="line"><a name="l01198"></a><span class="lineno"> 1198</span>&#160; ,</div><div class="line"><a name="l01199"></a><span class="lineno"> 1199</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a22f42fcf2077d951271df83b55c1a71a">IMAGE_DECLARATION</a>(sum_col)</div><div class="line"><a name="l01200"></a><span class="lineno"> 1200</span>&#160;#endif <span class="comment">// defined(A_OFFSET)</span></div><div class="line"><a name="l01201"></a><span class="lineno"> 1201</span>&#160;#<span class="keywordflow">if</span> defined(B_OFFSET)</div><div class="line"><a name="l01202"></a><span class="lineno"> 1202</span>&#160; ,</div><div class="line"><a name="l01203"></a><span class="lineno"> 1203</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a22f42fcf2077d951271df83b55c1a71a">IMAGE_DECLARATION</a>(sum_row)</div><div class="line"><a name="l01204"></a><span class="lineno"> 1204</span>&#160;#endif <span class="comment">// defined(B_OFFSET)</span></div><div class="line"><a name="l01205"></a><span class="lineno"> 1205</span>&#160;#<span class="keywordflow">if</span> defined(ADD_BIAS)</div><div class="line"><a name="l01206"></a><span class="lineno"> 1206</span>&#160; ,</div><div class="line"><a name="l01207"></a><span class="lineno"> 1207</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a40a6eb9f2a7712f08d6bb8ff6c9e6ca7">VECTOR_DECLARATION</a>(biases)</div><div class="line"><a name="l01208"></a><span class="lineno"> 1208</span>&#160;#endif <span class="comment">// defined(ADD_BIAS)</span></div><div class="line"><a name="l01209"></a><span class="lineno"> 1209</span>&#160;)</div><div class="line"><a name="l01210"></a><span class="lineno"> 1210</span>&#160;{</div><div class="line"><a name="l01211"></a><span class="lineno"> 1211</span>&#160; int4 a_offset_s32 = (int4)0;</div><div class="line"><a name="l01212"></a><span class="lineno"> 1212</span>&#160; int4 b_offset_s32 = (int4)0;</div><div class="line"><a name="l01213"></a><span class="lineno"> 1213</span>&#160;</div><div class="line"><a name="l01214"></a><span class="lineno"> 1214</span>&#160; <span class="keywordtype">int</span> batch_id = z;</div><div class="line"><a name="l01215"></a><span class="lineno"> 1215</span>&#160;<span class="preprocessor">#if defined(DEPTH_INPUT3D)</span></div><div class="line"><a name="l01216"></a><span class="lineno"> 1216</span>&#160; batch_id /= (int)DEPTH_INPUT3D;</div><div class="line"><a name="l01217"></a><span class="lineno"> 1217</span>&#160;<span class="preprocessor">#endif // defined(DEPTH_INPUT3D)</span></div><div class="line"><a name="l01218"></a><span class="lineno"> 1218</span>&#160;</div><div class="line"><a name="l01219"></a><span class="lineno"> 1219</span>&#160;<span class="preprocessor">#if defined(A_OFFSET)</span></div><div class="line"><a name="l01220"></a><span class="lineno"> 1220</span>&#160; <span class="comment">// Compute the offset contribution due to A_OFFSET</span></div><div class="line"><a name="l01221"></a><span class="lineno"> 1221</span>&#160; __global uchar *sum_col_addr = sum_col_ptr + sum_col_offset_first_element_in_bytes + x * <span class="keyword">sizeof</span>(int);</div><div class="line"><a name="l01222"></a><span class="lineno"> 1222</span>&#160;</div><div class="line"><a name="l01223"></a><span class="lineno"> 1223</span>&#160; <span class="comment">// Compute the offset contribution due to A_OFFSET</span></div><div class="line"><a name="l01224"></a><span class="lineno"> 1224</span>&#160;<span class="preprocessor">#if defined(SUM_COL_HAS_BATCHES)</span></div><div class="line"><a name="l01225"></a><span class="lineno"> 1225</span>&#160; a_offset_s32 = vload4(0, (__global <span class="keywordtype">int</span> *)(sum_col_addr + batch_id * sum_col_stride_y));</div><div class="line"><a name="l01226"></a><span class="lineno"> 1226</span>&#160;<span class="preprocessor">#else // defined(SUM_COL_HAS_BATCHES)</span></div><div class="line"><a name="l01227"></a><span class="lineno"> 1227</span>&#160; a_offset_s32 = vload4(0, (__global <span class="keywordtype">int</span> *)sum_col_addr);</div><div class="line"><a name="l01228"></a><span class="lineno"> 1228</span>&#160;<span class="preprocessor">#endif // defined(SUM_COL_HAS_BATCHES)</span></div><div class="line"><a name="l01229"></a><span class="lineno"> 1229</span>&#160;</div><div class="line"><a name="l01230"></a><span class="lineno"> 1230</span>&#160; a_offset_s32 *= (int4)A_OFFSET;</div><div class="line"><a name="l01231"></a><span class="lineno"> 1231</span>&#160;<span class="preprocessor">#endif // defined(A_OFFSET)</span></div><div class="line"><a name="l01232"></a><span class="lineno"> 1232</span>&#160;</div><div class="line"><a name="l01233"></a><span class="lineno"> 1233</span>&#160;<span class="preprocessor">#if defined(B_OFFSET)</span></div><div class="line"><a name="l01234"></a><span class="lineno"> 1234</span>&#160; <span class="comment">// Compute the offset contribution due to A_OFFSET</span></div><div class="line"><a name="l01235"></a><span class="lineno"> 1235</span>&#160; __global uchar *sum_row_addr = sum_row_ptr + sum_row_offset_first_element_in_bytes + y * <span class="keyword">sizeof</span>(int);</div><div class="line"><a name="l01236"></a><span class="lineno"> 1236</span>&#160;</div><div class="line"><a name="l01237"></a><span class="lineno"> 1237</span>&#160; <span class="comment">// Compute the offset contribution due to B_OFFSET</span></div><div class="line"><a name="l01238"></a><span class="lineno"> 1238</span>&#160;<span class="preprocessor">#if defined(HEIGHT_INPUT3D) &amp;&amp; defined(DEPTH_INPUT3D)</span></div><div class="line"><a name="l01239"></a><span class="lineno"> 1239</span>&#160; b_offset_s32 = (int4) * (((__global <span class="keywordtype">int</span> *)(sum_row_addr + batch_id * sum_row_stride_y)) + (z % (int)DEPTH_INPUT3D) * (int)HEIGHT_INPUT3D);</div><div class="line"><a name="l01240"></a><span class="lineno"> 1240</span>&#160;<span class="preprocessor">#else // defined(HEIGHT_INPUT3D) &amp;&amp; defined(DEPTH_INPUT3D)</span></div><div class="line"><a name="l01241"></a><span class="lineno"> 1241</span>&#160; b_offset_s32 = (int4) * (((__global <span class="keywordtype">int</span> *)(sum_row_addr + batch_id * sum_row_stride_y)));</div><div class="line"><a name="l01242"></a><span class="lineno"> 1242</span>&#160;<span class="preprocessor">#endif // defined(HEIGHT_INPUT3D) &amp;&amp; defined(DEPTH_INPUT3D)</span></div><div class="line"><a name="l01243"></a><span class="lineno"> 1243</span>&#160; b_offset_s32 *= (int4)B_OFFSET;</div><div class="line"><a name="l01244"></a><span class="lineno"> 1244</span>&#160;<span class="preprocessor">#endif // defined(B_OFFSET)</span></div><div class="line"><a name="l01245"></a><span class="lineno"> 1245</span>&#160;</div><div class="line"><a name="l01246"></a><span class="lineno"> 1246</span>&#160;<span class="preprocessor">#if defined(ADD_BIAS)</span></div><div class="line"><a name="l01247"></a><span class="lineno"> 1247</span>&#160; <span class="comment">// Add bias</span></div><div class="line"><a name="l01248"></a><span class="lineno"> 1248</span>&#160; __global uchar *bias_addr = biases_ptr + biases_offset_first_element_in_bytes + x * <span class="keyword">sizeof</span>(int);</div><div class="line"><a name="l01249"></a><span class="lineno"> 1249</span>&#160;</div><div class="line"><a name="l01250"></a><span class="lineno"> 1250</span>&#160; int4 biases_values = vload4(0, (__global <span class="keywordtype">int</span> *)bias_addr);</div><div class="line"><a name="l01251"></a><span class="lineno"> 1251</span>&#160; b_offset_s32 += (int4)biases_values;</div><div class="line"><a name="l01252"></a><span class="lineno"> 1252</span>&#160;<span class="preprocessor">#endif // defined(ADD_BIAS)</span></div><div class="line"><a name="l01253"></a><span class="lineno"> 1253</span>&#160;</div><div class="line"><a name="l01254"></a><span class="lineno"> 1254</span>&#160; <span class="keywordflow">return</span> (int4)K_OFFSET + a_offset_s32 + b_offset_s32;</div><div class="line"><a name="l01255"></a><span class="lineno"> 1255</span>&#160;}</div><div class="line"><a name="l01256"></a><span class="lineno"> 1256</span>&#160;</div><div class="line"><a name="l01257"></a><span class="lineno"> 1257</span>&#160;<span class="comment">/* OpenCL kernel used to add the offset contribution after @ref CLGEMMLowpMatrixMultiplyKernel. The computation is performed in-place</span></div><div class="line"><a name="l01258"></a><span class="lineno"> 1258</span>&#160;<span class="comment"> *</span></div><div class="line"><a name="l01259"></a><span class="lineno"> 1259</span>&#160;<span class="comment"> * This kernel takes a final int32 accumulator value (the output of @CLGEMMLowpMatrixMultiplyKernel),</span></div><div class="line"><a name="l01260"></a><span class="lineno"> 1260</span>&#160;<span class="comment"> * and adds to it the offset contribution of matrix A and matrix B in-place.</span></div><div class="line"><a name="l01261"></a><span class="lineno"> 1261</span>&#160;<span class="comment"> *</span></div><div class="line"><a name="l01262"></a><span class="lineno"> 1262</span>&#160;<span class="comment"> * @attention The k_offset = a_offset * b_offset * k (where k is the number of matrix A columns) needs to be passed at compile time using -DK_OFFSET (i.e. -DK_OFFSET=1200)</span></div><div class="line"><a name="l01263"></a><span class="lineno"> 1263</span>&#160;<span class="comment"> * @note In case the offset contribution due to a_offset is required, a_offset needs to be passed at compile time using -DA_OFFSET (i.e. -DA_OFFSET=1)</span></div><div class="line"><a name="l01264"></a><span class="lineno"> 1264</span>&#160;<span class="comment"> * @note In case the offset contribution due to b_offset is required, b_offset needs to be passed at compile time using -DB_OFFSET (i.e. -DB_OFFSET=6)</span></div><div class="line"><a name="l01265"></a><span class="lineno"> 1265</span>&#160;<span class="comment"> * @note In case sum_col has batches, -DSUM_COL_HAS_BATCHES must be passed at compile time. Usually if gemmlowp is used to accelerate convolution layer, sum_col will not have batches</span></div><div class="line"><a name="l01266"></a><span class="lineno"> 1266</span>&#160;<span class="comment"> *</span></div><div class="line"><a name="l01267"></a><span class="lineno"> 1267</span>&#160;<span class="comment"> * The final result is:</span></div><div class="line"><a name="l01268"></a><span class="lineno"> 1268</span>&#160;<span class="comment"> *</span></div><div class="line"><a name="l01269"></a><span class="lineno"> 1269</span>&#160;<span class="comment"> * mm_result[i][k] = mm_result[i][k] +</span></div><div class="line"><a name="l01270"></a><span class="lineno"> 1270</span>&#160;<span class="comment"> * (sum_col[k] * A_OFFSET) +</span></div><div class="line"><a name="l01271"></a><span class="lineno"> 1271</span>&#160;<span class="comment"> * (sum_row[i] * B_OFFSET) +</span></div><div class="line"><a name="l01272"></a><span class="lineno"> 1272</span>&#160;<span class="comment"> * (K_OFFSET)</span></div><div class="line"><a name="l01273"></a><span class="lineno"> 1273</span>&#160;<span class="comment"> *</span></div><div class="line"><a name="l01274"></a><span class="lineno"> 1274</span>&#160;<span class="comment"> * @param[in] mm_result_ptr Pointer to the source tensor. Supported data type: S32</span></div><div class="line"><a name="l01275"></a><span class="lineno"> 1275</span>&#160;<span class="comment"> * @param[in] mm_result_stride_x Stride of the source tensor in X dimension (in bytes)</span></div><div class="line"><a name="l01276"></a><span class="lineno"> 1276</span>&#160;<span class="comment"> * @param[in] mm_result_step_x mm_result_stride_x * number of elements along X processed per workitem(in bytes)</span></div><div class="line"><a name="l01277"></a><span class="lineno"> 1277</span>&#160;<span class="comment"> * @param[in] mm_result_stride_y Stride of the source tensor in Y dimension (in bytes)</span></div><div class="line"><a name="l01278"></a><span class="lineno"> 1278</span>&#160;<span class="comment"> * @param[in] mm_result_step_y mm_result_stride_y * number of elements along Y processed per workitem(in bytes)</span></div><div class="line"><a name="l01279"></a><span class="lineno"> 1279</span>&#160;<span class="comment"> * @param[in] mm_result_stride_z Stride of the source tensor in Z dimension (in bytes)</span></div><div class="line"><a name="l01280"></a><span class="lineno"> 1280</span>&#160;<span class="comment"> * @param[in] mm_result_step_z mm_result_stride_z * number of elements along Z processed per workitem(in bytes)</span></div><div class="line"><a name="l01281"></a><span class="lineno"> 1281</span>&#160;<span class="comment"> * @param[in] mm_result_offset_first_element_in_bytes The offset of the first element in the source tensor</span></div><div class="line"><a name="l01282"></a><span class="lineno"> 1282</span>&#160;<span class="comment"> * @param[in] sum_col_ptr (Optional) Pointer to the source tensor. Supported data type: same as @p mm_result_ptr</span></div><div class="line"><a name="l01283"></a><span class="lineno"> 1283</span>&#160;<span class="comment"> * @param[in] sum_col_stride_x (Optional) Stride of the source tensor in X dimension (in bytes)</span></div><div class="line"><a name="l01284"></a><span class="lineno"> 1284</span>&#160;<span class="comment"> * @param[in] sum_col_step_x (Optional) sum_col_stride_x * number of elements along X processed per workitem(in bytes)</span></div><div class="line"><a name="l01285"></a><span class="lineno"> 1285</span>&#160;<span class="comment"> * @param[in] sum_col_stride_y (Optional) Stride of the source tensor in Y dimension (in bytes)</span></div><div class="line"><a name="l01286"></a><span class="lineno"> 1286</span>&#160;<span class="comment"> * @param[in] sum_col_step_y (Optional) sum_col_stride_y * number of elements along Y processed per workitem(in bytes)</span></div><div class="line"><a name="l01287"></a><span class="lineno"> 1287</span>&#160;<span class="comment"> * @param[in] sum_col_offset_first_element_in_bytes (Optional) The offset of the first element in the source tensor</span></div><div class="line"><a name="l01288"></a><span class="lineno"> 1288</span>&#160;<span class="comment"> * @param[in] sum_row_ptr (Optional) Pointer to the source tensor. Supported data type: same as @p mm_result_ptr</span></div><div class="line"><a name="l01289"></a><span class="lineno"> 1289</span>&#160;<span class="comment"> * @param[in] sum_row_stride_x (Optional) Stride of the source tensor in X dimension (in bytes)</span></div><div class="line"><a name="l01290"></a><span class="lineno"> 1290</span>&#160;<span class="comment"> * @param[in] sum_row_step_x (Optional) sum_row_stride_x * number of elements along X processed per workitem(in bytes)</span></div><div class="line"><a name="l01291"></a><span class="lineno"> 1291</span>&#160;<span class="comment"> * @param[in] sum_row_stride_y (Optional) Stride of the source tensor in Y dimension (in bytes)</span></div><div class="line"><a name="l01292"></a><span class="lineno"> 1292</span>&#160;<span class="comment"> * @param[in] sum_row_step_y (Optional) sum_row_stride_y * number of elements along Y processed per workitem(in bytes)</span></div><div class="line"><a name="l01293"></a><span class="lineno"> 1293</span>&#160;<span class="comment"> * @param[in] sum_row_offset_first_element_in_bytes (Optional) The offset of the first element in the source tensor</span></div><div class="line"><a name="l01294"></a><span class="lineno"> 1294</span>&#160;<span class="comment"> * @param[in] biases_ptr (Optional) Pointer to the biases tensor. Supported data type: same as @p src_ptr</span></div><div class="line"><a name="l01295"></a><span class="lineno"> 1295</span>&#160;<span class="comment"> * @param[in] biases_stride_x (Optional) Stride of the biases tensor in X dimension (in bytes)</span></div><div class="line"><a name="l01296"></a><span class="lineno"> 1296</span>&#160;<span class="comment"> * @param[in] biases_step_x (Optional) biases_stride_x * number of elements along X processed per workitem(in bytes)</span></div><div class="line"><a name="l01297"></a><span class="lineno"> 1297</span>&#160;<span class="comment"> * @param[in] biases_offset_first_element_in_bytes (Optional) The offset of the first element in the biases tensor</span></div><div class="line"><a name="l01298"></a><span class="lineno"> 1298</span>&#160;<span class="comment"> */</span></div><div class="line"><a name="l01299"></a><span class="lineno"> 1299</span>&#160;__kernel <span class="keywordtype">void</span> gemmlowp_offset_contribution(<a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a6b83038822d1ae7ab619b684ed3b7fc0">TENSOR3D_DECLARATION</a>(mm_result)</div><div class="line"><a name="l01300"></a><span class="lineno"> 1300</span>&#160;#<span class="keywordflow">if</span> defined(A_OFFSET)</div><div class="line"><a name="l01301"></a><span class="lineno"> 1301</span>&#160; ,</div><div class="line"><a name="l01302"></a><span class="lineno"> 1302</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a22f42fcf2077d951271df83b55c1a71a">IMAGE_DECLARATION</a>(sum_col)</div><div class="line"><a name="l01303"></a><span class="lineno"> 1303</span>&#160;#endif <span class="comment">// defined(A_OFFSET)</span></div><div class="line"><a name="l01304"></a><span class="lineno"> 1304</span>&#160;#<span class="keywordflow">if</span> defined(B_OFFSET)</div><div class="line"><a name="l01305"></a><span class="lineno"> 1305</span>&#160; ,</div><div class="line"><a name="l01306"></a><span class="lineno"> 1306</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a22f42fcf2077d951271df83b55c1a71a">IMAGE_DECLARATION</a>(sum_row)</div><div class="line"><a name="l01307"></a><span class="lineno"> 1307</span>&#160;#endif <span class="comment">// defined(B_OFFSET)</span></div><div class="line"><a name="l01308"></a><span class="lineno"> 1308</span>&#160;#<span class="keywordflow">if</span> defined(ADD_BIAS)</div><div class="line"><a name="l01309"></a><span class="lineno"> 1309</span>&#160; ,</div><div class="line"><a name="l01310"></a><span class="lineno"> 1310</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a40a6eb9f2a7712f08d6bb8ff6c9e6ca7">VECTOR_DECLARATION</a>(biases)</div><div class="line"><a name="l01311"></a><span class="lineno"> 1311</span>&#160;#endif <span class="comment">// defined(ADD_BIAS))</span></div><div class="line"><a name="l01312"></a><span class="lineno"> 1312</span>&#160; )</div><div class="line"><a name="l01313"></a><span class="lineno"> 1313</span>&#160;{</div><div class="line"><a name="l01314"></a><span class="lineno"> 1314</span>&#160; <span class="keyword">const</span> <span class="keywordtype">int</span> x = get_global_id(0) * 4;</div><div class="line"><a name="l01315"></a><span class="lineno"> 1315</span>&#160; <span class="keyword">const</span> <span class="keywordtype">int</span> y = get_global_id(1);</div><div class="line"><a name="l01316"></a><span class="lineno"> 1316</span>&#160; <span class="keyword">const</span> <span class="keywordtype">int</span> z = get_global_id(2);</div><div class="line"><a name="l01317"></a><span class="lineno"> 1317</span>&#160;</div><div class="line"><a name="l01318"></a><span class="lineno"> 1318</span>&#160; <span class="comment">// Compute offset contribution</span></div><div class="line"><a name="l01319"></a><span class="lineno"> 1319</span>&#160; int4 offset_term_s32 = offset_contribution(</div><div class="line"><a name="l01320"></a><span class="lineno"> 1320</span>&#160; x, y, z</div><div class="line"><a name="l01321"></a><span class="lineno"> 1321</span>&#160;#<span class="keywordflow">if</span> defined(A_OFFSET)</div><div class="line"><a name="l01322"></a><span class="lineno"> 1322</span>&#160; ,</div><div class="line"><a name="l01323"></a><span class="lineno"> 1323</span>&#160; sum_col_ptr,</div><div class="line"><a name="l01324"></a><span class="lineno"> 1324</span>&#160; sum_col_stride_x,</div><div class="line"><a name="l01325"></a><span class="lineno"> 1325</span>&#160; sum_col_step_x,</div><div class="line"><a name="l01326"></a><span class="lineno"> 1326</span>&#160; sum_col_stride_y,</div><div class="line"><a name="l01327"></a><span class="lineno"> 1327</span>&#160; sum_col_step_y,</div><div class="line"><a name="l01328"></a><span class="lineno"> 1328</span>&#160; sum_col_offset_first_element_in_bytes</div><div class="line"><a name="l01329"></a><span class="lineno"> 1329</span>&#160;#endif <span class="comment">// defined(A_OFFSET)</span></div><div class="line"><a name="l01330"></a><span class="lineno"> 1330</span>&#160;#<span class="keywordflow">if</span> defined(B_OFFSET)</div><div class="line"><a name="l01331"></a><span class="lineno"> 1331</span>&#160; ,</div><div class="line"><a name="l01332"></a><span class="lineno"> 1332</span>&#160; sum_row_ptr,</div><div class="line"><a name="l01333"></a><span class="lineno"> 1333</span>&#160; sum_row_stride_x,</div><div class="line"><a name="l01334"></a><span class="lineno"> 1334</span>&#160; sum_row_step_x,</div><div class="line"><a name="l01335"></a><span class="lineno"> 1335</span>&#160; sum_row_stride_y,</div><div class="line"><a name="l01336"></a><span class="lineno"> 1336</span>&#160; sum_row_step_y,</div><div class="line"><a name="l01337"></a><span class="lineno"> 1337</span>&#160; sum_row_offset_first_element_in_bytes</div><div class="line"><a name="l01338"></a><span class="lineno"> 1338</span>&#160;#endif <span class="comment">// defined(B_OFFSET)</span></div><div class="line"><a name="l01339"></a><span class="lineno"> 1339</span>&#160;#<span class="keywordflow">if</span> defined(ADD_BIAS)</div><div class="line"><a name="l01340"></a><span class="lineno"> 1340</span>&#160; ,</div><div class="line"><a name="l01341"></a><span class="lineno"> 1341</span>&#160; biases_ptr,</div><div class="line"><a name="l01342"></a><span class="lineno"> 1342</span>&#160; biases_stride_x,</div><div class="line"><a name="l01343"></a><span class="lineno"> 1343</span>&#160; biases_step_x,</div><div class="line"><a name="l01344"></a><span class="lineno"> 1344</span>&#160; biases_offset_first_element_in_bytes</div><div class="line"><a name="l01345"></a><span class="lineno"> 1345</span>&#160;#endif <span class="comment">// defined(ADD_BIAS)</span></div><div class="line"><a name="l01346"></a><span class="lineno"> 1346</span>&#160; );</div><div class="line"><a name="l01347"></a><span class="lineno"> 1347</span>&#160;</div><div class="line"><a name="l01348"></a><span class="lineno"> 1348</span>&#160; __global uchar *mm_result_addr = mm_result_ptr + mm_result_offset_first_element_in_bytes + x * <span class="keyword">sizeof</span>(int) + y * mm_result_stride_y + z * mm_result_stride_z;</div><div class="line"><a name="l01349"></a><span class="lineno"> 1349</span>&#160;</div><div class="line"><a name="l01350"></a><span class="lineno"> 1350</span>&#160; int4 in_s32 = vload4(0, (__global <span class="keywordtype">int</span> *)mm_result_addr);</div><div class="line"><a name="l01351"></a><span class="lineno"> 1351</span>&#160;</div><div class="line"><a name="l01352"></a><span class="lineno"> 1352</span>&#160; <span class="comment">// Add the offset terms to GEMM&#39;s result</span></div><div class="line"><a name="l01353"></a><span class="lineno"> 1353</span>&#160; in_s32 += offset_term_s32;</div><div class="line"><a name="l01354"></a><span class="lineno"> 1354</span>&#160;</div><div class="line"><a name="l01355"></a><span class="lineno"> 1355</span>&#160; <span class="comment">// Store the result with the offset contribution</span></div><div class="line"><a name="l01356"></a><span class="lineno"> 1356</span>&#160; vstore4(in_s32, 0, (__global <span class="keywordtype">int</span> *)mm_result_addr);</div><div class="line"><a name="l01357"></a><span class="lineno"> 1357</span>&#160;}</div><div class="line"><a name="l01358"></a><span class="lineno"> 1358</span>&#160;</div><div class="line"><a name="l01359"></a><span class="lineno"> 1359</span>&#160;<span class="preprocessor">#if defined(RESULT_OFFSET) &amp;&amp; defined(RESULT_MULTIPLIER) &amp;&amp; defined(RESULT_SHIFT)</span></div><div class="line"><a name="l01360"></a><span class="lineno"> 1360</span>&#160;<span class="comment">/* OpenCL kernel used to add the offset contribution after @ref CLGEMMLowpMatrixMultiplyKernel and it quantizes down to uint8.</span></div><div class="line"><a name="l01361"></a><span class="lineno"> 1361</span>&#160;<span class="comment"> *</span></div><div class="line"><a name="l01362"></a><span class="lineno"> 1362</span>&#160;<span class="comment"> * This kernel takes a final int32 accumulator value (the output of @CLGEMMLowpMatrixMultiplyKernel), adds to it the offset contribution of matrix A and matrix B and quantizes to uint8 through the output stage.</span></div><div class="line"><a name="l01363"></a><span class="lineno"> 1363</span>&#160;<span class="comment"> *</span></div><div class="line"><a name="l01364"></a><span class="lineno"> 1364</span>&#160;<span class="comment"> *</span></div><div class="line"><a name="l01365"></a><span class="lineno"> 1365</span>&#160;<span class="comment"> * @attention The k_offset = a_offset * b_offset * k (where k is the number of matrix A columns) needs to be passed at compile time using -DK_OFFSET (i.e. -DK_OFFSET=1200)</span></div><div class="line"><a name="l01366"></a><span class="lineno"> 1366</span>&#160;<span class="comment"> * @note In case the offset contribution due to a_offset is required, a_offset needs to be passed at compile time using -DA_OFFSET (i.e. -DA_OFFSET=1)</span></div><div class="line"><a name="l01367"></a><span class="lineno"> 1367</span>&#160;<span class="comment"> * @note In case the offset contribution due to b_offset is required, b_offset needs to be passed at compile time using -DB_OFFSET (i.e. -DB_OFFSET=6)</span></div><div class="line"><a name="l01368"></a><span class="lineno"> 1368</span>&#160;<span class="comment"> * @note In case sum_col has batches, -DSUM_COL_HAS_BATCHES must be passed at compile time. Usually if gemmlowp is used to accelerate convolution layer, sum_col will not have batches</span></div><div class="line"><a name="l01369"></a><span class="lineno"> 1369</span>&#160;<span class="comment"> *</span></div><div class="line"><a name="l01370"></a><span class="lineno"> 1370</span>&#160;<span class="comment"> * The result before the output stage is:</span></div><div class="line"><a name="l01371"></a><span class="lineno"> 1371</span>&#160;<span class="comment"> *</span></div><div class="line"><a name="l01372"></a><span class="lineno"> 1372</span>&#160;<span class="comment"> * mm_result[i][k] = mm_result[i][k] +</span></div><div class="line"><a name="l01373"></a><span class="lineno"> 1373</span>&#160;<span class="comment"> * (sum_col[k] * A_OFFSET) +</span></div><div class="line"><a name="l01374"></a><span class="lineno"> 1374</span>&#160;<span class="comment"> * (sum_row[i] * B_OFFSET) +</span></div><div class="line"><a name="l01375"></a><span class="lineno"> 1375</span>&#160;<span class="comment"> * (K_OFFSET)</span></div><div class="line"><a name="l01376"></a><span class="lineno"> 1376</span>&#160;<span class="comment"> *</span></div><div class="line"><a name="l01377"></a><span class="lineno"> 1377</span>&#160;<span class="comment"> * This result is quantized down to uint8 using the output stage. The output stage computes the following operations:</span></div><div class="line"><a name="l01378"></a><span class="lineno"> 1378</span>&#160;<span class="comment"> *</span></div><div class="line"><a name="l01379"></a><span class="lineno"> 1379</span>&#160;<span class="comment"> * -# Add offset terms to final result</span></div><div class="line"><a name="l01380"></a><span class="lineno"> 1380</span>&#160;<span class="comment"> * -# Multiply each entry of result by result_mult_int</span></div><div class="line"><a name="l01381"></a><span class="lineno"> 1381</span>&#160;<span class="comment"> * -# Add bias to final result (if -DADD_BIAS is passed at compile time)</span></div><div class="line"><a name="l01382"></a><span class="lineno"> 1382</span>&#160;<span class="comment"> * -# Shift the int32 accumulator by result_shift</span></div><div class="line"><a name="l01383"></a><span class="lineno"> 1383</span>&#160;<span class="comment"> * -# Clamp the value between the specified min and max bounds (if -DMIN_BOUND and/or -DMAX_BOUND are passed at compile time)</span></div><div class="line"><a name="l01384"></a><span class="lineno"> 1384</span>&#160;<span class="comment"> * -# Clamp the resulting int32 values to the [0..255] range and cast to QASYMM8.</span></div><div class="line"><a name="l01385"></a><span class="lineno"> 1385</span>&#160;<span class="comment"> *</span></div><div class="line"><a name="l01386"></a><span class="lineno"> 1386</span>&#160;<span class="comment"> * @attention The offset, scalar scale factor and number of bits to shift right of output tensor must be passed at compile time using -DRESULT_OFFSET, -RESULT_MULT_INT and -DRESULT_SHIFT</span></div><div class="line"><a name="l01387"></a><span class="lineno"> 1387</span>&#160;<span class="comment"> *</span></div><div class="line"><a name="l01388"></a><span class="lineno"> 1388</span>&#160;<span class="comment"> * @note In case the addition of int32 biases is required, -DADD_BIAS should be passed at compile time</span></div><div class="line"><a name="l01389"></a><span class="lineno"> 1389</span>&#160;<span class="comment"> * @note In case the clamping of the result is required, the min and max bounds can be passed at compile time using -DMIN_BOUND and -DMAX_BOUND.</span></div><div class="line"><a name="l01390"></a><span class="lineno"> 1390</span>&#160;<span class="comment"> * These values can be used to implement &quot;rectified linear unit&quot; activation functions</span></div><div class="line"><a name="l01391"></a><span class="lineno"> 1391</span>&#160;<span class="comment"> *</span></div><div class="line"><a name="l01392"></a><span class="lineno"> 1392</span>&#160;<span class="comment"> * @param[in] mm_result_ptr Pointer to the source tensor. Supported data type: S32</span></div><div class="line"><a name="l01393"></a><span class="lineno"> 1393</span>&#160;<span class="comment"> * @param[in] mm_result_stride_x Stride of the source tensor in X dimension (in bytes)</span></div><div class="line"><a name="l01394"></a><span class="lineno"> 1394</span>&#160;<span class="comment"> * @param[in] mm_result_step_x mm_result_stride_x * number of elements along X processed per workitem(in bytes)</span></div><div class="line"><a name="l01395"></a><span class="lineno"> 1395</span>&#160;<span class="comment"> * @param[in] mm_result_stride_y Stride of the source tensor in Y dimension (in bytes)</span></div><div class="line"><a name="l01396"></a><span class="lineno"> 1396</span>&#160;<span class="comment"> * @param[in] mm_result_step_y mm_result_stride_y * number of elements along Y processed per workitem(in bytes)</span></div><div class="line"><a name="l01397"></a><span class="lineno"> 1397</span>&#160;<span class="comment"> * @param[in] mm_result_stride_z Stride of the source tensor in Z dimension (in bytes)</span></div><div class="line"><a name="l01398"></a><span class="lineno"> 1398</span>&#160;<span class="comment"> * @param[in] mm_result_step_z mm_result_stride_z * number of elements along Z processed per workitem(in bytes)</span></div><div class="line"><a name="l01399"></a><span class="lineno"> 1399</span>&#160;<span class="comment"> * @param[in] mm_result_offset_first_element_in_bytes The offset of the first element in the source tensor</span></div><div class="line"><a name="l01400"></a><span class="lineno"> 1400</span>&#160;<span class="comment"> * @param[in] sum_col_ptr (Optional) Pointer to the source tensor. Supported data type: same as @p mm_result_ptr</span></div><div class="line"><a name="l01401"></a><span class="lineno"> 1401</span>&#160;<span class="comment"> * @param[in] sum_col_stride_x (Optional) Stride of the source tensor in X dimension (in bytes)</span></div><div class="line"><a name="l01402"></a><span class="lineno"> 1402</span>&#160;<span class="comment"> * @param[in] sum_col_step_x (Optional) sum_col_stride_x * number of elements along X processed per workitem(in bytes)</span></div><div class="line"><a name="l01403"></a><span class="lineno"> 1403</span>&#160;<span class="comment"> * @param[in] sum_col_stride_y (Optional) Stride of the source tensor in Y dimension (in bytes)</span></div><div class="line"><a name="l01404"></a><span class="lineno"> 1404</span>&#160;<span class="comment"> * @param[in] sum_col_step_y (Optional) sum_col_stride_y * number of elements along Y processed per workitem(in bytes)</span></div><div class="line"><a name="l01405"></a><span class="lineno"> 1405</span>&#160;<span class="comment"> * @param[in] sum_col_offset_first_element_in_bytes (Optional) The offset of the first element in the source tensor</span></div><div class="line"><a name="l01406"></a><span class="lineno"> 1406</span>&#160;<span class="comment"> * @param[in] sum_row_ptr (Optional) Pointer to the source tensor. Supported data type: same as @p mm_result_ptr</span></div><div class="line"><a name="l01407"></a><span class="lineno"> 1407</span>&#160;<span class="comment"> * @param[in] sum_row_stride_x (Optional) Stride of the source tensor in X dimension (in bytes)</span></div><div class="line"><a name="l01408"></a><span class="lineno"> 1408</span>&#160;<span class="comment"> * @param[in] sum_row_step_x (Optional) sum_row_stride_x * number of elements along X processed per workitem(in bytes)</span></div><div class="line"><a name="l01409"></a><span class="lineno"> 1409</span>&#160;<span class="comment"> * @param[in] sum_row_stride_y (Optional) Stride of the source tensor in Y dimension (in bytes)</span></div><div class="line"><a name="l01410"></a><span class="lineno"> 1410</span>&#160;<span class="comment"> * @param[in] sum_row_step_y (Optional) sum_row_stride_y * number of elements along Y processed per workitem(in bytes)</span></div><div class="line"><a name="l01411"></a><span class="lineno"> 1411</span>&#160;<span class="comment"> * @param[in] sum_row_offset_first_element_in_bytes (Optional) The offset of the first element in the source tensor</span></div><div class="line"><a name="l01412"></a><span class="lineno"> 1412</span>&#160;<span class="comment"> * @param[in] biases_ptr (Optional) Pointer to the biases tensor. Supported data type: same as @p src_ptr</span></div><div class="line"><a name="l01413"></a><span class="lineno"> 1413</span>&#160;<span class="comment"> * @param[in] biases_stride_x (Optional) Stride of the biases tensor in X dimension (in bytes)</span></div><div class="line"><a name="l01414"></a><span class="lineno"> 1414</span>&#160;<span class="comment"> * @param[in] biases_step_x (Optional) biases_stride_x * number of elements along X processed per workitem(in bytes)</span></div><div class="line"><a name="l01415"></a><span class="lineno"> 1415</span>&#160;<span class="comment"> * @param[in] biases_offset_first_element_in_bytes (Optional) The offset of the first element in the biases tensor</span></div><div class="line"><a name="l01416"></a><span class="lineno"> 1416</span>&#160;<span class="comment"> * @param[out] dst_ptr Pointer to the destination tensor Supported data type: QASYMM8</span></div><div class="line"><a name="l01417"></a><span class="lineno"> 1417</span>&#160;<span class="comment"> * @param[in] dst_stride_x Stride of the destination tensor in X dimension (in bytes)</span></div><div class="line"><a name="l01418"></a><span class="lineno"> 1418</span>&#160;<span class="comment"> * @param[in] dst_step_x dst_gx_stride_x * number of elements along X processed per workitem(in bytes)</span></div><div class="line"><a name="l01419"></a><span class="lineno"> 1419</span>&#160;<span class="comment"> * @param[in] dst_stride_y Stride of the destination tensor in Y dimension (in bytes)</span></div><div class="line"><a name="l01420"></a><span class="lineno"> 1420</span>&#160;<span class="comment"> * @param[in] dst_step_y dst_gx_stride_y * number of elements along Y processed per workitem(in bytes)</span></div><div class="line"><a name="l01421"></a><span class="lineno"> 1421</span>&#160;<span class="comment"> * @param[in] dst_stride_z Stride of the source tensor in Z dimension (in bytes)</span></div><div class="line"><a name="l01422"></a><span class="lineno"> 1422</span>&#160;<span class="comment"> * @param[in] dst_step_z src_stride_z * number of elements along Z processed per workitem(in bytes)</span></div><div class="line"><a name="l01423"></a><span class="lineno"> 1423</span>&#160;<span class="comment"> * @param[in] dst_offset_first_element_in_bytes The offset of the first element in the destination tensor</span></div><div class="line"><a name="l01424"></a><span class="lineno"> 1424</span>&#160;<span class="comment"> */</span></div><div class="line"><a name="l01425"></a><span class="lineno"> 1425</span>&#160;__kernel <span class="keywordtype">void</span> gemmlowp_offset_contribution_quantize_down(<a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a6b83038822d1ae7ab619b684ed3b7fc0">TENSOR3D_DECLARATION</a>(mm_result)</div><div class="line"><a name="l01426"></a><span class="lineno"> 1426</span>&#160;#<span class="keywordflow">if</span> defined(A_OFFSET)</div><div class="line"><a name="l01427"></a><span class="lineno"> 1427</span>&#160; ,</div><div class="line"><a name="l01428"></a><span class="lineno"> 1428</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a22f42fcf2077d951271df83b55c1a71a">IMAGE_DECLARATION</a>(sum_col)</div><div class="line"><a name="l01429"></a><span class="lineno"> 1429</span>&#160;#endif <span class="comment">// defined(A_OFFSET)</span></div><div class="line"><a name="l01430"></a><span class="lineno"> 1430</span>&#160;#<span class="keywordflow">if</span> defined(B_OFFSET)</div><div class="line"><a name="l01431"></a><span class="lineno"> 1431</span>&#160; ,</div><div class="line"><a name="l01432"></a><span class="lineno"> 1432</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a22f42fcf2077d951271df83b55c1a71a">IMAGE_DECLARATION</a>(sum_row)</div><div class="line"><a name="l01433"></a><span class="lineno"> 1433</span>&#160;#endif <span class="comment">// defined(B_OFFSET)</span></div><div class="line"><a name="l01434"></a><span class="lineno"> 1434</span>&#160; ,</div><div class="line"><a name="l01435"></a><span class="lineno"> 1435</span>&#160;#<span class="keywordflow">if</span> defined(ADD_BIAS)</div><div class="line"><a name="l01436"></a><span class="lineno"> 1436</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a40a6eb9f2a7712f08d6bb8ff6c9e6ca7">VECTOR_DECLARATION</a>(biases),</div><div class="line"><a name="l01437"></a><span class="lineno"> 1437</span>&#160;#endif <span class="comment">// defined(ADD_BIAS)</span></div><div class="line"><a name="l01438"></a><span class="lineno"> 1438</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_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="l01439"></a><span class="lineno"> 1439</span>&#160;{</div><div class="line"><a name="l01440"></a><span class="lineno"> 1440</span>&#160; <span class="keyword">const</span> <span class="keywordtype">int</span> x = get_global_id(0) * 4;</div><div class="line"><a name="l01441"></a><span class="lineno"> 1441</span>&#160; <span class="keyword">const</span> <span class="keywordtype">int</span> y = get_global_id(1);</div><div class="line"><a name="l01442"></a><span class="lineno"> 1442</span>&#160; <span class="keyword">const</span> <span class="keywordtype">int</span> z = get_global_id(2);</div><div class="line"><a name="l01443"></a><span class="lineno"> 1443</span>&#160;</div><div class="line"><a name="l01444"></a><span class="lineno"> 1444</span>&#160; __global uchar *dst_addr = dst_ptr + dst_offset_first_element_in_bytes + x + y * dst_stride_y + z * dst_stride_z;</div><div class="line"><a name="l01445"></a><span class="lineno"> 1445</span>&#160;</div><div class="line"><a name="l01446"></a><span class="lineno"> 1446</span>&#160; <span class="comment">// Compute offset contribution</span></div><div class="line"><a name="l01447"></a><span class="lineno"> 1447</span>&#160; int4 offset_term_s32 = offset_contribution(</div><div class="line"><a name="l01448"></a><span class="lineno"> 1448</span>&#160; x, y, z</div><div class="line"><a name="l01449"></a><span class="lineno"> 1449</span>&#160;#<span class="keywordflow">if</span> defined(A_OFFSET)</div><div class="line"><a name="l01450"></a><span class="lineno"> 1450</span>&#160; ,</div><div class="line"><a name="l01451"></a><span class="lineno"> 1451</span>&#160; sum_col_ptr,</div><div class="line"><a name="l01452"></a><span class="lineno"> 1452</span>&#160; sum_col_stride_x,</div><div class="line"><a name="l01453"></a><span class="lineno"> 1453</span>&#160; sum_col_step_x,</div><div class="line"><a name="l01454"></a><span class="lineno"> 1454</span>&#160; sum_col_stride_y,</div><div class="line"><a name="l01455"></a><span class="lineno"> 1455</span>&#160; sum_col_step_y,</div><div class="line"><a name="l01456"></a><span class="lineno"> 1456</span>&#160; sum_col_offset_first_element_in_bytes</div><div class="line"><a name="l01457"></a><span class="lineno"> 1457</span>&#160;#endif <span class="comment">// defined(A_OFFSET)</span></div><div class="line"><a name="l01458"></a><span class="lineno"> 1458</span>&#160;#<span class="keywordflow">if</span> defined(B_OFFSET)</div><div class="line"><a name="l01459"></a><span class="lineno"> 1459</span>&#160; ,</div><div class="line"><a name="l01460"></a><span class="lineno"> 1460</span>&#160; sum_row_ptr,</div><div class="line"><a name="l01461"></a><span class="lineno"> 1461</span>&#160; sum_row_stride_x,</div><div class="line"><a name="l01462"></a><span class="lineno"> 1462</span>&#160; sum_row_step_x,</div><div class="line"><a name="l01463"></a><span class="lineno"> 1463</span>&#160; sum_row_stride_y,</div><div class="line"><a name="l01464"></a><span class="lineno"> 1464</span>&#160; sum_row_step_y,</div><div class="line"><a name="l01465"></a><span class="lineno"> 1465</span>&#160; sum_row_offset_first_element_in_bytes</div><div class="line"><a name="l01466"></a><span class="lineno"> 1466</span>&#160;#endif <span class="comment">// defined(B_OFFSET)</span></div><div class="line"><a name="l01467"></a><span class="lineno"> 1467</span>&#160;#<span class="keywordflow">if</span> defined(ADD_BIAS)</div><div class="line"><a name="l01468"></a><span class="lineno"> 1468</span>&#160; ,</div><div class="line"><a name="l01469"></a><span class="lineno"> 1469</span>&#160; biases_ptr,</div><div class="line"><a name="l01470"></a><span class="lineno"> 1470</span>&#160; biases_stride_x,</div><div class="line"><a name="l01471"></a><span class="lineno"> 1471</span>&#160; biases_step_x,</div><div class="line"><a name="l01472"></a><span class="lineno"> 1472</span>&#160; biases_offset_first_element_in_bytes</div><div class="line"><a name="l01473"></a><span class="lineno"> 1473</span>&#160;#endif <span class="comment">// defined(ADD_BIAS)</span></div><div class="line"><a name="l01474"></a><span class="lineno"> 1474</span>&#160; );</div><div class="line"><a name="l01475"></a><span class="lineno"> 1475</span>&#160;</div><div class="line"><a name="l01476"></a><span class="lineno"> 1476</span>&#160; __global uchar *mm_result_addr = mm_result_ptr + mm_result_offset_first_element_in_bytes + x * <span class="keyword">sizeof</span>(int) + y * mm_result_stride_y + z * mm_result_stride_z;</div><div class="line"><a name="l01477"></a><span class="lineno"> 1477</span>&#160;</div><div class="line"><a name="l01478"></a><span class="lineno"> 1478</span>&#160; int4 in_s32 = vload4(0, (__global <span class="keywordtype">int</span> *)mm_result_addr);</div><div class="line"><a name="l01479"></a><span class="lineno"> 1479</span>&#160;</div><div class="line"><a name="l01480"></a><span class="lineno"> 1480</span>&#160; <span class="comment">// Add the offset terms to GEMM&#39;s result</span></div><div class="line"><a name="l01481"></a><span class="lineno"> 1481</span>&#160; in_s32 += offset_term_s32;</div><div class="line"><a name="l01482"></a><span class="lineno"> 1482</span>&#160;</div><div class="line"><a name="l01483"></a><span class="lineno"> 1483</span>&#160; <span class="comment">// -------------- OUTPUT STAGE</span></div><div class="line"><a name="l01484"></a><span class="lineno"> 1484</span>&#160;</div><div class="line"><a name="l01485"></a><span class="lineno"> 1485</span>&#160; <span class="comment">// Add the offset terms to GEMM&#39;s result</span></div><div class="line"><a name="l01486"></a><span class="lineno"> 1486</span>&#160; in_s32 += (int4)RESULT_OFFSET;</div><div class="line"><a name="l01487"></a><span class="lineno"> 1487</span>&#160;</div><div class="line"><a name="l01488"></a><span class="lineno"> 1488</span>&#160; <span class="comment">// Multiply by result_mult_int and shift</span></div><div class="line"><a name="l01489"></a><span class="lineno"> 1489</span>&#160; in_s32 *= RESULT_MULTIPLIER;</div><div class="line"><a name="l01490"></a><span class="lineno"> 1490</span>&#160;</div><div class="line"><a name="l01491"></a><span class="lineno"> 1491</span>&#160; in_s32 &gt;&gt;= RESULT_SHIFT;</div><div class="line"><a name="l01492"></a><span class="lineno"> 1492</span>&#160;</div><div class="line"><a name="l01493"></a><span class="lineno"> 1493</span>&#160; uchar4 res = convert_uchar4_sat(in_s32);</div><div class="line"><a name="l01494"></a><span class="lineno"> 1494</span>&#160;</div><div class="line"><a name="l01495"></a><span class="lineno"> 1495</span>&#160;<span class="preprocessor">#if defined(MIN_BOUND)</span></div><div class="line"><a name="l01496"></a><span class="lineno"> 1496</span>&#160; res = max(res, (uchar4)MIN_BOUND);</div><div class="line"><a name="l01497"></a><span class="lineno"> 1497</span>&#160;<span class="preprocessor">#endif // defined(MIN_BOUND)</span></div><div class="line"><a name="l01498"></a><span class="lineno"> 1498</span>&#160;<span class="preprocessor">#if defined(MAX_BOUND)</span></div><div class="line"><a name="l01499"></a><span class="lineno"> 1499</span>&#160; res = min(res, (uchar4)MAX_BOUND);</div><div class="line"><a name="l01500"></a><span class="lineno"> 1500</span>&#160;<span class="preprocessor">#endif // defined(MAX_BOUND)</span></div><div class="line"><a name="l01501"></a><span class="lineno"> 1501</span>&#160;</div><div class="line"><a name="l01502"></a><span class="lineno"> 1502</span>&#160; <span class="comment">// Store the result</span></div><div class="line"><a name="l01503"></a><span class="lineno"> 1503</span>&#160; vstore4(res, 0, dst_addr);</div><div class="line"><a name="l01504"></a><span class="lineno"> 1504</span>&#160;}</div><div class="line"><a name="l01505"></a><span class="lineno"> 1505</span>&#160;</div><div class="line"><a name="l01506"></a><span class="lineno"> 1506</span>&#160;<span class="comment">/* OpenCL kernel used to add the offset contribution after @ref CLGEMMLowpMatrixMultiplyKernel and it quantizes down to uint8.</span></div><div class="line"><a name="l01507"></a><span class="lineno"> 1507</span>&#160;<span class="comment"> *</span></div><div class="line"><a name="l01508"></a><span class="lineno"> 1508</span>&#160;<span class="comment"> * This kernel takes a final int32 accumulator value (the output of @CLGEMMLowpMatrixMultiplyKernel), adds to it the offset contribution of matrix A and matrix B and quantizes to uint8 through the output stage.</span></div><div class="line"><a name="l01509"></a><span class="lineno"> 1509</span>&#160;<span class="comment"> *</span></div><div class="line"><a name="l01510"></a><span class="lineno"> 1510</span>&#160;<span class="comment"> *</span></div><div class="line"><a name="l01511"></a><span class="lineno"> 1511</span>&#160;<span class="comment"> * @attention The k_offset = a_offset * b_offset * k (where k is the number of matrix A columns) needs to be passed at compile time using -DK_OFFSET (i.e. -DK_OFFSET=1200)</span></div><div class="line"><a name="l01512"></a><span class="lineno"> 1512</span>&#160;<span class="comment"> * @note In case the offset contribution due to a_offset is required, a_offset needs to be passed at compile time using -DA_OFFSET (i.e. -DA_OFFSET=1)</span></div><div class="line"><a name="l01513"></a><span class="lineno"> 1513</span>&#160;<span class="comment"> * @note In case the offset contribution due to b_offset is required, b_offset needs to be passed at compile time using -DB_OFFSET (i.e. -DB_OFFSET=6)</span></div><div class="line"><a name="l01514"></a><span class="lineno"> 1514</span>&#160;<span class="comment"> * @note In case sum_col has batches, -DSUM_COL_HAS_BATCHES must be passed at compile time. Usually if gemmlowp is used to accelerate convolution layer, sum_col will not have batches</span></div><div class="line"><a name="l01515"></a><span class="lineno"> 1515</span>&#160;<span class="comment"> *</span></div><div class="line"><a name="l01516"></a><span class="lineno"> 1516</span>&#160;<span class="comment"> * The result before the output stage is:</span></div><div class="line"><a name="l01517"></a><span class="lineno"> 1517</span>&#160;<span class="comment"> *</span></div><div class="line"><a name="l01518"></a><span class="lineno"> 1518</span>&#160;<span class="comment"> * mm_result[i][k] = mm_result[i][k] +</span></div><div class="line"><a name="l01519"></a><span class="lineno"> 1519</span>&#160;<span class="comment"> * (sum_col[k] * A_OFFSET) +</span></div><div class="line"><a name="l01520"></a><span class="lineno"> 1520</span>&#160;<span class="comment"> * (sum_row[i] * B_OFFSET) +</span></div><div class="line"><a name="l01521"></a><span class="lineno"> 1521</span>&#160;<span class="comment"> * (K_OFFSET)</span></div><div class="line"><a name="l01522"></a><span class="lineno"> 1522</span>&#160;<span class="comment"> *</span></div><div class="line"><a name="l01523"></a><span class="lineno"> 1523</span>&#160;<span class="comment"> * This result is quantized down to uint8 using the output stage. The output stage computes the following operations:</span></div><div class="line"><a name="l01524"></a><span class="lineno"> 1524</span>&#160;<span class="comment"> *</span></div><div class="line"><a name="l01525"></a><span class="lineno"> 1525</span>&#160;<span class="comment"> * -# Compute fixed point multiplication between each entry of input by result_fixedpoint_multiplier</span></div><div class="line"><a name="l01526"></a><span class="lineno"> 1526</span>&#160;<span class="comment"> * -# Add bias to final result if bias tensor is not a nullptr</span></div><div class="line"><a name="l01527"></a><span class="lineno"> 1527</span>&#160;<span class="comment"> * -# Round to nearest division by a power-of-two using result_shift</span></div><div class="line"><a name="l01528"></a><span class="lineno"> 1528</span>&#160;<span class="comment"> * -# Add offset to each result</span></div><div class="line"><a name="l01529"></a><span class="lineno"> 1529</span>&#160;<span class="comment"> * -# Clamp the value between the specified min and max bounds</span></div><div class="line"><a name="l01530"></a><span class="lineno"> 1530</span>&#160;<span class="comment"> * -# Clamp the resulting int32 values to the [0..255] range and cast to QASYMM8.</span></div><div class="line"><a name="l01531"></a><span class="lineno"> 1531</span>&#160;<span class="comment"> *</span></div><div class="line"><a name="l01532"></a><span class="lineno"> 1532</span>&#160;<span class="comment"> * @attention The offset, scalar scale factor and number of bits to shift right of output tensor must be passed at compile time using -DRESULT_OFFSET, -RESULT_MULT_INT and -DRESULT_SHIFT</span></div><div class="line"><a name="l01533"></a><span class="lineno"> 1533</span>&#160;<span class="comment"> *</span></div><div class="line"><a name="l01534"></a><span class="lineno"> 1534</span>&#160;<span class="comment"> * @note In case the addition of int32 biases is required, -DADD_BIAS should be passed at compile time</span></div><div class="line"><a name="l01535"></a><span class="lineno"> 1535</span>&#160;<span class="comment"> * @note In case the clamping of the result is required, the min and max bounds can be passed at compile time using -DMIN_BOUND and -DMAX_BOUND.</span></div><div class="line"><a name="l01536"></a><span class="lineno"> 1536</span>&#160;<span class="comment"> * These values can be used to implement &quot;rectified linear unit&quot; activation functions</span></div><div class="line"><a name="l01537"></a><span class="lineno"> 1537</span>&#160;<span class="comment"> *</span></div><div class="line"><a name="l01538"></a><span class="lineno"> 1538</span>&#160;<span class="comment"> * @param[in] mm_result_ptr Pointer to the source tensor. Supported data type: S32</span></div><div class="line"><a name="l01539"></a><span class="lineno"> 1539</span>&#160;<span class="comment"> * @param[in] mm_result_stride_x Stride of the source tensor in X dimension (in bytes)</span></div><div class="line"><a name="l01540"></a><span class="lineno"> 1540</span>&#160;<span class="comment"> * @param[in] mm_result_step_x mm_result_stride_x * number of elements along X processed per workitem(in bytes)</span></div><div class="line"><a name="l01541"></a><span class="lineno"> 1541</span>&#160;<span class="comment"> * @param[in] mm_result_stride_y Stride of the source tensor in Y dimension (in bytes)</span></div><div class="line"><a name="l01542"></a><span class="lineno"> 1542</span>&#160;<span class="comment"> * @param[in] mm_result_step_y mm_result_stride_y * number of elements along Y processed per workitem(in bytes)</span></div><div class="line"><a name="l01543"></a><span class="lineno"> 1543</span>&#160;<span class="comment"> * @param[in] mm_result_stride_z Stride of the source tensor in Z dimension (in bytes)</span></div><div class="line"><a name="l01544"></a><span class="lineno"> 1544</span>&#160;<span class="comment"> * @param[in] mm_result_step_z mm_result_stride_z * number of elements along Z processed per workitem(in bytes)</span></div><div class="line"><a name="l01545"></a><span class="lineno"> 1545</span>&#160;<span class="comment"> * @param[in] mm_result_offset_first_element_in_bytes The offset of the first element in the source tensor</span></div><div class="line"><a name="l01546"></a><span class="lineno"> 1546</span>&#160;<span class="comment"> * @param[in] sum_col_ptr (Optional) Pointer to the source tensor. Supported data type: same as @p mm_result_ptr</span></div><div class="line"><a name="l01547"></a><span class="lineno"> 1547</span>&#160;<span class="comment"> * @param[in] sum_col_stride_x (Optional) Stride of the source tensor in X dimension (in bytes)</span></div><div class="line"><a name="l01548"></a><span class="lineno"> 1548</span>&#160;<span class="comment"> * @param[in] sum_col_step_x (Optional) sum_col_stride_x * number of elements along X processed per workitem(in bytes)</span></div><div class="line"><a name="l01549"></a><span class="lineno"> 1549</span>&#160;<span class="comment"> * @param[in] sum_col_stride_y (Optional) Stride of the source tensor in Y dimension (in bytes)</span></div><div class="line"><a name="l01550"></a><span class="lineno"> 1550</span>&#160;<span class="comment"> * @param[in] sum_col_step_y (Optional) sum_col_stride_y * number of elements along Y processed per workitem(in bytes)</span></div><div class="line"><a name="l01551"></a><span class="lineno"> 1551</span>&#160;<span class="comment"> * @param[in] sum_col_offset_first_element_in_bytes (Optional) The offset of the first element in the source tensor</span></div><div class="line"><a name="l01552"></a><span class="lineno"> 1552</span>&#160;<span class="comment"> * @param[in] sum_row_ptr (Optional) Pointer to the source tensor. Supported data type: same as @p mm_result_ptr</span></div><div class="line"><a name="l01553"></a><span class="lineno"> 1553</span>&#160;<span class="comment"> * @param[in] sum_row_stride_x (Optional) Stride of the source tensor in X dimension (in bytes)</span></div><div class="line"><a name="l01554"></a><span class="lineno"> 1554</span>&#160;<span class="comment"> * @param[in] sum_row_step_x (Optional) sum_row_stride_x * number of elements along X processed per workitem(in bytes)</span></div><div class="line"><a name="l01555"></a><span class="lineno"> 1555</span>&#160;<span class="comment"> * @param[in] sum_row_stride_y (Optional) Stride of the source tensor in Y dimension (in bytes)</span></div><div class="line"><a name="l01556"></a><span class="lineno"> 1556</span>&#160;<span class="comment"> * @param[in] sum_row_step_y (Optional) sum_row_stride_y * number of elements along Y processed per workitem(in bytes)</span></div><div class="line"><a name="l01557"></a><span class="lineno"> 1557</span>&#160;<span class="comment"> * @param[in] sum_row_offset_first_element_in_bytes (Optional) The offset of the first element in the source tensor</span></div><div class="line"><a name="l01558"></a><span class="lineno"> 1558</span>&#160;<span class="comment"> * @param[in] biases_ptr (Optional) Pointer to the biases tensor. Supported data type: same as @p src_ptr</span></div><div class="line"><a name="l01559"></a><span class="lineno"> 1559</span>&#160;<span class="comment"> * @param[in] biases_stride_x (Optional) Stride of the biases tensor in X dimension (in bytes)</span></div><div class="line"><a name="l01560"></a><span class="lineno"> 1560</span>&#160;<span class="comment"> * @param[in] biases_step_x (Optional) biases_stride_x * number of elements along X processed per workitem(in bytes)</span></div><div class="line"><a name="l01561"></a><span class="lineno"> 1561</span>&#160;<span class="comment"> * @param[in] biases_offset_first_element_in_bytes (Optional) The offset of the first element in the biases tensor</span></div><div class="line"><a name="l01562"></a><span class="lineno"> 1562</span>&#160;<span class="comment"> * @param[out] dst_ptr Pointer to the destination tensor Supported data type: QASYMM8</span></div><div class="line"><a name="l01563"></a><span class="lineno"> 1563</span>&#160;<span class="comment"> * @param[in] dst_stride_x Stride of the destination tensor in X dimension (in bytes)</span></div><div class="line"><a name="l01564"></a><span class="lineno"> 1564</span>&#160;<span class="comment"> * @param[in] dst_step_x dst_gx_stride_x * number of elements along X processed per workitem(in bytes)</span></div><div class="line"><a name="l01565"></a><span class="lineno"> 1565</span>&#160;<span class="comment"> * @param[in] dst_stride_y Stride of the destination tensor in Y dimension (in bytes)</span></div><div class="line"><a name="l01566"></a><span class="lineno"> 1566</span>&#160;<span class="comment"> * @param[in] dst_step_y dst_gx_stride_y * number of elements along Y processed per workitem(in bytes)</span></div><div class="line"><a name="l01567"></a><span class="lineno"> 1567</span>&#160;<span class="comment"> * @param[in] dst_stride_z Stride of the source tensor in Z dimension (in bytes)</span></div><div class="line"><a name="l01568"></a><span class="lineno"> 1568</span>&#160;<span class="comment"> * @param[in] dst_step_z src_stride_z * number of elements along Z processed per workitem(in bytes)</span></div><div class="line"><a name="l01569"></a><span class="lineno"> 1569</span>&#160;<span class="comment"> * @param[in] dst_offset_first_element_in_bytes The offset of the first element in the destination tensor</span></div><div class="line"><a name="l01570"></a><span class="lineno"> 1570</span>&#160;<span class="comment"> */</span></div><div class="line"><a name="l01571"></a><span class="lineno"> 1571</span>&#160;__kernel <span class="keywordtype">void</span> gemmlowp_offset_contribution_quantize_down_fixedpoint(<a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a6b83038822d1ae7ab619b684ed3b7fc0">TENSOR3D_DECLARATION</a>(mm_result)</div><div class="line"><a name="l01572"></a><span class="lineno"> 1572</span>&#160;#<span class="keywordflow">if</span> defined(A_OFFSET)</div><div class="line"><a name="l01573"></a><span class="lineno"> 1573</span>&#160; ,</div><div class="line"><a name="l01574"></a><span class="lineno"> 1574</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a22f42fcf2077d951271df83b55c1a71a">IMAGE_DECLARATION</a>(sum_col)</div><div class="line"><a name="l01575"></a><span class="lineno"> 1575</span>&#160;#endif <span class="comment">// defined(A_OFFSET)</span></div><div class="line"><a name="l01576"></a><span class="lineno"> 1576</span>&#160;#<span class="keywordflow">if</span> defined(B_OFFSET)</div><div class="line"><a name="l01577"></a><span class="lineno"> 1577</span>&#160; ,</div><div class="line"><a name="l01578"></a><span class="lineno"> 1578</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a22f42fcf2077d951271df83b55c1a71a">IMAGE_DECLARATION</a>(sum_row)</div><div class="line"><a name="l01579"></a><span class="lineno"> 1579</span>&#160;#endif <span class="comment">// defined(B_OFFSET)</span></div><div class="line"><a name="l01580"></a><span class="lineno"> 1580</span>&#160; ,</div><div class="line"><a name="l01581"></a><span class="lineno"> 1581</span>&#160;#<span class="keywordflow">if</span> defined(ADD_BIAS)</div><div class="line"><a name="l01582"></a><span class="lineno"> 1582</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a40a6eb9f2a7712f08d6bb8ff6c9e6ca7">VECTOR_DECLARATION</a>(biases),</div><div class="line"><a name="l01583"></a><span class="lineno"> 1583</span>&#160;#endif <span class="comment">// defined(ADD_BIAS)</span></div><div class="line"><a name="l01584"></a><span class="lineno"> 1584</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_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="l01585"></a><span class="lineno"> 1585</span>&#160;{</div><div class="line"><a name="l01586"></a><span class="lineno"> 1586</span>&#160; <span class="keyword">const</span> <span class="keywordtype">int</span> x = get_global_id(0) * 4;</div><div class="line"><a name="l01587"></a><span class="lineno"> 1587</span>&#160; <span class="keyword">const</span> <span class="keywordtype">int</span> y = get_global_id(1);</div><div class="line"><a name="l01588"></a><span class="lineno"> 1588</span>&#160; <span class="keyword">const</span> <span class="keywordtype">int</span> z = get_global_id(2);</div><div class="line"><a name="l01589"></a><span class="lineno"> 1589</span>&#160;</div><div class="line"><a name="l01590"></a><span class="lineno"> 1590</span>&#160; <span class="comment">// Compute offset contribution</span></div><div class="line"><a name="l01591"></a><span class="lineno"> 1591</span>&#160; int4 offset_term_s32 = offset_contribution(</div><div class="line"><a name="l01592"></a><span class="lineno"> 1592</span>&#160; x, y, z</div><div class="line"><a name="l01593"></a><span class="lineno"> 1593</span>&#160;#<span class="keywordflow">if</span> defined(A_OFFSET)</div><div class="line"><a name="l01594"></a><span class="lineno"> 1594</span>&#160; ,</div><div class="line"><a name="l01595"></a><span class="lineno"> 1595</span>&#160; sum_col_ptr,</div><div class="line"><a name="l01596"></a><span class="lineno"> 1596</span>&#160; sum_col_stride_x,</div><div class="line"><a name="l01597"></a><span class="lineno"> 1597</span>&#160; sum_col_step_x,</div><div class="line"><a name="l01598"></a><span class="lineno"> 1598</span>&#160; sum_col_stride_y,</div><div class="line"><a name="l01599"></a><span class="lineno"> 1599</span>&#160; sum_col_step_y,</div><div class="line"><a name="l01600"></a><span class="lineno"> 1600</span>&#160; sum_col_offset_first_element_in_bytes</div><div class="line"><a name="l01601"></a><span class="lineno"> 1601</span>&#160;#endif <span class="comment">// defined(A_OFFSET)</span></div><div class="line"><a name="l01602"></a><span class="lineno"> 1602</span>&#160;#<span class="keywordflow">if</span> defined(B_OFFSET)</div><div class="line"><a name="l01603"></a><span class="lineno"> 1603</span>&#160; ,</div><div class="line"><a name="l01604"></a><span class="lineno"> 1604</span>&#160; sum_row_ptr,</div><div class="line"><a name="l01605"></a><span class="lineno"> 1605</span>&#160; sum_row_stride_x,</div><div class="line"><a name="l01606"></a><span class="lineno"> 1606</span>&#160; sum_row_step_x,</div><div class="line"><a name="l01607"></a><span class="lineno"> 1607</span>&#160; sum_row_stride_y,</div><div class="line"><a name="l01608"></a><span class="lineno"> 1608</span>&#160; sum_row_step_y,</div><div class="line"><a name="l01609"></a><span class="lineno"> 1609</span>&#160; sum_row_offset_first_element_in_bytes</div><div class="line"><a name="l01610"></a><span class="lineno"> 1610</span>&#160;#endif <span class="comment">// defined(B_OFFSET)</span></div><div class="line"><a name="l01611"></a><span class="lineno"> 1611</span>&#160;#<span class="keywordflow">if</span> defined(ADD_BIAS)</div><div class="line"><a name="l01612"></a><span class="lineno"> 1612</span>&#160; ,</div><div class="line"><a name="l01613"></a><span class="lineno"> 1613</span>&#160; biases_ptr,</div><div class="line"><a name="l01614"></a><span class="lineno"> 1614</span>&#160; biases_stride_x,</div><div class="line"><a name="l01615"></a><span class="lineno"> 1615</span>&#160; biases_step_x,</div><div class="line"><a name="l01616"></a><span class="lineno"> 1616</span>&#160; biases_offset_first_element_in_bytes</div><div class="line"><a name="l01617"></a><span class="lineno"> 1617</span>&#160;#endif <span class="comment">// defined(ADD_BIAS)</span></div><div class="line"><a name="l01618"></a><span class="lineno"> 1618</span>&#160; );</div><div class="line"><a name="l01619"></a><span class="lineno"> 1619</span>&#160;</div><div class="line"><a name="l01620"></a><span class="lineno"> 1620</span>&#160; __global uchar *mm_result_addr = mm_result_ptr + mm_result_offset_first_element_in_bytes + x * <span class="keyword">sizeof</span>(int) + y * mm_result_stride_y + z * mm_result_stride_z;</div><div class="line"><a name="l01621"></a><span class="lineno"> 1621</span>&#160;</div><div class="line"><a name="l01622"></a><span class="lineno"> 1622</span>&#160; __global uchar *dst_addr = dst_ptr + dst_offset_first_element_in_bytes + x + y * dst_stride_y + z * dst_stride_z;</div><div class="line"><a name="l01623"></a><span class="lineno"> 1623</span>&#160;</div><div class="line"><a name="l01624"></a><span class="lineno"> 1624</span>&#160; int4 in_s32 = vload4(0, (__global <span class="keywordtype">int</span> *)mm_result_addr);</div><div class="line"><a name="l01625"></a><span class="lineno"> 1625</span>&#160;</div><div class="line"><a name="l01626"></a><span class="lineno"> 1626</span>&#160; <span class="comment">// Add the offset terms to GEMM&#39;s result</span></div><div class="line"><a name="l01627"></a><span class="lineno"> 1627</span>&#160; in_s32 += offset_term_s32;</div><div class="line"><a name="l01628"></a><span class="lineno"> 1628</span>&#160;</div><div class="line"><a name="l01629"></a><span class="lineno"> 1629</span>&#160; <span class="comment">// -------------- OUTPUT STAGE</span></div><div class="line"><a name="l01630"></a><span class="lineno"> 1630</span>&#160;</div><div class="line"><a name="l01631"></a><span class="lineno"> 1631</span>&#160; <span class="comment">// Multiply by result_mult_int and shift</span></div><div class="line"><a name="l01632"></a><span class="lineno"> 1632</span>&#160; in_s32 = <a class="code" href="helpers__asymm_8h.xhtml#a86de0ffca367bfcf27a8ae5dd1fdef2d">ASYMM_MULT_BY_QUANT_MULTIPLIER_LESS_THAN_ONE</a>(in_s32, RESULT_MULTIPLIER, RESULT_SHIFT, 4);</div><div class="line"><a name="l01633"></a><span class="lineno"> 1633</span>&#160;</div><div class="line"><a name="l01634"></a><span class="lineno"> 1634</span>&#160; <span class="comment">// Add the offset terms to GEMM&#39;s result</span></div><div class="line"><a name="l01635"></a><span class="lineno"> 1635</span>&#160; in_s32 += (int4)RESULT_OFFSET;</div><div class="line"><a name="l01636"></a><span class="lineno"> 1636</span>&#160;</div><div class="line"><a name="l01637"></a><span class="lineno"> 1637</span>&#160; uchar4 res = convert_uchar4_sat(in_s32);</div><div class="line"><a name="l01638"></a><span class="lineno"> 1638</span>&#160;</div><div class="line"><a name="l01639"></a><span class="lineno"> 1639</span>&#160;<span class="preprocessor">#if defined(MIN_BOUND)</span></div><div class="line"><a name="l01640"></a><span class="lineno"> 1640</span>&#160; res = max(res, (uchar4)MIN_BOUND);</div><div class="line"><a name="l01641"></a><span class="lineno"> 1641</span>&#160;<span class="preprocessor">#endif // defined(MIN_BOUND)</span></div><div class="line"><a name="l01642"></a><span class="lineno"> 1642</span>&#160;<span class="preprocessor">#if defined(MAX_BOUND)</span></div><div class="line"><a name="l01643"></a><span class="lineno"> 1643</span>&#160; res = min(res, (uchar4)MAX_BOUND);</div><div class="line"><a name="l01644"></a><span class="lineno"> 1644</span>&#160;<span class="preprocessor">#endif // defined(MAX_BOUND)</span></div><div class="line"><a name="l01645"></a><span class="lineno"> 1645</span>&#160;</div><div class="line"><a name="l01646"></a><span class="lineno"> 1646</span>&#160; <span class="comment">// Store the result</span></div><div class="line"><a name="l01647"></a><span class="lineno"> 1647</span>&#160; vstore4(res, 0, dst_addr);</div><div class="line"><a name="l01648"></a><span class="lineno"> 1648</span>&#160;}</div><div class="line"><a name="l01649"></a><span class="lineno"> 1649</span>&#160;<span class="preprocessor">#endif // defined(K_OFFSET) &amp;&amp; defined(RESULT_OFFSET) &amp;&amp; defined(RESULT_MULTIPLIER) &amp;&amp; defined(RESULT_SHIFT)</span></div><div class="line"><a name="l01650"></a><span class="lineno"> 1650</span>&#160;<span class="preprocessor">#endif // defined(K_OFFSET)</span></div><div class="line"><a name="l01651"></a><span class="lineno"> 1651</span>&#160;</div><div class="line"><a name="l01652"></a><span class="lineno"> 1652</span>&#160;<span class="preprocessor">#if defined(RESULT_OFFSET) &amp;&amp; defined(RESULT_MULT_INT) &amp;&amp; defined(RESULT_SHIFT)</span></div><div class="line"><a name="l01653"></a><span class="lineno"> 1653</span>&#160;</div><div class="line"><a name="l01692"></a><span class="lineno"> 1692</span>&#160;__kernel <span class="keywordtype">void</span> gemmlowp_output_stage_quantize_down(<a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a6b83038822d1ae7ab619b684ed3b7fc0">TENSOR3D_DECLARATION</a>(<a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#a989ab3e96426615bb98e04e0235088ca">src</a>),</div><div class="line"><a name="l01693"></a><span class="lineno"> 1693</span>&#160;#<span class="keywordflow">if</span> defined(ADD_BIAS)</div><div class="line"><a name="l01694"></a><span class="lineno"> 1694</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a40a6eb9f2a7712f08d6bb8ff6c9e6ca7">VECTOR_DECLARATION</a>(biases),</div><div class="line"><a name="l01695"></a><span class="lineno"> 1695</span>&#160;#endif <span class="comment">// defined(ADD_BIAS)</span></div><div class="line"><a name="l01696"></a><span class="lineno"> 1696</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_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="l01697"></a><span class="lineno"> 1697</span>&#160;{</div><div class="line"><a name="l01698"></a><span class="lineno"> 1698</span>&#160; <span class="comment">// Compute source and destination addresses</span></div><div class="line"><a name="l01699"></a><span class="lineno"> 1699</span>&#160; <span class="keywordtype">int</span> x = get_global_id(0) * 4;</div><div class="line"><a name="l01700"></a><span class="lineno"> 1700</span>&#160; <span class="keywordtype">int</span> y = get_global_id(1);</div><div class="line"><a name="l01701"></a><span class="lineno"> 1701</span>&#160; <span class="keywordtype">int</span> z = get_global_id(2);</div><div class="line"><a name="l01702"></a><span class="lineno"> 1702</span>&#160;</div><div class="line"><a name="l01703"></a><span class="lineno"> 1703</span>&#160; __global uchar *src_addr = src_ptr + src_offset_first_element_in_bytes + x * <span class="keyword">sizeof</span>(int) + y * src_stride_y + z * src_stride_z;</div><div class="line"><a name="l01704"></a><span class="lineno"> 1704</span>&#160;</div><div class="line"><a name="l01705"></a><span class="lineno"> 1705</span>&#160; __global uchar *dst_addr = dst_ptr + dst_offset_first_element_in_bytes + x + y * dst_stride_y + z * dst_stride_z;</div><div class="line"><a name="l01706"></a><span class="lineno"> 1706</span>&#160;</div><div class="line"><a name="l01707"></a><span class="lineno"> 1707</span>&#160; int4 input_values = vload4(0, (__global <span class="keywordtype">int</span> *)src_addr);</div><div class="line"><a name="l01708"></a><span class="lineno"> 1708</span>&#160;</div><div class="line"><a name="l01709"></a><span class="lineno"> 1709</span>&#160;<span class="preprocessor">#if defined(ADD_BIAS)</span></div><div class="line"><a name="l01710"></a><span class="lineno"> 1710</span>&#160; <span class="comment">// Add bias</span></div><div class="line"><a name="l01711"></a><span class="lineno"> 1711</span>&#160; __global uchar *bias_addr = biases_ptr + biases_offset_first_element_in_bytes + x * <span class="keyword">sizeof</span>(int);</div><div class="line"><a name="l01712"></a><span class="lineno"> 1712</span>&#160;</div><div class="line"><a name="l01713"></a><span class="lineno"> 1713</span>&#160; int4 biases_values = vload4(0, (__global <span class="keywordtype">int</span> *)bias_addr);</div><div class="line"><a name="l01714"></a><span class="lineno"> 1714</span>&#160; input_values += (int4)biases_values;</div><div class="line"><a name="l01715"></a><span class="lineno"> 1715</span>&#160;<span class="preprocessor">#endif // defined(ADD_BIAS)</span></div><div class="line"><a name="l01716"></a><span class="lineno"> 1716</span>&#160;</div><div class="line"><a name="l01717"></a><span class="lineno"> 1717</span>&#160; <span class="comment">// Add the offset terms to GEMM&#39;s result</span></div><div class="line"><a name="l01718"></a><span class="lineno"> 1718</span>&#160; input_values += (int4)RESULT_OFFSET;</div><div class="line"><a name="l01719"></a><span class="lineno"> 1719</span>&#160;</div><div class="line"><a name="l01720"></a><span class="lineno"> 1720</span>&#160; <span class="comment">// Multiply by result_mult_int and shift</span></div><div class="line"><a name="l01721"></a><span class="lineno"> 1721</span>&#160; input_values *= RESULT_MULT_INT;</div><div class="line"><a name="l01722"></a><span class="lineno"> 1722</span>&#160;</div><div class="line"><a name="l01723"></a><span class="lineno"> 1723</span>&#160; input_values &gt;&gt;= RESULT_SHIFT;</div><div class="line"><a name="l01724"></a><span class="lineno"> 1724</span>&#160;</div><div class="line"><a name="l01725"></a><span class="lineno"> 1725</span>&#160; uchar4 res = convert_uchar4_sat(input_values);</div><div class="line"><a name="l01726"></a><span class="lineno"> 1726</span>&#160;</div><div class="line"><a name="l01727"></a><span class="lineno"> 1727</span>&#160;<span class="preprocessor">#if defined(MIN_BOUND)</span></div><div class="line"><a name="l01728"></a><span class="lineno"> 1728</span>&#160; res = max(res, (uchar4)MIN_BOUND);</div><div class="line"><a name="l01729"></a><span class="lineno"> 1729</span>&#160;<span class="preprocessor">#endif // defined(MIN_BOUND)</span></div><div class="line"><a name="l01730"></a><span class="lineno"> 1730</span>&#160;<span class="preprocessor">#if defined(MAX_BOUND)</span></div><div class="line"><a name="l01731"></a><span class="lineno"> 1731</span>&#160; res = min(res, (uchar4)MAX_BOUND);</div><div class="line"><a name="l01732"></a><span class="lineno"> 1732</span>&#160;<span class="preprocessor">#endif // defined(MAX_BOUND)</span></div><div class="line"><a name="l01733"></a><span class="lineno"> 1733</span>&#160;</div><div class="line"><a name="l01734"></a><span class="lineno"> 1734</span>&#160; <span class="comment">// Store the result</span></div><div class="line"><a name="l01735"></a><span class="lineno"> 1735</span>&#160; vstore4(res, 0, dst_addr);</div><div class="line"><a name="l01736"></a><span class="lineno"> 1736</span>&#160;}</div><div class="line"><a name="l01737"></a><span class="lineno"> 1737</span>&#160;<span class="preprocessor">#endif // defined(RESULT_OFFSET) &amp;&amp; defined(RESULT_MULT_INT) &amp;&amp; defined(RESULT_SHIFT)</span></div><div class="line"><a name="l01738"></a><span class="lineno"> 1738</span>&#160;</div><div class="line"><a name="l01739"></a><span class="lineno"> 1739</span>&#160;<span class="preprocessor">#if defined(RESULT_OFFSET_AFTER_SHIFT) &amp;&amp; defined(RESULT_FIXEDPOINT_MULTIPLIER) &amp;&amp; defined(RESULT_SHIFT)</span></div><div class="line"><a name="l01740"></a><span class="lineno"> 1740</span>&#160;</div><div class="line"><a name="l01779"></a><span class="lineno"> 1779</span>&#160;__kernel <span class="keywordtype">void</span> gemmlowp_output_stage_quantize_down_fixedpoint(<a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a6b83038822d1ae7ab619b684ed3b7fc0">TENSOR3D_DECLARATION</a>(<a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#a989ab3e96426615bb98e04e0235088ca">src</a>),</div><div class="line"><a name="l01780"></a><span class="lineno"> 1780</span>&#160;#<span class="keywordflow">if</span> defined(ADD_BIAS)</div><div class="line"><a name="l01781"></a><span class="lineno"> 1781</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a40a6eb9f2a7712f08d6bb8ff6c9e6ca7">VECTOR_DECLARATION</a>(biases),</div><div class="line"><a name="l01782"></a><span class="lineno"> 1782</span>&#160;#endif <span class="comment">// defined(ADD_BIAS)</span></div><div class="line"><a name="l01783"></a><span class="lineno"> 1783</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_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="l01784"></a><span class="lineno"> 1784</span>&#160;{</div><div class="line"><a name="l01785"></a><span class="lineno"> 1785</span>&#160; <span class="comment">// Compute source and destination addresses</span></div><div class="line"><a name="l01786"></a><span class="lineno"> 1786</span>&#160; <span class="keywordtype">int</span> x = get_global_id(0) * 4;</div><div class="line"><a name="l01787"></a><span class="lineno"> 1787</span>&#160; <span class="keywordtype">int</span> y = get_global_id(1);</div><div class="line"><a name="l01788"></a><span class="lineno"> 1788</span>&#160; <span class="keywordtype">int</span> z = get_global_id(2);</div><div class="line"><a name="l01789"></a><span class="lineno"> 1789</span>&#160;</div><div class="line"><a name="l01790"></a><span class="lineno"> 1790</span>&#160; __global uchar *src_addr = src_ptr + src_offset_first_element_in_bytes + x * <span class="keyword">sizeof</span>(int) + y * src_stride_y + z * src_stride_z;</div><div class="line"><a name="l01791"></a><span class="lineno"> 1791</span>&#160;</div><div class="line"><a name="l01792"></a><span class="lineno"> 1792</span>&#160; __global uchar *dst_addr = dst_ptr + dst_offset_first_element_in_bytes + x + y * dst_stride_y + z * dst_stride_z;</div><div class="line"><a name="l01793"></a><span class="lineno"> 1793</span>&#160;</div><div class="line"><a name="l01794"></a><span class="lineno"> 1794</span>&#160; int4 input_values = vload4(0, (__global <span class="keywordtype">int</span> *)src_addr);</div><div class="line"><a name="l01795"></a><span class="lineno"> 1795</span>&#160;</div><div class="line"><a name="l01796"></a><span class="lineno"> 1796</span>&#160;<span class="preprocessor">#if defined(ADD_BIAS)</span></div><div class="line"><a name="l01797"></a><span class="lineno"> 1797</span>&#160; <span class="comment">// Add bias</span></div><div class="line"><a name="l01798"></a><span class="lineno"> 1798</span>&#160; __global uchar *bias_addr = biases_ptr + biases_offset_first_element_in_bytes + x * <span class="keyword">sizeof</span>(int);</div><div class="line"><a name="l01799"></a><span class="lineno"> 1799</span>&#160;</div><div class="line"><a name="l01800"></a><span class="lineno"> 1800</span>&#160; int4 biases_values = vload4(0, (__global <span class="keywordtype">int</span> *)bias_addr);</div><div class="line"><a name="l01801"></a><span class="lineno"> 1801</span>&#160; input_values += (int4)biases_values;</div><div class="line"><a name="l01802"></a><span class="lineno"> 1802</span>&#160;<span class="preprocessor">#endif // defined(ADD_BIAS)</span></div><div class="line"><a name="l01803"></a><span class="lineno"> 1803</span>&#160;</div><div class="line"><a name="l01804"></a><span class="lineno"> 1804</span>&#160; <span class="comment">// Multiply by result_mult_int and shift</span></div><div class="line"><a name="l01805"></a><span class="lineno"> 1805</span>&#160; input_values = <a class="code" href="helpers__asymm_8h.xhtml#a86de0ffca367bfcf27a8ae5dd1fdef2d">ASYMM_MULT_BY_QUANT_MULTIPLIER_LESS_THAN_ONE</a>(input_values, RESULT_FIXEDPOINT_MULTIPLIER, RESULT_SHIFT, 4);</div><div class="line"><a name="l01806"></a><span class="lineno"> 1806</span>&#160;</div><div class="line"><a name="l01807"></a><span class="lineno"> 1807</span>&#160; <span class="comment">// Add the offset terms to GEMM&#39;s result</span></div><div class="line"><a name="l01808"></a><span class="lineno"> 1808</span>&#160; input_values += (int4)RESULT_OFFSET_AFTER_SHIFT;</div><div class="line"><a name="l01809"></a><span class="lineno"> 1809</span>&#160;</div><div class="line"><a name="l01810"></a><span class="lineno"> 1810</span>&#160; uchar4 res = convert_uchar4_sat(input_values);</div><div class="line"><a name="l01811"></a><span class="lineno"> 1811</span>&#160;</div><div class="line"><a name="l01812"></a><span class="lineno"> 1812</span>&#160;<span class="preprocessor">#if defined(MIN_BOUND)</span></div><div class="line"><a name="l01813"></a><span class="lineno"> 1813</span>&#160; res = max(res, (uchar4)MIN_BOUND);</div><div class="line"><a name="l01814"></a><span class="lineno"> 1814</span>&#160;<span class="preprocessor">#endif // defined(MIN_BOUND)</span></div><div class="line"><a name="l01815"></a><span class="lineno"> 1815</span>&#160;<span class="preprocessor">#if defined(MAX_BOUND)</span></div><div class="line"><a name="l01816"></a><span class="lineno"> 1816</span>&#160; res = min(res, (uchar4)MAX_BOUND);</div><div class="line"><a name="l01817"></a><span class="lineno"> 1817</span>&#160;<span class="preprocessor">#endif // defined(MAX_BOUND)</span></div><div class="line"><a name="l01818"></a><span class="lineno"> 1818</span>&#160;</div><div class="line"><a name="l01819"></a><span class="lineno"> 1819</span>&#160; <span class="comment">// Store the result</span></div><div class="line"><a name="l01820"></a><span class="lineno"> 1820</span>&#160; vstore4(res, 0, dst_addr);</div><div class="line"><a name="l01821"></a><span class="lineno"> 1821</span>&#160;}</div><div class="line"><a name="l01822"></a><span class="lineno"> 1822</span>&#160;<span class="preprocessor">#endif // defined(RESULT_OFFSET_AFTER_SHIFT) &amp;&amp; defined(RESULT_FIXEDPOINT_MULTIPLIER) &amp;&amp; defined(RESULT_SHIFT)</span></div><div class="line"><a name="l01823"></a><span class="lineno"> 1823</span>&#160;</div><div class="line"><a name="l01824"></a><span class="lineno"> 1824</span>&#160;<span class="preprocessor">#if defined(RESULT_FIXEDPOINT_MULTIPLIER) &amp;&amp; defined(RESULT_SHIFT)</span></div><div class="line"><a name="l01825"></a><span class="lineno"> 1825</span>&#160;</div><div class="line"><a name="l01865"></a><span class="lineno"> 1865</span>&#160;__kernel <span class="keywordtype">void</span> gemmlowp_output_stage_quantize_down_fixedpoint_qsymm16(<a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a6b83038822d1ae7ab619b684ed3b7fc0">TENSOR3D_DECLARATION</a>(<a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#a989ab3e96426615bb98e04e0235088ca">src</a>),</div><div class="line"><a name="l01866"></a><span class="lineno"> 1866</span>&#160;#<span class="keywordflow">if</span> defined(ADD_BIAS)</div><div class="line"><a name="l01867"></a><span class="lineno"> 1867</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a40a6eb9f2a7712f08d6bb8ff6c9e6ca7">VECTOR_DECLARATION</a>(biases),</div><div class="line"><a name="l01868"></a><span class="lineno"> 1868</span>&#160;#endif <span class="comment">// defined(ADD_BIAS)</span></div><div class="line"><a name="l01869"></a><span class="lineno"> 1869</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_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="l01870"></a><span class="lineno"> 1870</span>&#160;{</div><div class="line"><a name="l01871"></a><span class="lineno"> 1871</span>&#160; <span class="comment">// Compute source and destination addresses</span></div><div class="line"><a name="l01872"></a><span class="lineno"> 1872</span>&#160; <span class="keywordtype">int</span> x = get_global_id(0) * 4;</div><div class="line"><a name="l01873"></a><span class="lineno"> 1873</span>&#160; <span class="keywordtype">int</span> y = get_global_id(1);</div><div class="line"><a name="l01874"></a><span class="lineno"> 1874</span>&#160; <span class="keywordtype">int</span> z = get_global_id(2);</div><div class="line"><a name="l01875"></a><span class="lineno"> 1875</span>&#160;</div><div class="line"><a name="l01876"></a><span class="lineno"> 1876</span>&#160; __global uchar *src_addr = src_ptr + src_offset_first_element_in_bytes + x * <span class="keyword">sizeof</span>(int) + y * src_stride_y + z * src_stride_z;</div><div class="line"><a name="l01877"></a><span class="lineno"> 1877</span>&#160;</div><div class="line"><a name="l01878"></a><span class="lineno"> 1878</span>&#160; __global uchar *dst_addr = dst_ptr + dst_offset_first_element_in_bytes + x * 2 + y * dst_stride_y + z * dst_stride_z;</div><div class="line"><a name="l01879"></a><span class="lineno"> 1879</span>&#160;</div><div class="line"><a name="l01880"></a><span class="lineno"> 1880</span>&#160; int4 input_values = vload4(0, (__global <span class="keywordtype">int</span> *)src_addr);</div><div class="line"><a name="l01881"></a><span class="lineno"> 1881</span>&#160;</div><div class="line"><a name="l01882"></a><span class="lineno"> 1882</span>&#160;<span class="preprocessor">#if defined(ADD_BIAS)</span></div><div class="line"><a name="l01883"></a><span class="lineno"> 1883</span>&#160; <span class="comment">// Add bias</span></div><div class="line"><a name="l01884"></a><span class="lineno"> 1884</span>&#160; __global uchar *bias_addr = biases_ptr + biases_offset_first_element_in_bytes + x * <span class="keyword">sizeof</span>(int);</div><div class="line"><a name="l01885"></a><span class="lineno"> 1885</span>&#160;</div><div class="line"><a name="l01886"></a><span class="lineno"> 1886</span>&#160; int4 biases_values = vload4(0, (__global <span class="keywordtype">int</span> *)bias_addr);</div><div class="line"><a name="l01887"></a><span class="lineno"> 1887</span>&#160; input_values += (int4)biases_values;</div><div class="line"><a name="l01888"></a><span class="lineno"> 1888</span>&#160;<span class="preprocessor">#endif // defined(ADD_BIAS)</span></div><div class="line"><a name="l01889"></a><span class="lineno"> 1889</span>&#160;</div><div class="line"><a name="l01890"></a><span class="lineno"> 1890</span>&#160; <span class="comment">// Multiply by result_mult_int and shift</span></div><div class="line"><a name="l01891"></a><span class="lineno"> 1891</span>&#160; input_values = <a class="code" href="helpers__asymm_8h.xhtml#a86de0ffca367bfcf27a8ae5dd1fdef2d">ASYMM_MULT_BY_QUANT_MULTIPLIER_LESS_THAN_ONE</a>(input_values, RESULT_FIXEDPOINT_MULTIPLIER, RESULT_SHIFT, 4);</div><div class="line"><a name="l01892"></a><span class="lineno"> 1892</span>&#160;</div><div class="line"><a name="l01893"></a><span class="lineno"> 1893</span>&#160; short4 res = convert_short4_sat(input_values);</div><div class="line"><a name="l01894"></a><span class="lineno"> 1894</span>&#160;</div><div class="line"><a name="l01895"></a><span class="lineno"> 1895</span>&#160;<span class="preprocessor">#if defined(MIN_BOUND)</span></div><div class="line"><a name="l01896"></a><span class="lineno"> 1896</span>&#160; res = max(res, (short4)MIN_BOUND);</div><div class="line"><a name="l01897"></a><span class="lineno"> 1897</span>&#160;<span class="preprocessor">#endif // defined(MIN_BOUND)</span></div><div class="line"><a name="l01898"></a><span class="lineno"> 1898</span>&#160;<span class="preprocessor">#if defined(MAX_BOUND)</span></div><div class="line"><a name="l01899"></a><span class="lineno"> 1899</span>&#160; res = min(res, (short4)MAX_BOUND);</div><div class="line"><a name="l01900"></a><span class="lineno"> 1900</span>&#160;<span class="preprocessor">#endif // defined(MAX_BOUND)</span></div><div class="line"><a name="l01901"></a><span class="lineno"> 1901</span>&#160;</div><div class="line"><a name="l01902"></a><span class="lineno"> 1902</span>&#160; <span class="comment">// Store the result</span></div><div class="line"><a name="l01903"></a><span class="lineno"> 1903</span>&#160; vstore4(res, 0, (__global <span class="keywordtype">short</span> *)dst_addr);</div><div class="line"><a name="l01904"></a><span class="lineno"> 1904</span>&#160;}</div><div class="line"><a name="l01905"></a><span class="lineno"> 1905</span>&#160;<span class="preprocessor">#endif // defined(RESULT_FIXEDPOINT_MULTIPLIER) &amp;&amp; defined(RESULT_SHIFT)</span></div><div class="line"><a name="l01906"></a><span class="lineno"> 1906</span>&#160;</div><div class="line"><a name="l01907"></a><span class="lineno"> 1907</span>&#160;<span class="preprocessor">#if defined(REAL_MULTIPLIER) &amp;&amp; defined(OUTPUT_OFFSET)</span></div><div class="line"><a name="l01908"></a><span class="lineno"> 1908</span>&#160;</div><div class="line"><a name="l01949"></a><span class="lineno"> 1949</span>&#160;__kernel <span class="keywordtype">void</span> gemmlowp_output_stage_quantize_down_float(<a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a6b83038822d1ae7ab619b684ed3b7fc0">TENSOR3D_DECLARATION</a>(<a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#a989ab3e96426615bb98e04e0235088ca">src</a>),</div><div class="line"><a name="l01950"></a><span class="lineno"> 1950</span>&#160;#<span class="keywordflow">if</span> defined(ADD_BIAS)</div><div class="line"><a name="l01951"></a><span class="lineno"> 1951</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a40a6eb9f2a7712f08d6bb8ff6c9e6ca7">VECTOR_DECLARATION</a>(biases),</div><div class="line"><a name="l01952"></a><span class="lineno"> 1952</span>&#160;#endif <span class="comment">// defined(ADD_BIAS)</span></div><div class="line"><a name="l01953"></a><span class="lineno"> 1953</span>&#160;#<span class="keywordflow">if</span> defined(DST_HEIGHT)</div><div class="line"><a name="l01954"></a><span class="lineno"> 1954</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a481bdc6d61b3df9dcdbdb244f0f97790">TENSOR4D_DECLARATION</a>(<a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#adbf67dcee294e673cf796f1ed8aeb6a4">dst</a>))</div><div class="line"><a name="l01955"></a><span class="lineno"> 1955</span>&#160;<span class="preprocessor">#else // defined(DST_HEIGHT)</span></div><div class="line"><a name="l01956"></a><span class="lineno"> 1956</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_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="l01957"></a><span class="lineno"> 1957</span>&#160;#endif <span class="comment">// defined(DST_HEIGHT)</span></div><div class="line"><a name="l01958"></a><span class="lineno"> 1958</span>&#160;{</div><div class="line"><a name="l01959"></a><span class="lineno"> 1959</span>&#160; <span class="comment">// Compute source and destination addresses</span></div><div class="line"><a name="l01960"></a><span class="lineno"> 1960</span>&#160; <span class="keywordtype">int</span> x = get_global_id(0) * 4;</div><div class="line"><a name="l01961"></a><span class="lineno"> 1961</span>&#160; <span class="keywordtype">int</span> y = get_global_id(1);</div><div class="line"><a name="l01962"></a><span class="lineno"> 1962</span>&#160; <span class="keywordtype">int</span> z = get_global_id(2);</div><div class="line"><a name="l01963"></a><span class="lineno"> 1963</span>&#160;</div><div class="line"><a name="l01964"></a><span class="lineno"> 1964</span>&#160; __global uchar *src_addr = src_ptr + src_offset_first_element_in_bytes + x * <span class="keyword">sizeof</span>(int) + y * src_stride_y + z * src_stride_z;</div><div class="line"><a name="l01965"></a><span class="lineno"> 1965</span>&#160;</div><div class="line"><a name="l01966"></a><span class="lineno"> 1966</span>&#160; __global uchar *dst_addr = dst_ptr + dst_offset_first_element_in_bytes + x + y * dst_stride_y + z * dst_stride_z;</div><div class="line"><a name="l01967"></a><span class="lineno"> 1967</span>&#160;</div><div class="line"><a name="l01968"></a><span class="lineno"> 1968</span>&#160; int4 input_values = vload4(0, (__global <span class="keywordtype">int</span> *)src_addr);</div><div class="line"><a name="l01969"></a><span class="lineno"> 1969</span>&#160;</div><div class="line"><a name="l01970"></a><span class="lineno"> 1970</span>&#160;<span class="preprocessor">#if defined(ADD_BIAS)</span></div><div class="line"><a name="l01971"></a><span class="lineno"> 1971</span>&#160; <span class="comment">// Add bias</span></div><div class="line"><a name="l01972"></a><span class="lineno"> 1972</span>&#160; __global uchar *bias_addr = biases_ptr + biases_offset_first_element_in_bytes + x * <span class="keyword">sizeof</span>(int);</div><div class="line"><a name="l01973"></a><span class="lineno"> 1973</span>&#160;</div><div class="line"><a name="l01974"></a><span class="lineno"> 1974</span>&#160; int4 biases_values = vload4(0, (__global <span class="keywordtype">int</span> *)bias_addr);</div><div class="line"><a name="l01975"></a><span class="lineno"> 1975</span>&#160; input_values += (int4)biases_values;</div><div class="line"><a name="l01976"></a><span class="lineno"> 1976</span>&#160;<span class="preprocessor">#endif // defined(ADD_BIAS)</span></div><div class="line"><a name="l01977"></a><span class="lineno"> 1977</span>&#160;</div><div class="line"><a name="l01978"></a><span class="lineno"> 1978</span>&#160; <span class="comment">// Convert to float</span></div><div class="line"><a name="l01979"></a><span class="lineno"> 1979</span>&#160; float16 input_values_f = convert_float4(input_values);</div><div class="line"><a name="l01980"></a><span class="lineno"> 1980</span>&#160; input_values_f = <a class="code" href="namespacearm__compute.xhtml#aaae2b6b1c3f4404121346a4c27b22647">round</a>(input_values_f * (<span class="keywordtype">float</span>)REAL_MULTIPLIER + (<span class="keywordtype">float</span>)OUTPUT_OFFSET);</div><div class="line"><a name="l01981"></a><span class="lineno"> 1981</span>&#160;</div><div class="line"><a name="l01982"></a><span class="lineno"> 1982</span>&#160; uchar4 res = convert_uchar4_sat(input_values_f);</div><div class="line"><a name="l01983"></a><span class="lineno"> 1983</span>&#160;</div><div class="line"><a name="l01984"></a><span class="lineno"> 1984</span>&#160;<span class="preprocessor">#if defined(MIN_BOUND)</span></div><div class="line"><a name="l01985"></a><span class="lineno"> 1985</span>&#160; res = max(res, (uchar4)MIN_BOUND);</div><div class="line"><a name="l01986"></a><span class="lineno"> 1986</span>&#160;<span class="preprocessor">#endif // defined(MIN_BOUND)</span></div><div class="line"><a name="l01987"></a><span class="lineno"> 1987</span>&#160;<span class="preprocessor">#if defined(MAX_BOUND)</span></div><div class="line"><a name="l01988"></a><span class="lineno"> 1988</span>&#160; res = min(res, (uchar4)MAX_BOUND);</div><div class="line"><a name="l01989"></a><span class="lineno"> 1989</span>&#160;<span class="preprocessor">#endif // defined(MAX_BOUND)</span></div><div class="line"><a name="l01990"></a><span class="lineno"> 1990</span>&#160;</div><div class="line"><a name="l01991"></a><span class="lineno"> 1991</span>&#160; <span class="comment">// Store the result</span></div><div class="line"><a name="l01992"></a><span class="lineno"> 1992</span>&#160; vstore4(res, 0, dst_addr);</div><div class="line"><a name="l01993"></a><span class="lineno"> 1993</span>&#160;}</div><div class="line"><a name="l01994"></a><span class="lineno"> 1994</span>&#160;<span class="preprocessor">#endif // defined(REAL_MULTIPLIER) &amp;&amp; defined(OUTPUT_OFFSET)</span></div><div class="ttc" id="repeat_8h_xhtml_a4a61a8df72f9bc1b1dcc4cc4ec6ac3b9"><div class="ttname"><a href="repeat_8h.xhtml#a4a61a8df72f9bc1b1dcc4cc4ec6ac3b9">REPEAT_VAR_INIT_TO_CONST</a></div><div class="ttdeci">#define REPEAT_VAR_INIT_TO_CONST(N, TYPE, VAR, VAL)</div><div class="ttdef"><b>Definition:</b> <a href="repeat_8h_source.xhtml#l00081">repeat.h:81</a></div></div>
<div class="ttc" id="src_2core_2_c_l_2cl__kernels_2_helpers_8h_xhtml_aa8d95ba04fc73845abc6045952cae5be"><div class="ttname"><a href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#aa8d95ba04fc73845abc6045952cae5be">CONVERT</a></div><div class="ttdeci">#define CONVERT(x, type)</div><div class="ttdef"><b>Definition:</b> <a href="src_2core_2_c_l_2cl__kernels_2_helpers_8h_source.xhtml#l00079">helpers.h:79</a></div></div>
<div class="ttc" id="repeat_8h_xhtml"><div class="ttname"><a href="repeat_8h.xhtml">repeat.h</a></div></div>
<div class="ttc" id="src_2core_2_c_l_2cl__kernels_2_helpers_8h_xhtml_aebe814363556c244be043b13e7969197"><div class="ttname"><a href="src_2core_2_c_l_2cl__kernels_2_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="src_2core_2_c_l_2cl__kernels_2_helpers_8h_source.xhtml#l00129">helpers.h:129</a></div></div>
<div class="ttc" id="namespacearm__compute_1_1test_1_1validation_xhtml_aa76b4a6e74940dabc5b7fc6b2dab3545"><div class="ttname"><a href="namespacearm__compute_1_1test_1_1validation.xhtml#aa76b4a6e74940dabc5b7fc6b2dab3545">arm_compute::test::validation::b</a></div><div class="ttdeci">SimpleTensor&lt; float &gt; b</div><div class="ttdef"><b>Definition:</b> <a href="_c_p_p_2_d_f_t_8cpp_source.xhtml#l00157">DFT.cpp:157</a></div></div>
<div class="ttc" id="gemm__helpers_8h_xhtml"><div class="ttname"><a href="gemm__helpers_8h.xhtml">gemm_helpers.h</a></div></div>
<div class="ttc" id="src_2core_2_c_l_2cl__kernels_2_helpers_8h_xhtml_a22f42fcf2077d951271df83b55c1a71a"><div class="ttname"><a href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a22f42fcf2077d951271df83b55c1a71a">IMAGE_DECLARATION</a></div><div class="ttdeci">#define IMAGE_DECLARATION(name)</div><div class="ttdef"><b>Definition:</b> <a href="src_2core_2_c_l_2cl__kernels_2_helpers_8h_source.xhtml#l00093">helpers.h:93</a></div></div>
<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="src_2core_2_c_l_2cl__kernels_2_helpers_8h_source.xhtml#l00176">helpers.h:176</a></div></div>
<div class="ttc" id="gemm__helpers_8h_xhtml_af83fba9e0a00fc38a71258f0052b9c24"><div class="ttname"><a href="gemm__helpers_8h.xhtml#af83fba9e0a00fc38a71258f0052b9c24">TRANSPOSE_K0XN0</a></div><div class="ttdeci">#define TRANSPOSE_K0XN0(K0, N0, BASENAME, B)</div><div class="ttdef"><b>Definition:</b> <a href="gemm__helpers_8h_source.xhtml#l00480">gemm_helpers.h:480</a></div></div>
<div class="ttc" id="gemmlowp_8cl_xhtml_af110f6648cbcc21216a66fb3342632b0"><div class="ttname"><a href="gemmlowp_8cl.xhtml#af110f6648cbcc21216a66fb3342632b0">ARM_MM_K0XN0XM0</a></div><div class="ttdeci">#define ARM_MM_K0XN0XM0(m0, n0, k0, a, b, c)</div><div class="ttdef"><b>Definition:</b> <a href="gemmlowp_8cl_source.xhtml#l00190">gemmlowp.cl:190</a></div></div>
<div class="ttc" id="src_2core_2_c_l_2cl__kernels_2_helpers_8h_xhtml_a40a6eb9f2a7712f08d6bb8ff6c9e6ca7"><div class="ttname"><a href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a40a6eb9f2a7712f08d6bb8ff6c9e6ca7">VECTOR_DECLARATION</a></div><div class="ttdeci">#define VECTOR_DECLARATION(name)</div><div class="ttdef"><b>Definition:</b> <a href="src_2core_2_c_l_2cl__kernels_2_helpers_8h_source.xhtml#l00087">helpers.h:87</a></div></div>
<div class="ttc" id="gemm__helpers_8h_xhtml_a9fe7ded27e552f48fa285d34fb00427d"><div class="ttname"><a href="gemm__helpers_8h.xhtml#a9fe7ded27e552f48fa285d34fb00427d">LOAD_BLOCK</a></div><div class="ttdeci">#define LOAD_BLOCK(M0, N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Z)</div><div class="ttdoc">Load Blocks of M0 consecutive rows and N0 consecutive columns when using Z offset as well Supported c...</div><div class="ttdef"><b>Definition:</b> <a href="gemm__helpers_8h_source.xhtml#l00113">gemm_helpers.h:113</a></div></div>
<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_absolute_difference_8cpp_source.xhtml#l00102">AbsoluteDifference.cpp:102</a></div></div>
<div class="ttc" id="gemm__helpers_8h_xhtml_ade58486d6b2d13d7941106a9d2ec86d3"><div class="ttname"><a href="gemm__helpers_8h.xhtml#ade58486d6b2d13d7941106a9d2ec86d3">CALCULATE_Z_OFFSET</a></div><div class="ttdeci">#define CALCULATE_Z_OFFSET(M0, DATA_TYPE, Z, Y, HEIGHT_GEMM3D, DEPTH_GEMM3D, CROSS_PLANE_PAD, STRIDE_Y)</div><div class="ttdoc">The Z offsets are expected to have consecutive names, For e.g.</div><div class="ttdef"><b>Definition:</b> <a href="gemm__helpers_8h_source.xhtml#l00181">gemm_helpers.h:181</a></div></div>
<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="src_2core_2_c_l_2cl__kernels_2_helpers_8h_source.xhtml#l00167">helpers.h:167</a></div></div>
<div class="ttc" id="namespacearm__compute_xhtml_aaae2b6b1c3f4404121346a4c27b22647"><div class="ttname"><a href="namespacearm__compute.xhtml#aaae2b6b1c3f4404121346a4c27b22647">arm_compute::round</a></div><div class="ttdeci">int round(float x, RoundingPolicy rounding_policy)</div><div class="ttdoc">Return a rounded value of x.</div><div class="ttdef"><b>Definition:</b> <a href="_rounding_8cpp_source.xhtml#l00035">Rounding.cpp:35</a></div></div>
<div class="ttc" id="src_2core_2_c_l_2cl__kernels_2_helpers_8h_xhtml_a31c8c760f08fb1a331b16b7c204321dc"><div class="ttname"><a href="src_2core_2_c_l_2cl__kernels_2_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="src_2core_2_c_l_2cl__kernels_2_helpers_8h_source.xhtml#l00144">helpers.h:144</a></div></div>
<div class="ttc" id="helpers__asymm_8h_xhtml"><div class="ttname"><a href="helpers__asymm_8h.xhtml">helpers_asymm.h</a></div></div>
<div class="ttc" id="src_2core_2_c_l_2cl__kernels_2_helpers_8h_xhtml_acb282042d1edeeaa3cc979a206f78b54"><div class="ttname"><a href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#acb282042d1edeeaa3cc979a206f78b54">VSTORE</a></div><div class="ttdeci">#define VSTORE(size)</div><div class="ttdef"><b>Definition:</b> <a href="src_2core_2_c_l_2cl__kernels_2_helpers_8h_source.xhtml#l00056">helpers.h:56</a></div></div>
<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#l00297">helpers_asymm.h:297</a></div></div>
<div class="ttc" id="src_2core_2_c_l_2cl__kernels_2_helpers_8h_xhtml_a481bdc6d61b3df9dcdbdb244f0f97790"><div class="ttname"><a href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a481bdc6d61b3df9dcdbdb244f0f97790">TENSOR4D_DECLARATION</a></div><div class="ttdeci">#define TENSOR4D_DECLARATION(name)</div><div class="ttdef"><b>Definition:</b> <a href="src_2core_2_c_l_2cl__kernels_2_helpers_8h_source.xhtml#l00111">helpers.h:111</a></div></div>
<div class="ttc" id="gemm__helpers_8h_xhtml_af5c0a8dfefbd611d38bb91212684ef7d"><div class="ttname"><a href="gemm__helpers_8h.xhtml#af5c0a8dfefbd611d38bb91212684ef7d">CONVERT_STORE_BLOCK</a></div><div class="ttdeci">#define CONVERT_STORE_BLOCK(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)</div><div class="ttdoc">Convert and store a block of size M0 (rows) x NO (columns).</div><div class="ttdef"><b>Definition:</b> <a href="gemm__helpers_8h_source.xhtml#l00361">gemm_helpers.h:361</a></div></div>
<div class="ttc" id="src_2core_2_c_l_2cl__kernels_2_helpers_8h_xhtml_a287e2fc366c312b468382c95bb90f91f"><div class="ttname"><a href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a287e2fc366c312b468382c95bb90f91f">VLOAD</a></div><div class="ttdeci">#define VLOAD(size)</div><div class="ttdef"><b>Definition:</b> <a href="src_2core_2_c_l_2cl__kernels_2_helpers_8h_source.xhtml#l00053">helpers.h:53</a></div></div>
<div class="ttc" id="src_2core_2_c_l_2cl__kernels_2_helpers_8h_xhtml_a6b83038822d1ae7ab619b684ed3b7fc0"><div class="ttname"><a href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a6b83038822d1ae7ab619b684ed3b7fc0">TENSOR3D_DECLARATION</a></div><div class="ttdeci">#define TENSOR3D_DECLARATION(name)</div><div class="ttdef"><b>Definition:</b> <a href="src_2core_2_c_l_2cl__kernels_2_helpers_8h_source.xhtml#l00101">helpers.h:101</a></div></div>
<div class="ttc" id="namespacearm__compute_1_1test_1_1validation_xhtml_a989ab3e96426615bb98e04e0235088ca"><div class="ttname"><a href="namespacearm__compute_1_1test_1_1validation.xhtml#a989ab3e96426615bb98e04e0235088ca">arm_compute::test::validation::src</a></div><div class="ttdeci">cast configure &amp; src</div><div class="ttdef"><b>Definition:</b> <a href="_c_l_2_cast_8cpp_source.xhtml#l00169">Cast.cpp:169</a></div></div>
<div class="ttc" id="src_2core_2_c_l_2cl__kernels_2_helpers_8h_xhtml_a36f754c05b6fddf6df0d8d0a74f8159f"><div class="ttname"><a href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a></div><div class="ttdeci">#define VEC_DATA_TYPE(type, size)</div><div class="ttdef"><b>Definition:</b> <a href="src_2core_2_c_l_2cl__kernels_2_helpers_8h_source.xhtml#l00073">helpers.h:73</a></div></div>
</div><!-- fragment --></div><!-- contents -->
</div><!-- doc-content -->
<!-- start footer part -->
<div id="nav-path" class="navpath"><!-- id is needed for treeview function! -->
<ul>
<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="gemmlowp_8cl.xhtml">gemmlowp.cl</a></li>
<li class="footer">Generated on Mon Sep 2 2019 11:47:21 for Compute Library by
<a href="http://www.doxygen.org/index.html">
<img class="footer" src="doxygen.png" alt="doxygen"/></a> 1.8.15 </li>
</ul>
</div>
</body>
</html>