Gitiles
Code Review Sign In
gerrit-public.fairphone.software / platform / external / ComputeLibrary / 514be65ad8d3340f53fd9591035352ed285811ba / . / documentation / gemm_8cl_source.xhtml
blob: 1d8f876cdf047939418bb1967899c0d63a8509f4 [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/gemm.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">19.02</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('gemm_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">gemm.cl</div> </div>
</div><!--header-->
<div class="contents">
<a href="gemm_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="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml">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="repeat_8h.xhtml">repeat.h</a>&quot;</span></div><div class="line"><a name="l00026"></a><span class="lineno"> 26</span>&#160;</div><div class="line"><a name="l00027"></a><span class="lineno"> 27</span>&#160;<span class="preprocessor">#if defined(M0) &amp;&amp; defined(K0) &amp;&amp; defined(V0) &amp;&amp; defined(DATA_TYPE) &amp;&amp; defined(SRC_WIDTH)</span></div><div class="line"><a name="l00028"></a><span class="lineno"> 28</span>&#160;<span class="preprocessor">#define INC2 (VEC_DATA_TYPE(uint, 2))(0, 1)</span></div><div class="line"><a name="l00029"></a><span class="lineno"> 29</span>&#160;<span class="preprocessor">#define INC3 (VEC_DATA_TYPE(uint, 3))(0, 1, 2)</span></div><div class="line"><a name="l00030"></a><span class="lineno"> 30</span>&#160;<span class="preprocessor">#define INC4 (VEC_DATA_TYPE(uint, 4))(0, 1, 2, 3)</span></div><div class="line"><a name="l00031"></a><span class="lineno"> 31</span>&#160;<span class="preprocessor">#define INC8 (VEC_DATA_TYPE(uint, 8))(0, 1, 2, 3, 4, 5, 6, 7)</span></div><div class="line"><a name="l00032"></a><span class="lineno"> 32</span>&#160;<span class="preprocessor">#define INC16 (VEC_DATA_TYPE(uint, 16))(0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15)</span></div><div class="line"><a name="l00033"></a><span class="lineno"> 33</span>&#160;<span class="preprocessor">#define CONCAT_INC(K0) INC##K0</span></div><div class="line"><a name="l00034"></a><span class="lineno"> 34</span>&#160;<span class="preprocessor">#define INC(K0) CONCAT_INC(K0)</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(SRC_WIDTH % K0)</span></div><div class="line"><a name="l00037"></a><span class="lineno"> 37</span>&#160;<span class="preprocessor">#define BOUNDARY_CONDITION_X(x, a) \</span></div><div class="line"><a name="l00038"></a><span class="lineno"> 38</span>&#160;<span class="preprocessor"> ({ \</span></div><div class="line"><a name="l00039"></a><span class="lineno"> 39</span>&#160;<span class="preprocessor"> a = select(0, a, CONVERT(((x * (VEC_DATA_TYPE(uint, K0))K0 + INC(K0)) &lt; (VEC_DATA_TYPE(uint, K0))SRC_WIDTH), VEC_DATA_TYPE(DATA_TYPE, K0))); \</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">#else // (SRC_WIDTH % K0)</span></div><div class="line"><a name="l00042"></a><span class="lineno"> 42</span>&#160;<span class="preprocessor">#define BOUNDARY_CONDITION_X(x, a) \</span></div><div class="line"><a name="l00043"></a><span class="lineno"> 43</span>&#160;<span class="preprocessor"> ({})</span></div><div class="line"><a name="l00044"></a><span class="lineno"> 44</span>&#160;<span class="preprocessor">#endif // (SRC_WIDTH % K0)</span></div><div class="line"><a name="l00045"></a><span class="lineno"> 45</span>&#160;</div><div class="line"><a name="l00082"></a><span class="lineno"> 82</span>&#160;__kernel <span class="keywordtype">void</span> gemm_reshape_lhs_matrix_nt(<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#ae9d2dc29c2789c253406f9b304cc75a8">src</a>),</div><div class="line"><a name="l00083"></a><span class="lineno"> 83</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="l00084"></a><span class="lineno"> 84</span>&#160;#<span class="keywordflow">if</span> defined(REINTERPRET_INPUT_AS_3D)</div><div class="line"><a name="l00085"></a><span class="lineno"> 85</span>&#160; ,</div><div class="line"><a name="l00086"></a><span class="lineno"> 86</span>&#160; uint cross_plane_pad</div><div class="line"><a name="l00087"></a><span class="lineno"> 87</span>&#160;#endif <span class="comment">// REINTERPRET_INPUT_AS_3D</span></div><div class="line"><a name="l00088"></a><span class="lineno"> 88</span>&#160; )</div><div class="line"><a name="l00089"></a><span class="lineno"> 89</span>&#160;{</div><div class="line"><a name="l00090"></a><span class="lineno"> 90</span>&#160; <span class="comment">// Block size</span></div><div class="line"><a name="l00091"></a><span class="lineno"> 91</span>&#160;<span class="preprocessor">#define BLOCK_SIZE ((M0) * (K0))</span></div><div class="line"><a name="l00092"></a><span class="lineno"> 92</span>&#160;</div><div class="line"><a name="l00093"></a><span class="lineno"> 93</span>&#160; <span class="comment">// Output offset X</span></div><div class="line"><a name="l00094"></a><span class="lineno"> 94</span>&#160;<span class="preprocessor">#if defined(INTERLEAVE)</span></div><div class="line"><a name="l00095"></a><span class="lineno"> 95</span>&#160;<span class="preprocessor">#define OUTPUT_OFFSET_X (K0)</span></div><div class="line"><a name="l00096"></a><span class="lineno"> 96</span>&#160;<span class="preprocessor">#else // defined(INTERLEAVE)</span></div><div class="line"><a name="l00097"></a><span class="lineno"> 97</span>&#160;<span class="preprocessor">#define OUTPUT_OFFSET_X (BLOCK_SIZE)</span></div><div class="line"><a name="l00098"></a><span class="lineno"> 98</span>&#160;<span class="preprocessor">#endif // defined(INTERLEAVE)</span></div><div class="line"><a name="l00099"></a><span class="lineno"> 99</span>&#160;</div><div class="line"><a name="l00100"></a><span class="lineno"> 100</span>&#160; <span class="comment">// Output step X</span></div><div class="line"><a name="l00101"></a><span class="lineno"> 101</span>&#160;<span class="preprocessor">#if defined(INTERLEAVE)</span></div><div class="line"><a name="l00102"></a><span class="lineno"> 102</span>&#160;<span class="preprocessor">#define OUTPUT_STEP_X (K0) * (V0)</span></div><div class="line"><a name="l00103"></a><span class="lineno"> 103</span>&#160;<span class="preprocessor">#else // Do not interleave</span></div><div class="line"><a name="l00104"></a><span class="lineno"> 104</span>&#160;<span class="preprocessor">#define OUTPUT_STEP_X (K0)</span></div><div class="line"><a name="l00105"></a><span class="lineno"> 105</span>&#160;<span class="preprocessor">#endif // defined(INTERLEAVE)</span></div><div class="line"><a name="l00106"></a><span class="lineno"> 106</span>&#160;</div><div class="line"><a name="l00107"></a><span class="lineno"> 107</span>&#160; <span class="comment">// Compute source and destination addresses</span></div><div class="line"><a name="l00108"></a><span class="lineno"> 108</span>&#160; uint x = get_global_id(0);</div><div class="line"><a name="l00109"></a><span class="lineno"> 109</span>&#160; uint y = get_global_id(1);</div><div class="line"><a name="l00110"></a><span class="lineno"> 110</span>&#160; uint z = get_global_id(2);</div><div class="line"><a name="l00111"></a><span class="lineno"> 111</span>&#160;</div><div class="line"><a name="l00112"></a><span class="lineno"> 112</span>&#160; <span class="comment">// ------------------ Compute input/output addresses ---------------------------</span></div><div class="line"><a name="l00113"></a><span class="lineno"> 113</span>&#160;</div><div class="line"><a name="l00114"></a><span class="lineno"> 114</span>&#160; <span class="comment">// Compute the input address</span></div><div class="line"><a name="l00115"></a><span class="lineno"> 115</span>&#160; __global uchar *input_ptr = src_ptr + src_offset_first_element_in_bytes + x * (uint)K0 * <span class="keyword">sizeof</span>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>) + y * (uint)M0 * src_stride_y;</div><div class="line"><a name="l00116"></a><span class="lineno"> 116</span>&#160;</div><div class="line"><a name="l00117"></a><span class="lineno"> 117</span>&#160; <span class="comment">// Compute the output address</span></div><div class="line"><a name="l00118"></a><span class="lineno"> 118</span>&#160; __global uchar *output_ptr = dst_ptr + dst_offset_first_element_in_bytes + (x * (uint)BLOCK_SIZE * (uint)V0 * <span class="keyword">sizeof</span>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>)) + ((y / (uint)V0) * (uint)dst_stride_y) + ((y % V0) *</div><div class="line"><a name="l00119"></a><span class="lineno"> 119</span>&#160; (uint)OUTPUT_OFFSET_X * <span class="keyword">sizeof</span>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>));</div><div class="line"><a name="l00120"></a><span class="lineno"> 120</span>&#160;</div><div class="line"><a name="l00121"></a><span class="lineno"> 121</span>&#160; <span class="comment">// Create variables: uint zin0=0, zin1=0, zin2=0...zin(M0-1)=0;</span></div><div class="line"><a name="l00122"></a><span class="lineno"> 122</span>&#160; <a class="code" href="repeat_8h.xhtml#a4a61a8df72f9bc1b1dcc4cc4ec6ac3b9">REPEAT_VAR_INIT_TO_CONST</a>(M0, uint, zin, 0);</div><div class="line"><a name="l00123"></a><span class="lineno"> 123</span>&#160;</div><div class="line"><a name="l00124"></a><span class="lineno"> 124</span>&#160;<span class="preprocessor">#if defined(REINTERPRET_INPUT_AS_3D)</span></div><div class="line"><a name="l00125"></a><span class="lineno"> 125</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="l00126"></a><span class="lineno"> 126</span>&#160; <span class="comment">// multiply src_stride_z by DEPTH_GEMM3D</span></div><div class="line"><a name="l00127"></a><span class="lineno"> 127</span>&#160;</div><div class="line"><a name="l00128"></a><span class="lineno"> 128</span>&#160; <span class="comment">// Note for the REINTERPRET_INPUT_AS_3D case</span></div><div class="line"><a name="l00129"></a><span class="lineno"> 129</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="l00130"></a><span class="lineno"> 130</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="l00131"></a><span class="lineno"> 131</span>&#160; <span class="comment">//</span></div><div class="line"><a name="l00132"></a><span class="lineno"> 132</span>&#160; <span class="comment">// | |</span></div><div class="line"><a name="l00133"></a><span class="lineno"> 133</span>&#160; <span class="comment">// | plane0 |</span></div><div class="line"><a name="l00134"></a><span class="lineno"> 134</span>&#160; <span class="comment">// | |</span></div><div class="line"><a name="l00135"></a><span class="lineno"> 135</span>&#160; <span class="comment">// |__________________|</span></div><div class="line"><a name="l00136"></a><span class="lineno"> 136</span>&#160; <span class="comment">// |******************|</span></div><div class="line"><a name="l00137"></a><span class="lineno"> 137</span>&#160; <span class="comment">// | cross_plane_pad |</span></div><div class="line"><a name="l00138"></a><span class="lineno"> 138</span>&#160; <span class="comment">// |******************|</span></div><div class="line"><a name="l00139"></a><span class="lineno"> 139</span>&#160; <span class="comment">// | |</span></div><div class="line"><a name="l00140"></a><span class="lineno"> 140</span>&#160; <span class="comment">// | plane1 |</span></div><div class="line"><a name="l00141"></a><span class="lineno"> 141</span>&#160; <span class="comment">// | |</span></div><div class="line"><a name="l00142"></a><span class="lineno"> 142</span>&#160; <span class="comment">// |__________________|</span></div><div class="line"><a name="l00143"></a><span class="lineno"> 143</span>&#160;</div><div class="line"><a name="l00144"></a><span class="lineno"> 144</span>&#160; input_ptr += z * (uint)src_stride_z * DEPTH_GEMM3D;</div><div class="line"><a name="l00145"></a><span class="lineno"> 145</span>&#160;</div><div class="line"><a name="l00146"></a><span class="lineno"> 146</span>&#160; <span class="comment">// The plane (zin) is calculated dividing M (y * M0) by HEIGHT_GEMM3D</span></div><div class="line"><a name="l00147"></a><span class="lineno"> 147</span>&#160; zin0 = (0 + (uint)(y * M0)) / (uint)HEIGHT_GEMM3D;</div><div class="line"><a name="l00148"></a><span class="lineno"> 148</span>&#160; zin0 = min((uint)(DEPTH_GEMM3D - 1), zin0);</div><div class="line"><a name="l00149"></a><span class="lineno"> 149</span>&#160; zin0 *= (cross_plane_pad * src_stride_y);</div><div class="line"><a name="l00150"></a><span class="lineno"> 150</span>&#160;<span class="preprocessor">#if M0 &gt; 1</span></div><div class="line"><a name="l00151"></a><span class="lineno"> 151</span>&#160; zin1 = (1 + (uint)(y * M0)) / (uint)HEIGHT_GEMM3D;</div><div class="line"><a name="l00152"></a><span class="lineno"> 152</span>&#160; zin1 = min((uint)(DEPTH_GEMM3D - 1), zin1);</div><div class="line"><a name="l00153"></a><span class="lineno"> 153</span>&#160; zin1 *= (cross_plane_pad * src_stride_y);</div><div class="line"><a name="l00154"></a><span class="lineno"> 154</span>&#160;<span class="preprocessor">#endif // M0 &gt; 1</span></div><div class="line"><a name="l00155"></a><span class="lineno"> 155</span>&#160;<span class="preprocessor">#if M0 &gt; 2</span></div><div class="line"><a name="l00156"></a><span class="lineno"> 156</span>&#160; zin2 = (2 + (uint)(y * M0)) / (uint)HEIGHT_GEMM3D;</div><div class="line"><a name="l00157"></a><span class="lineno"> 157</span>&#160; zin2 = min((uint)(DEPTH_GEMM3D - 1), zin2);</div><div class="line"><a name="l00158"></a><span class="lineno"> 158</span>&#160; zin2 *= (cross_plane_pad * src_stride_y);</div><div class="line"><a name="l00159"></a><span class="lineno"> 159</span>&#160;<span class="preprocessor">#endif // M0 &gt; 2</span></div><div class="line"><a name="l00160"></a><span class="lineno"> 160</span>&#160;<span class="preprocessor">#if M0 &gt; 3</span></div><div class="line"><a name="l00161"></a><span class="lineno"> 161</span>&#160; zin3 = (3 + (uint)(y * M0)) / (uint)HEIGHT_GEMM3D;</div><div class="line"><a name="l00162"></a><span class="lineno"> 162</span>&#160; zin3 = min((uint)(DEPTH_GEMM3D - 1), zin3);</div><div class="line"><a name="l00163"></a><span class="lineno"> 163</span>&#160; zin3 *= (cross_plane_pad * src_stride_y);</div><div class="line"><a name="l00164"></a><span class="lineno"> 164</span>&#160;<span class="preprocessor">#endif // M0 &gt; 3</span></div><div class="line"><a name="l00165"></a><span class="lineno"> 165</span>&#160;<span class="preprocessor">#if M0 &gt; 4</span></div><div class="line"><a name="l00166"></a><span class="lineno"> 166</span>&#160; zin4 = (4 + (uint)(y * M0)) / (uint)HEIGHT_GEMM3D;</div><div class="line"><a name="l00167"></a><span class="lineno"> 167</span>&#160; zin4 = min((uint)(DEPTH_GEMM3D - 1), zin4);</div><div class="line"><a name="l00168"></a><span class="lineno"> 168</span>&#160; zin4 *= (cross_plane_pad * src_stride_y);</div><div class="line"><a name="l00169"></a><span class="lineno"> 169</span>&#160;<span class="preprocessor">#endif // M0 &gt; 4</span></div><div class="line"><a name="l00170"></a><span class="lineno"> 170</span>&#160;<span class="preprocessor">#if M0 &gt; 5</span></div><div class="line"><a name="l00171"></a><span class="lineno"> 171</span>&#160; zin5 = (5 + (uint)(y * M0)) / (uint)HEIGHT_GEMM3D;</div><div class="line"><a name="l00172"></a><span class="lineno"> 172</span>&#160; zin5 = min((uint)(DEPTH_GEMM3D - 1), zin5);</div><div class="line"><a name="l00173"></a><span class="lineno"> 173</span>&#160; zin5 *= (cross_plane_pad * src_stride_y);</div><div class="line"><a name="l00174"></a><span class="lineno"> 174</span>&#160;<span class="preprocessor">#endif // M0 &gt; 5</span></div><div class="line"><a name="l00175"></a><span class="lineno"> 175</span>&#160;<span class="preprocessor">#if M0 &gt; 6</span></div><div class="line"><a name="l00176"></a><span class="lineno"> 176</span>&#160; zin6 = (6 + (uint)(y * M0)) / (uint)HEIGHT_GEMM3D;</div><div class="line"><a name="l00177"></a><span class="lineno"> 177</span>&#160; zin6 = min((uint)(DEPTH_GEMM3D - 1), zin6);</div><div class="line"><a name="l00178"></a><span class="lineno"> 178</span>&#160; zin6 *= (cross_plane_pad * src_stride_y);</div><div class="line"><a name="l00179"></a><span class="lineno"> 179</span>&#160;<span class="preprocessor">#endif // M0 &gt; 6</span></div><div class="line"><a name="l00180"></a><span class="lineno"> 180</span>&#160;<span class="preprocessor">#if M0 &gt; 7</span></div><div class="line"><a name="l00181"></a><span class="lineno"> 181</span>&#160; zin7 = (7 + (uint)(y * M0)) / (uint)HEIGHT_GEMM3D;</div><div class="line"><a name="l00182"></a><span class="lineno"> 182</span>&#160; zin7 = min((uint)(DEPTH_GEMM3D - 1), zin7);</div><div class="line"><a name="l00183"></a><span class="lineno"> 183</span>&#160; zin7 *= (cross_plane_pad * src_stride_y);</div><div class="line"><a name="l00184"></a><span class="lineno"> 184</span>&#160;<span class="preprocessor">#endif // M0 &gt; 7</span></div><div class="line"><a name="l00185"></a><span class="lineno"> 185</span>&#160;</div><div class="line"><a name="l00186"></a><span class="lineno"> 186</span>&#160;<span class="preprocessor">#else // defined(REINTERPRET_INPUT_AS_3D)</span></div><div class="line"><a name="l00187"></a><span class="lineno"> 187</span>&#160;</div><div class="line"><a name="l00188"></a><span class="lineno"> 188</span>&#160; input_ptr += z * (uint)src_stride_z;</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"> 190</span>&#160;<span class="preprocessor">#endif // defined(REINTERPRET_INPUT_AS_3D)</span></div><div class="line"><a name="l00191"></a><span class="lineno"> 191</span>&#160;</div><div class="line"><a name="l00192"></a><span class="lineno"> 192</span>&#160; <span class="comment">// Add offset for batched GEMM</span></div><div class="line"><a name="l00193"></a><span class="lineno"> 193</span>&#160; output_ptr += z * (uint)dst_stride_z;</div><div class="line"><a name="l00194"></a><span class="lineno"> 194</span>&#160;</div><div class="line"><a name="l00195"></a><span class="lineno"> 195</span>&#160; <span class="comment">// ---------------------------Load input values --------------------------------</span></div><div class="line"><a name="l00196"></a><span class="lineno"> 196</span>&#160;</div><div class="line"><a name="l00197"></a><span class="lineno"> 197</span>&#160; <span class="comment">// Load values from the LHS matrix</span></div><div class="line"><a name="l00198"></a><span class="lineno"> 198</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, K0)</div><div class="line"><a name="l00199"></a><span class="lineno"> 199</span>&#160; a0 = <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a287e2fc366c312b468382c95bb90f91f">VLOAD</a>(K0)(0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(input_ptr + 0 * src_stride_y + zin0));</div><div class="line"><a name="l00200"></a><span class="lineno"> 200</span>&#160; BOUNDARY_CONDITION_X(x, a0);</div><div class="line"><a name="l00201"></a><span class="lineno"> 201</span>&#160;<span class="preprocessor">#if M0 &gt; 1</span></div><div class="line"><a name="l00202"></a><span class="lineno"> 202</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, K0)</div><div class="line"><a name="l00203"></a><span class="lineno"> 203</span>&#160; a1 = <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a287e2fc366c312b468382c95bb90f91f">VLOAD</a>(K0)(0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(input_ptr + 1 * src_stride_y + zin1));</div><div class="line"><a name="l00204"></a><span class="lineno"> 204</span>&#160; BOUNDARY_CONDITION_X(x, a1);</div><div class="line"><a name="l00205"></a><span class="lineno"> 205</span>&#160;<span class="preprocessor">#endif // M0 &gt; 1</span></div><div class="line"><a name="l00206"></a><span class="lineno"> 206</span>&#160;<span class="preprocessor">#if M0 &gt; 2</span></div><div class="line"><a name="l00207"></a><span class="lineno"> 207</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, K0)</div><div class="line"><a name="l00208"></a><span class="lineno"> 208</span>&#160; a2 = <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a287e2fc366c312b468382c95bb90f91f">VLOAD</a>(K0)(0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(input_ptr + 2 * src_stride_y + zin2));</div><div class="line"><a name="l00209"></a><span class="lineno"> 209</span>&#160; BOUNDARY_CONDITION_X(x, a2);</div><div class="line"><a name="l00210"></a><span class="lineno"> 210</span>&#160;<span class="preprocessor">#endif // M0 &gt; 2</span></div><div class="line"><a name="l00211"></a><span class="lineno"> 211</span>&#160;<span class="preprocessor">#if M0 &gt; 3</span></div><div class="line"><a name="l00212"></a><span class="lineno"> 212</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, K0)</div><div class="line"><a name="l00213"></a><span class="lineno"> 213</span>&#160; a3 = <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a287e2fc366c312b468382c95bb90f91f">VLOAD</a>(K0)(0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(input_ptr + 3 * src_stride_y + zin3));</div><div class="line"><a name="l00214"></a><span class="lineno"> 214</span>&#160; BOUNDARY_CONDITION_X(x, a3);</div><div class="line"><a name="l00215"></a><span class="lineno"> 215</span>&#160;<span class="preprocessor">#endif // M0 &gt; 3</span></div><div class="line"><a name="l00216"></a><span class="lineno"> 216</span>&#160;<span class="preprocessor">#if M0 &gt; 4</span></div><div class="line"><a name="l00217"></a><span class="lineno"> 217</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, K0)</div><div class="line"><a name="l00218"></a><span class="lineno"> 218</span>&#160; a4 = <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a287e2fc366c312b468382c95bb90f91f">VLOAD</a>(K0)(0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(input_ptr + 4 * src_stride_y + zin4));</div><div class="line"><a name="l00219"></a><span class="lineno"> 219</span>&#160; BOUNDARY_CONDITION_X(x, a4);</div><div class="line"><a name="l00220"></a><span class="lineno"> 220</span>&#160;<span class="preprocessor">#endif // M0 &gt; 4</span></div><div class="line"><a name="l00221"></a><span class="lineno"> 221</span>&#160;<span class="preprocessor">#if M0 &gt; 5</span></div><div class="line"><a name="l00222"></a><span class="lineno"> 222</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, K0)</div><div class="line"><a name="l00223"></a><span class="lineno"> 223</span>&#160; a5 = <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a287e2fc366c312b468382c95bb90f91f">VLOAD</a>(K0)(0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(input_ptr + 5 * src_stride_y + zin5));</div><div class="line"><a name="l00224"></a><span class="lineno"> 224</span>&#160; BOUNDARY_CONDITION_X(x, a5);</div><div class="line"><a name="l00225"></a><span class="lineno"> 225</span>&#160;<span class="preprocessor">#endif // M0 &gt; 5</span></div><div class="line"><a name="l00226"></a><span class="lineno"> 226</span>&#160;<span class="preprocessor">#if M0 &gt; 6</span></div><div class="line"><a name="l00227"></a><span class="lineno"> 227</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, K0)</div><div class="line"><a name="l00228"></a><span class="lineno"> 228</span>&#160; a6 = <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a287e2fc366c312b468382c95bb90f91f">VLOAD</a>(K0)(0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(input_ptr + 6 * src_stride_y + zin6));</div><div class="line"><a name="l00229"></a><span class="lineno"> 229</span>&#160; BOUNDARY_CONDITION_X(x, a6);</div><div class="line"><a name="l00230"></a><span class="lineno"> 230</span>&#160;<span class="preprocessor">#endif // M0 &gt; 6</span></div><div class="line"><a name="l00231"></a><span class="lineno"> 231</span>&#160;<span class="preprocessor">#if M0 &gt; 7</span></div><div class="line"><a name="l00232"></a><span class="lineno"> 232</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, K0)</div><div class="line"><a name="l00233"></a><span class="lineno"> 233</span>&#160; a7 = <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a287e2fc366c312b468382c95bb90f91f">VLOAD</a>(K0)(0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(input_ptr + 7 * src_stride_y + zin7));</div><div class="line"><a name="l00234"></a><span class="lineno"> 234</span>&#160; BOUNDARY_CONDITION_X(x, a7);</div><div class="line"><a name="l00235"></a><span class="lineno"> 235</span>&#160;<span class="preprocessor">#endif // M0 &gt; 7</span></div><div class="line"><a name="l00236"></a><span class="lineno"> 236</span>&#160;</div><div class="line"><a name="l00237"></a><span class="lineno"> 237</span>&#160; <span class="comment">// ---------------------------Store output values ------------------------------</span></div><div class="line"><a name="l00238"></a><span class="lineno"> 238</span>&#160;</div><div class="line"><a name="l00239"></a><span class="lineno"> 239</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#acb282042d1edeeaa3cc979a206f78b54">VSTORE</a>(K0)</div><div class="line"><a name="l00240"></a><span class="lineno"> 240</span>&#160; (a0, 0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(output_ptr + 0 * OUTPUT_STEP_X * <span class="keyword">sizeof</span>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>)));</div><div class="line"><a name="l00241"></a><span class="lineno"> 241</span>&#160;<span class="preprocessor">#if M0 &gt; 1</span></div><div class="line"><a name="l00242"></a><span class="lineno"> 242</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#acb282042d1edeeaa3cc979a206f78b54">VSTORE</a>(K0)</div><div class="line"><a name="l00243"></a><span class="lineno"> 243</span>&#160; (a1, 0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(output_ptr + 1 * OUTPUT_STEP_X * <span class="keyword">sizeof</span>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>)));</div><div class="line"><a name="l00244"></a><span class="lineno"> 244</span>&#160;<span class="preprocessor">#endif // M0 &gt; 1</span></div><div class="line"><a name="l00245"></a><span class="lineno"> 245</span>&#160;<span class="preprocessor">#if M0 &gt; 2</span></div><div class="line"><a name="l00246"></a><span class="lineno"> 246</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#acb282042d1edeeaa3cc979a206f78b54">VSTORE</a>(K0)</div><div class="line"><a name="l00247"></a><span class="lineno"> 247</span>&#160; (a2, 0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(output_ptr + 2 * OUTPUT_STEP_X * <span class="keyword">sizeof</span>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>)));</div><div class="line"><a name="l00248"></a><span class="lineno"> 248</span>&#160;<span class="preprocessor">#endif // M0 &gt; 2</span></div><div class="line"><a name="l00249"></a><span class="lineno"> 249</span>&#160;<span class="preprocessor">#if M0 &gt; 3</span></div><div class="line"><a name="l00250"></a><span class="lineno"> 250</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#acb282042d1edeeaa3cc979a206f78b54">VSTORE</a>(K0)</div><div class="line"><a name="l00251"></a><span class="lineno"> 251</span>&#160; (a3, 0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(output_ptr + 3 * OUTPUT_STEP_X * <span class="keyword">sizeof</span>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>)));</div><div class="line"><a name="l00252"></a><span class="lineno"> 252</span>&#160;<span class="preprocessor">#endif // M0 &gt; 3</span></div><div class="line"><a name="l00253"></a><span class="lineno"> 253</span>&#160;<span class="preprocessor">#if M0 &gt; 4</span></div><div class="line"><a name="l00254"></a><span class="lineno"> 254</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#acb282042d1edeeaa3cc979a206f78b54">VSTORE</a>(K0)</div><div class="line"><a name="l00255"></a><span class="lineno"> 255</span>&#160; (a4, 0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(output_ptr + 4 * OUTPUT_STEP_X * <span class="keyword">sizeof</span>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>)));</div><div class="line"><a name="l00256"></a><span class="lineno"> 256</span>&#160;<span class="preprocessor">#endif // M0 &gt; 4</span></div><div class="line"><a name="l00257"></a><span class="lineno"> 257</span>&#160;<span class="preprocessor">#if M0 &gt; 5</span></div><div class="line"><a name="l00258"></a><span class="lineno"> 258</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#acb282042d1edeeaa3cc979a206f78b54">VSTORE</a>(K0)</div><div class="line"><a name="l00259"></a><span class="lineno"> 259</span>&#160; (a5, 0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(output_ptr + 5 * OUTPUT_STEP_X * <span class="keyword">sizeof</span>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>)));</div><div class="line"><a name="l00260"></a><span class="lineno"> 260</span>&#160;<span class="preprocessor">#endif // M0 &gt; 5</span></div><div class="line"><a name="l00261"></a><span class="lineno"> 261</span>&#160;<span class="preprocessor">#if M0 &gt; 6</span></div><div class="line"><a name="l00262"></a><span class="lineno"> 262</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#acb282042d1edeeaa3cc979a206f78b54">VSTORE</a>(K0)</div><div class="line"><a name="l00263"></a><span class="lineno"> 263</span>&#160; (a6, 0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(output_ptr + 6 * OUTPUT_STEP_X * <span class="keyword">sizeof</span>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>)));</div><div class="line"><a name="l00264"></a><span class="lineno"> 264</span>&#160;<span class="preprocessor">#endif // M0 &gt; 6</span></div><div class="line"><a name="l00265"></a><span class="lineno"> 265</span>&#160;<span class="preprocessor">#if M0 &gt; 7</span></div><div class="line"><a name="l00266"></a><span class="lineno"> 266</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#acb282042d1edeeaa3cc979a206f78b54">VSTORE</a>(K0)</div><div class="line"><a name="l00267"></a><span class="lineno"> 267</span>&#160; (a7, 0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(output_ptr + 7 * OUTPUT_STEP_X * <span class="keyword">sizeof</span>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>)));</div><div class="line"><a name="l00268"></a><span class="lineno"> 268</span>&#160;<span class="preprocessor">#endif // M0 &gt; 7</span></div><div class="line"><a name="l00269"></a><span class="lineno"> 269</span>&#160;</div><div class="line"><a name="l00270"></a><span class="lineno"> 270</span>&#160;<span class="preprocessor">#undef BLOCK_SIZE</span></div><div class="line"><a name="l00271"></a><span class="lineno"> 271</span>&#160;<span class="preprocessor">#undef OUTPUT_OFFSET_X</span></div><div class="line"><a name="l00272"></a><span class="lineno"> 272</span>&#160;<span class="preprocessor">#undef OUTPUT_STEP_X</span></div><div class="line"><a name="l00273"></a><span class="lineno"> 273</span>&#160;}</div><div class="line"><a name="l00274"></a><span class="lineno"> 274</span>&#160;</div><div class="line"><a name="l00275"></a><span class="lineno"> 275</span>&#160;<span class="preprocessor">#if M0 == 2</span></div><div class="line"><a name="l00276"></a><span class="lineno"> 276</span>&#160;<span class="preprocessor">#define TRANSPOSE_COLUMN_AND_STORE(output_ptr, output_step_x, i) \</span></div><div class="line"><a name="l00277"></a><span class="lineno"> 277</span>&#160;<span class="preprocessor"> ({ \</span></div><div class="line"><a name="l00278"></a><span class="lineno"> 278</span>&#160;<span class="preprocessor"> VEC_DATA_TYPE(DATA_TYPE, M0) \</span></div><div class="line"><a name="l00279"></a><span class="lineno"> 279</span>&#160;<span class="preprocessor"> res = (VEC_DATA_TYPE(DATA_TYPE, M0))(a0.s##i, a1.s##i); \</span></div><div class="line"><a name="l00280"></a><span class="lineno"> 280</span>&#160;<span class="preprocessor"> VSTORE(M0) \</span></div><div class="line"><a name="l00281"></a><span class="lineno"> 281</span>&#160;<span class="preprocessor"> (res, 0, (__global DATA_TYPE *)(output_ptr + 0x##i * output_step_x * sizeof(DATA_TYPE))); \</span></div><div class="line"><a name="l00282"></a><span class="lineno"> 282</span>&#160;<span class="preprocessor"> })</span></div><div class="line"><a name="l00283"></a><span class="lineno"> 283</span>&#160;<span class="preprocessor">#elif M0 == 3 // M0 == 3</span></div><div class="line"><a name="l00284"></a><span class="lineno"> 284</span>&#160;<span class="preprocessor">#define TRANSPOSE_COLUMN_AND_STORE(output_ptr, output_step_x, i) \</span></div><div class="line"><a name="l00285"></a><span class="lineno"> 285</span>&#160;<span class="preprocessor"> ({ \</span></div><div class="line"><a name="l00286"></a><span class="lineno"> 286</span>&#160;<span class="preprocessor"> VEC_DATA_TYPE(DATA_TYPE, M0) \</span></div><div class="line"><a name="l00287"></a><span class="lineno"> 287</span>&#160;<span class="preprocessor"> res = (VEC_DATA_TYPE(DATA_TYPE, M0))(a0.s##i, a1.s##i, a2.s##i); \</span></div><div class="line"><a name="l00288"></a><span class="lineno"> 288</span>&#160;<span class="preprocessor"> VSTORE(M0) \</span></div><div class="line"><a name="l00289"></a><span class="lineno"> 289</span>&#160;<span class="preprocessor"> (res, 0, (__global DATA_TYPE *)(output_ptr + 0x##i * output_step_x * sizeof(DATA_TYPE))); \</span></div><div class="line"><a name="l00290"></a><span class="lineno"> 290</span>&#160;<span class="preprocessor"> })</span></div><div class="line"><a name="l00291"></a><span class="lineno"> 291</span>&#160;<span class="preprocessor">#elif M0 == 4 // M0 == 4</span></div><div class="line"><a name="l00292"></a><span class="lineno"> 292</span>&#160;<span class="preprocessor">#define TRANSPOSE_COLUMN_AND_STORE(output_ptr, output_step_x, i) \</span></div><div class="line"><a name="l00293"></a><span class="lineno"> 293</span>&#160;<span class="preprocessor"> ({ \</span></div><div class="line"><a name="l00294"></a><span class="lineno"> 294</span>&#160;<span class="preprocessor"> VEC_DATA_TYPE(DATA_TYPE, M0) \</span></div><div class="line"><a name="l00295"></a><span class="lineno"> 295</span>&#160;<span class="preprocessor"> res = (VEC_DATA_TYPE(DATA_TYPE, M0))(a0.s##i, a1.s##i, a2.s##i, a3.s##i); \</span></div><div class="line"><a name="l00296"></a><span class="lineno"> 296</span>&#160;<span class="preprocessor"> VSTORE(M0) \</span></div><div class="line"><a name="l00297"></a><span class="lineno"> 297</span>&#160;<span class="preprocessor"> (res, 0, (__global DATA_TYPE *)(output_ptr + 0x##i * output_step_x * sizeof(DATA_TYPE))); \</span></div><div class="line"><a name="l00298"></a><span class="lineno"> 298</span>&#160;<span class="preprocessor"> })</span></div><div class="line"><a name="l00299"></a><span class="lineno"> 299</span>&#160;<span class="preprocessor">#elif M0 == 5 // M0 == 5</span></div><div class="line"><a name="l00300"></a><span class="lineno"> 300</span>&#160;<span class="preprocessor">#define TRANSPOSE_COLUMN_AND_STORE(output_ptr, output_step_x, i) \</span></div><div class="line"><a name="l00301"></a><span class="lineno"> 301</span>&#160;<span class="preprocessor"> ({ \</span></div><div class="line"><a name="l00302"></a><span class="lineno"> 302</span>&#160;<span class="preprocessor"> VEC_DATA_TYPE(DATA_TYPE, 4) \</span></div><div class="line"><a name="l00303"></a><span class="lineno"> 303</span>&#160;<span class="preprocessor"> res0 = (VEC_DATA_TYPE(DATA_TYPE, 4))(a0.s##i, a1.s##i, a2.s##i, a3.s##i); \</span></div><div class="line"><a name="l00304"></a><span class="lineno"> 304</span>&#160;<span class="preprocessor"> DATA_TYPE res1 = a4.s##i; \</span></div><div class="line"><a name="l00305"></a><span class="lineno"> 305</span>&#160;<span class="preprocessor"> VSTORE(4) \</span></div><div class="line"><a name="l00306"></a><span class="lineno"> 306</span>&#160;<span class="preprocessor"> (res0, 0, (__global DATA_TYPE *)(output_ptr + 0x##i * output_step_x * sizeof(DATA_TYPE))); \</span></div><div class="line"><a name="l00307"></a><span class="lineno"> 307</span>&#160;<span class="preprocessor"> *((__global DATA_TYPE *)(output_ptr + 0x##i * output_step_x * sizeof(DATA_TYPE)) + 4) = res1; \</span></div><div class="line"><a name="l00308"></a><span class="lineno"> 308</span>&#160;<span class="preprocessor"> })</span></div><div class="line"><a name="l00309"></a><span class="lineno"> 309</span>&#160;<span class="preprocessor">#elif M0 == 6 // M0 == 6</span></div><div class="line"><a name="l00310"></a><span class="lineno"> 310</span>&#160;<span class="preprocessor">#define TRANSPOSE_COLUMN_AND_STORE(output_ptr, output_step_x, i) \</span></div><div class="line"><a name="l00311"></a><span class="lineno"> 311</span>&#160;<span class="preprocessor"> ({ \</span></div><div class="line"><a name="l00312"></a><span class="lineno"> 312</span>&#160;<span class="preprocessor"> VEC_DATA_TYPE(DATA_TYPE, 4) \</span></div><div class="line"><a name="l00313"></a><span class="lineno"> 313</span>&#160;<span class="preprocessor"> res0 = (VEC_DATA_TYPE(DATA_TYPE, 4))(a0.s##i, a1.s##i, a2.s##i, a3.s##i); \</span></div><div class="line"><a name="l00314"></a><span class="lineno"> 314</span>&#160;<span class="preprocessor"> VEC_DATA_TYPE(DATA_TYPE, 2) \</span></div><div class="line"><a name="l00315"></a><span class="lineno"> 315</span>&#160;<span class="preprocessor"> res1 = (VEC_DATA_TYPE(DATA_TYPE, 2))(a4.s##i, a5.s##i); \</span></div><div class="line"><a name="l00316"></a><span class="lineno"> 316</span>&#160;<span class="preprocessor"> VSTORE(4) \</span></div><div class="line"><a name="l00317"></a><span class="lineno"> 317</span>&#160;<span class="preprocessor"> (res0, 0, (__global DATA_TYPE *)(output_ptr + 0x##i * output_step_x * sizeof(DATA_TYPE))); \</span></div><div class="line"><a name="l00318"></a><span class="lineno"> 318</span>&#160;<span class="preprocessor"> VSTORE(2) \</span></div><div class="line"><a name="l00319"></a><span class="lineno"> 319</span>&#160;<span class="preprocessor"> (res1, 0, (__global DATA_TYPE *)(output_ptr + 0x##i * output_step_x * sizeof(DATA_TYPE)) + 4); \</span></div><div class="line"><a name="l00320"></a><span class="lineno"> 320</span>&#160;<span class="preprocessor"> })</span></div><div class="line"><a name="l00321"></a><span class="lineno"> 321</span>&#160;<span class="preprocessor">#elif M0 == 7 // M0 == 7</span></div><div class="line"><a name="l00322"></a><span class="lineno"> 322</span>&#160;<span class="preprocessor">#define TRANSPOSE_COLUMN_AND_STORE(output_ptr, output_step_x, i) \</span></div><div class="line"><a name="l00323"></a><span class="lineno"> 323</span>&#160;<span class="preprocessor"> ({ \</span></div><div class="line"><a name="l00324"></a><span class="lineno"> 324</span>&#160;<span class="preprocessor"> VEC_DATA_TYPE(DATA_TYPE, 4) \</span></div><div class="line"><a name="l00325"></a><span class="lineno"> 325</span>&#160;<span class="preprocessor"> res0 = (VEC_DATA_TYPE(DATA_TYPE, 4))(a0.s##i, a1.s##i, a2.s##i, a3.s##i); \</span></div><div class="line"><a name="l00326"></a><span class="lineno"> 326</span>&#160;<span class="preprocessor"> VEC_DATA_TYPE(DATA_TYPE, 3) \</span></div><div class="line"><a name="l00327"></a><span class="lineno"> 327</span>&#160;<span class="preprocessor"> res1 = (VEC_DATA_TYPE(DATA_TYPE, 3))(a4.s##i, a5.s##i, a6.s##i); \</span></div><div class="line"><a name="l00328"></a><span class="lineno"> 328</span>&#160;<span class="preprocessor"> VSTORE(4) \</span></div><div class="line"><a name="l00329"></a><span class="lineno"> 329</span>&#160;<span class="preprocessor"> (res0, 0, (__global DATA_TYPE *)(output_ptr + 0x##i * output_step_x * sizeof(DATA_TYPE))); \</span></div><div class="line"><a name="l00330"></a><span class="lineno"> 330</span>&#160;<span class="preprocessor"> VSTORE(3) \</span></div><div class="line"><a name="l00331"></a><span class="lineno"> 331</span>&#160;<span class="preprocessor"> (res1, 0, (__global DATA_TYPE *)(output_ptr + 0x##i * output_step_x * sizeof(DATA_TYPE)) + 4); \</span></div><div class="line"><a name="l00332"></a><span class="lineno"> 332</span>&#160;<span class="preprocessor"> })</span></div><div class="line"><a name="l00333"></a><span class="lineno"> 333</span>&#160;<span class="preprocessor">#elif M0 == 8 // M0 == 8</span></div><div class="line"><a name="l00334"></a><span class="lineno"> 334</span>&#160;<span class="preprocessor">#define TRANSPOSE_COLUMN_AND_STORE(output_ptr, output_step_x, i) \</span></div><div class="line"><a name="l00335"></a><span class="lineno"> 335</span>&#160;<span class="preprocessor"> ({ \</span></div><div class="line"><a name="l00336"></a><span class="lineno"> 336</span>&#160;<span class="preprocessor"> VEC_DATA_TYPE(DATA_TYPE, M0) \</span></div><div class="line"><a name="l00337"></a><span class="lineno"> 337</span>&#160;<span class="preprocessor"> res = (VEC_DATA_TYPE(DATA_TYPE, M0))(a0.s##i, a1.s##i, a2.s##i, a3.s##i, a4.s##i, a5.s##i, a6.s##i, a7.s##i); \</span></div><div class="line"><a name="l00338"></a><span class="lineno"> 338</span>&#160;<span class="preprocessor"> VSTORE(M0) \</span></div><div class="line"><a name="l00339"></a><span class="lineno"> 339</span>&#160;<span class="preprocessor"> (res, 0, (__global DATA_TYPE *)(output_ptr + 0x##i * output_step_x * sizeof(DATA_TYPE))); \</span></div><div class="line"><a name="l00340"></a><span class="lineno"> 340</span>&#160;<span class="preprocessor"> })</span></div><div class="line"><a name="l00341"></a><span class="lineno"> 341</span>&#160;<span class="preprocessor">#else // M0 not supported</span></div><div class="line"><a name="l00342"></a><span class="lineno"> 342</span>&#160;<span class="preprocessor">#error &quot;M0 value not supported&quot;</span></div><div class="line"><a name="l00343"></a><span class="lineno"> 343</span>&#160;<span class="preprocessor">#endif // N0 conditions</span></div><div class="line"><a name="l00344"></a><span class="lineno"> 344</span>&#160;</div><div class="line"><a name="l00381"></a><span class="lineno"> 381</span>&#160;__kernel <span class="keywordtype">void</span> gemm_reshape_lhs_matrix_t(<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#ae9d2dc29c2789c253406f9b304cc75a8">src</a>),</div><div class="line"><a name="l00382"></a><span class="lineno"> 382</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="l00383"></a><span class="lineno"> 383</span>&#160;#<span class="keywordflow">if</span> defined(REINTERPRET_INPUT_AS_3D)</div><div class="line"><a name="l00384"></a><span class="lineno"> 384</span>&#160; ,</div><div class="line"><a name="l00385"></a><span class="lineno"> 385</span>&#160; uint cross_plane_pad</div><div class="line"><a name="l00386"></a><span class="lineno"> 386</span>&#160;#endif <span class="comment">// REINTERPRET_INPUT_AS_3D</span></div><div class="line"><a name="l00387"></a><span class="lineno"> 387</span>&#160; )</div><div class="line"><a name="l00388"></a><span class="lineno"> 388</span>&#160;{</div><div class="line"><a name="l00389"></a><span class="lineno"> 389</span>&#160; <span class="comment">// Block size</span></div><div class="line"><a name="l00390"></a><span class="lineno"> 390</span>&#160;<span class="preprocessor">#define BLOCK_SIZE ((M0) * (K0))</span></div><div class="line"><a name="l00391"></a><span class="lineno"> 391</span>&#160;</div><div class="line"><a name="l00392"></a><span class="lineno"> 392</span>&#160; <span class="comment">// Output offset X</span></div><div class="line"><a name="l00393"></a><span class="lineno"> 393</span>&#160;<span class="preprocessor">#if defined(INTERLEAVE)</span></div><div class="line"><a name="l00394"></a><span class="lineno"> 394</span>&#160;<span class="preprocessor">#define OUTPUT_OFFSET_X (M0)</span></div><div class="line"><a name="l00395"></a><span class="lineno"> 395</span>&#160;<span class="preprocessor">#else // defined(INTERLEAVE)</span></div><div class="line"><a name="l00396"></a><span class="lineno"> 396</span>&#160;<span class="preprocessor">#define OUTPUT_OFFSET_X (BLOCK_SIZE)</span></div><div class="line"><a name="l00397"></a><span class="lineno"> 397</span>&#160;<span class="preprocessor">#endif // defined(INTERLEAVE)</span></div><div class="line"><a name="l00398"></a><span class="lineno"> 398</span>&#160;</div><div class="line"><a name="l00399"></a><span class="lineno"> 399</span>&#160; <span class="comment">// Output step X</span></div><div class="line"><a name="l00400"></a><span class="lineno"> 400</span>&#160;<span class="preprocessor">#if defined(INTERLEAVE)</span></div><div class="line"><a name="l00401"></a><span class="lineno"> 401</span>&#160;<span class="preprocessor">#define OUTPUT_STEP_X (M0) * (V0)</span></div><div class="line"><a name="l00402"></a><span class="lineno"> 402</span>&#160;<span class="preprocessor">#else // Do not interleave</span></div><div class="line"><a name="l00403"></a><span class="lineno"> 403</span>&#160;<span class="preprocessor">#define OUTPUT_STEP_X (M0)</span></div><div class="line"><a name="l00404"></a><span class="lineno"> 404</span>&#160;<span class="preprocessor">#endif // defined(INTERLEAVE)</span></div><div class="line"><a name="l00405"></a><span class="lineno"> 405</span>&#160;</div><div class="line"><a name="l00406"></a><span class="lineno"> 406</span>&#160; <span class="comment">// Compute source and destination addresses</span></div><div class="line"><a name="l00407"></a><span class="lineno"> 407</span>&#160; uint x = get_global_id(0);</div><div class="line"><a name="l00408"></a><span class="lineno"> 408</span>&#160; uint y = get_global_id(1);</div><div class="line"><a name="l00409"></a><span class="lineno"> 409</span>&#160; uint z = get_global_id(2);</div><div class="line"><a name="l00410"></a><span class="lineno"> 410</span>&#160;</div><div class="line"><a name="l00411"></a><span class="lineno"> 411</span>&#160; <span class="comment">// ------------------ Compute input/output addresses ---------------------------</span></div><div class="line"><a name="l00412"></a><span class="lineno"> 412</span>&#160;</div><div class="line"><a name="l00413"></a><span class="lineno"> 413</span>&#160; <span class="comment">// Compute the input address</span></div><div class="line"><a name="l00414"></a><span class="lineno"> 414</span>&#160; __global uchar *input_ptr = src_ptr + src_offset_first_element_in_bytes + x * (uint)K0 * <span class="keyword">sizeof</span>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>) + y * (uint)M0 * src_stride_y;</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">// Compute the output address</span></div><div class="line"><a name="l00417"></a><span class="lineno"> 417</span>&#160; __global uchar *output_ptr = dst_ptr + dst_offset_first_element_in_bytes + (x * (uint)BLOCK_SIZE * (uint)V0 * <span class="keyword">sizeof</span>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>)) + ((y / (uint)V0) * (uint)dst_stride_y) + ((y % V0) *</div><div class="line"><a name="l00418"></a><span class="lineno"> 418</span>&#160; (uint)OUTPUT_OFFSET_X * <span class="keyword">sizeof</span>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>));</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">// Create variables: uint zin0=0, zin1=0, zin2=0...zin(M0-1)=0;</span></div><div class="line"><a name="l00421"></a><span class="lineno"> 421</span>&#160; <a class="code" href="repeat_8h.xhtml#a4a61a8df72f9bc1b1dcc4cc4ec6ac3b9">REPEAT_VAR_INIT_TO_CONST</a>(M0, uint, zin, 0);</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="preprocessor">#if defined(REINTERPRET_INPUT_AS_3D)</span></div><div class="line"><a name="l00424"></a><span class="lineno"> 424</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="l00425"></a><span class="lineno"> 425</span>&#160; <span class="comment">// multiply src_stride_z by DEPTH_GEMM3D</span></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">// Note for the REINTERPRET_INPUT_AS_3D case</span></div><div class="line"><a name="l00428"></a><span class="lineno"> 428</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="l00429"></a><span class="lineno"> 429</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="l00430"></a><span class="lineno"> 430</span>&#160; <span class="comment">//</span></div><div class="line"><a name="l00431"></a><span class="lineno"> 431</span>&#160; <span class="comment">// | |</span></div><div class="line"><a name="l00432"></a><span class="lineno"> 432</span>&#160; <span class="comment">// | plane0 |</span></div><div class="line"><a name="l00433"></a><span class="lineno"> 433</span>&#160; <span class="comment">// | |</span></div><div class="line"><a name="l00434"></a><span class="lineno"> 434</span>&#160; <span class="comment">// |__________________|</span></div><div class="line"><a name="l00435"></a><span class="lineno"> 435</span>&#160; <span class="comment">// |******************|</span></div><div class="line"><a name="l00436"></a><span class="lineno"> 436</span>&#160; <span class="comment">// | cross_plane_pad |</span></div><div class="line"><a name="l00437"></a><span class="lineno"> 437</span>&#160; <span class="comment">// |******************|</span></div><div class="line"><a name="l00438"></a><span class="lineno"> 438</span>&#160; <span class="comment">// | |</span></div><div class="line"><a name="l00439"></a><span class="lineno"> 439</span>&#160; <span class="comment">// | plane1 |</span></div><div class="line"><a name="l00440"></a><span class="lineno"> 440</span>&#160; <span class="comment">// | |</span></div><div class="line"><a name="l00441"></a><span class="lineno"> 441</span>&#160; <span class="comment">// |__________________|</span></div><div class="line"><a name="l00442"></a><span class="lineno"> 442</span>&#160;</div><div class="line"><a name="l00443"></a><span class="lineno"> 443</span>&#160; input_ptr += z * (uint)src_stride_z * DEPTH_GEMM3D;</div><div class="line"><a name="l00444"></a><span class="lineno"> 444</span>&#160;</div><div class="line"><a name="l00445"></a><span class="lineno"> 445</span>&#160; <span class="comment">// The plane (zin) is calculated dividing M (y * M0) by HEIGHT_GEMM3D</span></div><div class="line"><a name="l00446"></a><span class="lineno"> 446</span>&#160; zin0 = (0 + (uint)(y * M0)) / (uint)HEIGHT_GEMM3D;</div><div class="line"><a name="l00447"></a><span class="lineno"> 447</span>&#160; zin0 = min((uint)(DEPTH_GEMM3D - 1), zin0);</div><div class="line"><a name="l00448"></a><span class="lineno"> 448</span>&#160; zin0 *= (cross_plane_pad * src_stride_y);</div><div class="line"><a name="l00449"></a><span class="lineno"> 449</span>&#160;<span class="preprocessor">#if M0 &gt; 1</span></div><div class="line"><a name="l00450"></a><span class="lineno"> 450</span>&#160; zin1 = (1 + (uint)(y * M0)) / (uint)HEIGHT_GEMM3D;</div><div class="line"><a name="l00451"></a><span class="lineno"> 451</span>&#160; zin1 = min((uint)(DEPTH_GEMM3D - 1), zin1);</div><div class="line"><a name="l00452"></a><span class="lineno"> 452</span>&#160; zin1 *= (cross_plane_pad * src_stride_y);</div><div class="line"><a name="l00453"></a><span class="lineno"> 453</span>&#160;<span class="preprocessor">#endif // M0 &gt; 1</span></div><div class="line"><a name="l00454"></a><span class="lineno"> 454</span>&#160;<span class="preprocessor">#if M0 &gt; 2</span></div><div class="line"><a name="l00455"></a><span class="lineno"> 455</span>&#160; zin2 = (2 + (uint)(y * M0)) / (uint)HEIGHT_GEMM3D;</div><div class="line"><a name="l00456"></a><span class="lineno"> 456</span>&#160; zin2 = min((uint)(DEPTH_GEMM3D - 1), zin2);</div><div class="line"><a name="l00457"></a><span class="lineno"> 457</span>&#160; zin2 *= (cross_plane_pad * src_stride_y);</div><div class="line"><a name="l00458"></a><span class="lineno"> 458</span>&#160;<span class="preprocessor">#endif // M0 &gt; 2</span></div><div class="line"><a name="l00459"></a><span class="lineno"> 459</span>&#160;<span class="preprocessor">#if M0 &gt; 3</span></div><div class="line"><a name="l00460"></a><span class="lineno"> 460</span>&#160; zin3 = (3 + (uint)(y * M0)) / (uint)HEIGHT_GEMM3D;</div><div class="line"><a name="l00461"></a><span class="lineno"> 461</span>&#160; zin3 = min((uint)(DEPTH_GEMM3D - 1), zin3);</div><div class="line"><a name="l00462"></a><span class="lineno"> 462</span>&#160; zin3 *= (cross_plane_pad * src_stride_y);</div><div class="line"><a name="l00463"></a><span class="lineno"> 463</span>&#160;<span class="preprocessor">#endif // M0 &gt; 3</span></div><div class="line"><a name="l00464"></a><span class="lineno"> 464</span>&#160;<span class="preprocessor">#if M0 &gt; 4</span></div><div class="line"><a name="l00465"></a><span class="lineno"> 465</span>&#160; zin4 = (4 + (uint)(y * M0)) / (uint)HEIGHT_GEMM3D;</div><div class="line"><a name="l00466"></a><span class="lineno"> 466</span>&#160; zin4 = min((uint)(DEPTH_GEMM3D - 1), zin4);</div><div class="line"><a name="l00467"></a><span class="lineno"> 467</span>&#160; zin4 *= (cross_plane_pad * src_stride_y);</div><div class="line"><a name="l00468"></a><span class="lineno"> 468</span>&#160;<span class="preprocessor">#endif // M0 &gt; 4</span></div><div class="line"><a name="l00469"></a><span class="lineno"> 469</span>&#160;<span class="preprocessor">#if M0 &gt; 5</span></div><div class="line"><a name="l00470"></a><span class="lineno"> 470</span>&#160; zin5 = (5 + (uint)(y * M0)) / (uint)HEIGHT_GEMM3D;</div><div class="line"><a name="l00471"></a><span class="lineno"> 471</span>&#160; zin5 = min((uint)(DEPTH_GEMM3D - 1), zin5);</div><div class="line"><a name="l00472"></a><span class="lineno"> 472</span>&#160; zin5 *= (cross_plane_pad * src_stride_y);</div><div class="line"><a name="l00473"></a><span class="lineno"> 473</span>&#160;<span class="preprocessor">#endif // M0 &gt; 5</span></div><div class="line"><a name="l00474"></a><span class="lineno"> 474</span>&#160;<span class="preprocessor">#if M0 &gt; 6</span></div><div class="line"><a name="l00475"></a><span class="lineno"> 475</span>&#160; zin6 = (6 + (uint)(y * M0)) / (uint)HEIGHT_GEMM3D;</div><div class="line"><a name="l00476"></a><span class="lineno"> 476</span>&#160; zin6 = min((uint)(DEPTH_GEMM3D - 1), zin6);</div><div class="line"><a name="l00477"></a><span class="lineno"> 477</span>&#160; zin6 *= (cross_plane_pad * src_stride_y);</div><div class="line"><a name="l00478"></a><span class="lineno"> 478</span>&#160;<span class="preprocessor">#endif // M0 &gt; 6</span></div><div class="line"><a name="l00479"></a><span class="lineno"> 479</span>&#160;<span class="preprocessor">#if M0 &gt; 7</span></div><div class="line"><a name="l00480"></a><span class="lineno"> 480</span>&#160; zin7 = (7 + (uint)(y * M0)) / (uint)HEIGHT_GEMM3D;</div><div class="line"><a name="l00481"></a><span class="lineno"> 481</span>&#160; zin7 = min((uint)(DEPTH_GEMM3D - 1), zin7);</div><div class="line"><a name="l00482"></a><span class="lineno"> 482</span>&#160; zin7 *= (cross_plane_pad * src_stride_y);</div><div class="line"><a name="l00483"></a><span class="lineno"> 483</span>&#160;<span class="preprocessor">#endif // M0 &gt; 7</span></div><div class="line"><a name="l00484"></a><span class="lineno"> 484</span>&#160;</div><div class="line"><a name="l00485"></a><span class="lineno"> 485</span>&#160;<span class="preprocessor">#else // defined(REINTERPRET_INPUT_AS_3D)</span></div><div class="line"><a name="l00486"></a><span class="lineno"> 486</span>&#160;</div><div class="line"><a name="l00487"></a><span class="lineno"> 487</span>&#160; input_ptr += z * (uint)src_stride_z;</div><div class="line"><a name="l00488"></a><span class="lineno"> 488</span>&#160;</div><div class="line"><a name="l00489"></a><span class="lineno"> 489</span>&#160;<span class="preprocessor">#endif // defined(REINTERPRET_INPUT_AS_3D)</span></div><div class="line"><a name="l00490"></a><span class="lineno"> 490</span>&#160;</div><div class="line"><a name="l00491"></a><span class="lineno"> 491</span>&#160; <span class="comment">// Add offset for batched GEMM</span></div><div class="line"><a name="l00492"></a><span class="lineno"> 492</span>&#160; output_ptr += z * (uint)dst_stride_z;</div><div class="line"><a name="l00493"></a><span class="lineno"> 493</span>&#160;</div><div class="line"><a name="l00494"></a><span class="lineno"> 494</span>&#160; <span class="comment">// ---------------------------Load input values --------------------------------</span></div><div class="line"><a name="l00495"></a><span class="lineno"> 495</span>&#160;</div><div class="line"><a name="l00496"></a><span class="lineno"> 496</span>&#160; <span class="comment">// Load values from the LHS matrix</span></div><div class="line"><a name="l00497"></a><span class="lineno"> 497</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, K0)</div><div class="line"><a name="l00498"></a><span class="lineno"> 498</span>&#160; a0 = <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a287e2fc366c312b468382c95bb90f91f">VLOAD</a>(K0)(0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(input_ptr + 0 * src_stride_y + zin0));</div><div class="line"><a name="l00499"></a><span class="lineno"> 499</span>&#160; BOUNDARY_CONDITION_X(x, a0);</div><div class="line"><a name="l00500"></a><span class="lineno"> 500</span>&#160;<span class="preprocessor">#if M0 &gt; 1</span></div><div class="line"><a name="l00501"></a><span class="lineno"> 501</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, K0)</div><div class="line"><a name="l00502"></a><span class="lineno"> 502</span>&#160; a1 = <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a287e2fc366c312b468382c95bb90f91f">VLOAD</a>(K0)(0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(input_ptr + 1 * src_stride_y + zin1));</div><div class="line"><a name="l00503"></a><span class="lineno"> 503</span>&#160; BOUNDARY_CONDITION_X(x, a1);</div><div class="line"><a name="l00504"></a><span class="lineno"> 504</span>&#160;<span class="preprocessor">#endif // M0 &gt; 1</span></div><div class="line"><a name="l00505"></a><span class="lineno"> 505</span>&#160;<span class="preprocessor">#if M0 &gt; 2</span></div><div class="line"><a name="l00506"></a><span class="lineno"> 506</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, K0)</div><div class="line"><a name="l00507"></a><span class="lineno"> 507</span>&#160; a2 = <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a287e2fc366c312b468382c95bb90f91f">VLOAD</a>(K0)(0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(input_ptr + 2 * src_stride_y + zin2));</div><div class="line"><a name="l00508"></a><span class="lineno"> 508</span>&#160; BOUNDARY_CONDITION_X(x, a2);</div><div class="line"><a name="l00509"></a><span class="lineno"> 509</span>&#160;<span class="preprocessor">#endif // M0 &gt; 2</span></div><div class="line"><a name="l00510"></a><span class="lineno"> 510</span>&#160;<span class="preprocessor">#if M0 &gt; 3</span></div><div class="line"><a name="l00511"></a><span class="lineno"> 511</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, K0)</div><div class="line"><a name="l00512"></a><span class="lineno"> 512</span>&#160; a3 = <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a287e2fc366c312b468382c95bb90f91f">VLOAD</a>(K0)(0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(input_ptr + 3 * src_stride_y + zin3));</div><div class="line"><a name="l00513"></a><span class="lineno"> 513</span>&#160; BOUNDARY_CONDITION_X(x, a3);</div><div class="line"><a name="l00514"></a><span class="lineno"> 514</span>&#160;<span class="preprocessor">#endif // M0 &gt; 3</span></div><div class="line"><a name="l00515"></a><span class="lineno"> 515</span>&#160;<span class="preprocessor">#if M0 &gt; 4</span></div><div class="line"><a name="l00516"></a><span class="lineno"> 516</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, K0)</div><div class="line"><a name="l00517"></a><span class="lineno"> 517</span>&#160; a4 = <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a287e2fc366c312b468382c95bb90f91f">VLOAD</a>(K0)(0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(input_ptr + 4 * src_stride_y + zin4));</div><div class="line"><a name="l00518"></a><span class="lineno"> 518</span>&#160; BOUNDARY_CONDITION_X(x, a4);</div><div class="line"><a name="l00519"></a><span class="lineno"> 519</span>&#160;<span class="preprocessor">#endif // M0 &gt; 4</span></div><div class="line"><a name="l00520"></a><span class="lineno"> 520</span>&#160;<span class="preprocessor">#if M0 &gt; 5</span></div><div class="line"><a name="l00521"></a><span class="lineno"> 521</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, K0)</div><div class="line"><a name="l00522"></a><span class="lineno"> 522</span>&#160; a5 = <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a287e2fc366c312b468382c95bb90f91f">VLOAD</a>(K0)(0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(input_ptr + 5 * src_stride_y + zin5));</div><div class="line"><a name="l00523"></a><span class="lineno"> 523</span>&#160; BOUNDARY_CONDITION_X(x, a5);</div><div class="line"><a name="l00524"></a><span class="lineno"> 524</span>&#160;<span class="preprocessor">#endif // M0 &gt; 5</span></div><div class="line"><a name="l00525"></a><span class="lineno"> 525</span>&#160;<span class="preprocessor">#if M0 &gt; 6</span></div><div class="line"><a name="l00526"></a><span class="lineno"> 526</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, K0)</div><div class="line"><a name="l00527"></a><span class="lineno"> 527</span>&#160; a6 = <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a287e2fc366c312b468382c95bb90f91f">VLOAD</a>(K0)(0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(input_ptr + 6 * src_stride_y + zin6));</div><div class="line"><a name="l00528"></a><span class="lineno"> 528</span>&#160; BOUNDARY_CONDITION_X(x, a6);</div><div class="line"><a name="l00529"></a><span class="lineno"> 529</span>&#160;<span class="preprocessor">#endif // M0 &gt; 6</span></div><div class="line"><a name="l00530"></a><span class="lineno"> 530</span>&#160;<span class="preprocessor">#if M0 &gt; 7</span></div><div class="line"><a name="l00531"></a><span class="lineno"> 531</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, K0)</div><div class="line"><a name="l00532"></a><span class="lineno"> 532</span>&#160; a7 = <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a287e2fc366c312b468382c95bb90f91f">VLOAD</a>(K0)(0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(input_ptr + 7 * src_stride_y + zin7));</div><div class="line"><a name="l00533"></a><span class="lineno"> 533</span>&#160; BOUNDARY_CONDITION_X(x, a7);</div><div class="line"><a name="l00534"></a><span class="lineno"> 534</span>&#160;<span class="preprocessor">#endif // M0 &gt; 7</span></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">// ---------------------------Transpose and store block -----------------------</span></div><div class="line"><a name="l00537"></a><span class="lineno"> 537</span>&#160;</div><div class="line"><a name="l00538"></a><span class="lineno"> 538</span>&#160; TRANSPOSE_COLUMN_AND_STORE(output_ptr, OUTPUT_STEP_X, 0);</div><div class="line"><a name="l00539"></a><span class="lineno"> 539</span>&#160; TRANSPOSE_COLUMN_AND_STORE(output_ptr, OUTPUT_STEP_X, 1);</div><div class="line"><a name="l00540"></a><span class="lineno"> 540</span>&#160;<span class="preprocessor">#if K0 &gt; 2</span></div><div class="line"><a name="l00541"></a><span class="lineno"> 541</span>&#160; TRANSPOSE_COLUMN_AND_STORE(output_ptr, OUTPUT_STEP_X, 2);</div><div class="line"><a name="l00542"></a><span class="lineno"> 542</span>&#160;<span class="preprocessor">#endif // K0 &gt; 2</span></div><div class="line"><a name="l00543"></a><span class="lineno"> 543</span>&#160;<span class="preprocessor">#if K0 &gt; 3</span></div><div class="line"><a name="l00544"></a><span class="lineno"> 544</span>&#160; TRANSPOSE_COLUMN_AND_STORE(output_ptr, OUTPUT_STEP_X, 3);</div><div class="line"><a name="l00545"></a><span class="lineno"> 545</span>&#160;<span class="preprocessor">#endif // K0 &gt; 3</span></div><div class="line"><a name="l00546"></a><span class="lineno"> 546</span>&#160;<span class="preprocessor">#if K0 &gt; 4</span></div><div class="line"><a name="l00547"></a><span class="lineno"> 547</span>&#160; TRANSPOSE_COLUMN_AND_STORE(output_ptr, OUTPUT_STEP_X, 4);</div><div class="line"><a name="l00548"></a><span class="lineno"> 548</span>&#160; TRANSPOSE_COLUMN_AND_STORE(output_ptr, OUTPUT_STEP_X, 5);</div><div class="line"><a name="l00549"></a><span class="lineno"> 549</span>&#160; TRANSPOSE_COLUMN_AND_STORE(output_ptr, OUTPUT_STEP_X, 6);</div><div class="line"><a name="l00550"></a><span class="lineno"> 550</span>&#160; TRANSPOSE_COLUMN_AND_STORE(output_ptr, OUTPUT_STEP_X, 7);</div><div class="line"><a name="l00551"></a><span class="lineno"> 551</span>&#160;<span class="preprocessor">#endif // K0 &gt; 4</span></div><div class="line"><a name="l00552"></a><span class="lineno"> 552</span>&#160;<span class="preprocessor">#if K0 &gt; 8</span></div><div class="line"><a name="l00553"></a><span class="lineno"> 553</span>&#160; TRANSPOSE_COLUMN_AND_STORE(output_ptr, OUTPUT_STEP_X, 8);</div><div class="line"><a name="l00554"></a><span class="lineno"> 554</span>&#160; TRANSPOSE_COLUMN_AND_STORE(output_ptr, OUTPUT_STEP_X, 9);</div><div class="line"><a name="l00555"></a><span class="lineno"> 555</span>&#160; TRANSPOSE_COLUMN_AND_STORE(output_ptr, OUTPUT_STEP_X, A);</div><div class="line"><a name="l00556"></a><span class="lineno"> 556</span>&#160; TRANSPOSE_COLUMN_AND_STORE(output_ptr, OUTPUT_STEP_X, B);</div><div class="line"><a name="l00557"></a><span class="lineno"> 557</span>&#160; TRANSPOSE_COLUMN_AND_STORE(output_ptr, OUTPUT_STEP_X, C);</div><div class="line"><a name="l00558"></a><span class="lineno"> 558</span>&#160; TRANSPOSE_COLUMN_AND_STORE(output_ptr, OUTPUT_STEP_X, D);</div><div class="line"><a name="l00559"></a><span class="lineno"> 559</span>&#160; TRANSPOSE_COLUMN_AND_STORE(output_ptr, OUTPUT_STEP_X, E);</div><div class="line"><a name="l00560"></a><span class="lineno"> 560</span>&#160; TRANSPOSE_COLUMN_AND_STORE(output_ptr, OUTPUT_STEP_X, F);</div><div class="line"><a name="l00561"></a><span class="lineno"> 561</span>&#160;<span class="preprocessor">#endif // K0 &gt; 8</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;<span class="preprocessor">#undef BLOCK_SIZE</span></div><div class="line"><a name="l00564"></a><span class="lineno"> 564</span>&#160;<span class="preprocessor">#undef OUTPUT_OFFSET_X</span></div><div class="line"><a name="l00565"></a><span class="lineno"> 565</span>&#160;<span class="preprocessor">#undef OUTPUT_STEP_X</span></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">#endif // defined(M0) &amp;&amp; defined(K0) &amp;&amp; defined(V0) &amp;&amp; defined(DATA_TYPE) &amp;&amp; defined(SRC_WIDTH)</span></div><div class="line"><a name="l00568"></a><span class="lineno"> 568</span>&#160;</div><div class="line"><a name="l00569"></a><span class="lineno"> 569</span>&#160;<span class="preprocessor">#if defined(K0) &amp;&amp; defined(N0) &amp;&amp; defined(H0) &amp;&amp; defined(DATA_TYPE) &amp;&amp; defined(SRC_HEIGHT)</span></div><div class="line"><a name="l00570"></a><span class="lineno"> 570</span>&#160;</div><div class="line"><a name="l00600"></a><span class="lineno"> 600</span>&#160;__kernel <span class="keywordtype">void</span> gemm_reshape_rhs_matrix_nt(<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#ae9d2dc29c2789c253406f9b304cc75a8">src</a>),</div><div class="line"><a name="l00601"></a><span class="lineno"> 601</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="l00602"></a><span class="lineno"> 602</span>&#160;{</div><div class="line"><a name="l00603"></a><span class="lineno"> 603</span>&#160; <span class="comment">// Block size</span></div><div class="line"><a name="l00604"></a><span class="lineno"> 604</span>&#160;<span class="preprocessor">#define BLOCK_SIZE ((K0) * (N0))</span></div><div class="line"><a name="l00605"></a><span class="lineno"> 605</span>&#160;</div><div class="line"><a name="l00606"></a><span class="lineno"> 606</span>&#160; <span class="comment">// Output offset X</span></div><div class="line"><a name="l00607"></a><span class="lineno"> 607</span>&#160;<span class="preprocessor">#if defined(INTERLEAVE)</span></div><div class="line"><a name="l00608"></a><span class="lineno"> 608</span>&#160;<span class="preprocessor">#define OUTPUT_OFFSET_X (N0)</span></div><div class="line"><a name="l00609"></a><span class="lineno"> 609</span>&#160;<span class="preprocessor">#else // defined(INTERLEAVE)</span></div><div class="line"><a name="l00610"></a><span class="lineno"> 610</span>&#160;<span class="preprocessor">#define OUTPUT_OFFSET_X (BLOCK_SIZE)</span></div><div class="line"><a name="l00611"></a><span class="lineno"> 611</span>&#160;<span class="preprocessor">#endif // defined(INTERLEAVE)</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="comment">// Output step X</span></div><div class="line"><a name="l00614"></a><span class="lineno"> 614</span>&#160;<span class="preprocessor">#if defined(INTERLEAVE)</span></div><div class="line"><a name="l00615"></a><span class="lineno"> 615</span>&#160;<span class="preprocessor">#define OUTPUT_STEP_X (N0) * (H0)</span></div><div class="line"><a name="l00616"></a><span class="lineno"> 616</span>&#160;<span class="preprocessor">#else // Do not interleave</span></div><div class="line"><a name="l00617"></a><span class="lineno"> 617</span>&#160;<span class="preprocessor">#define OUTPUT_STEP_X (N0)</span></div><div class="line"><a name="l00618"></a><span class="lineno"> 618</span>&#160;<span class="preprocessor">#endif // defined(INTERLEAVE)</span></div><div class="line"><a name="l00619"></a><span class="lineno"> 619</span>&#160;</div><div class="line"><a name="l00620"></a><span class="lineno"> 620</span>&#160; <span class="comment">// Compute source and destination addresses</span></div><div class="line"><a name="l00621"></a><span class="lineno"> 621</span>&#160; uint x = get_global_id(0);</div><div class="line"><a name="l00622"></a><span class="lineno"> 622</span>&#160; uint y = get_global_id(1);</div><div class="line"><a name="l00623"></a><span class="lineno"> 623</span>&#160; uint z = get_global_id(2);</div><div class="line"><a name="l00624"></a><span class="lineno"> 624</span>&#160;</div><div class="line"><a name="l00625"></a><span class="lineno"> 625</span>&#160; <span class="comment">// ------------------ Compute input/output addresses ---------------------------</span></div><div class="line"><a name="l00626"></a><span class="lineno"> 626</span>&#160;</div><div class="line"><a name="l00627"></a><span class="lineno"> 627</span>&#160; <span class="comment">// Compute the input address</span></div><div class="line"><a name="l00628"></a><span class="lineno"> 628</span>&#160; __global uchar *input_ptr = src_ptr + src_offset_first_element_in_bytes + x * (uint)N0 * <span class="keyword">sizeof</span>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>) + y * (uint)K0 * src_stride_y + z * (uint)src_stride_z;</div><div class="line"><a name="l00629"></a><span class="lineno"> 629</span>&#160;</div><div class="line"><a name="l00630"></a><span class="lineno"> 630</span>&#160; <span class="comment">// Compute the output address</span></div><div class="line"><a name="l00631"></a><span class="lineno"> 631</span>&#160; __global uchar *output_ptr = dst_ptr + dst_offset_first_element_in_bytes + (y * (uint)BLOCK_SIZE * (uint)H0 * <span class="keyword">sizeof</span>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>)) + ((x % (uint)H0) * (uint)OUTPUT_OFFSET_X * <span class="keyword">sizeof</span>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>)) + ((</div><div class="line"><a name="l00632"></a><span class="lineno"> 632</span>&#160; x / (uint)H0)</div><div class="line"><a name="l00633"></a><span class="lineno"> 633</span>&#160; * (uint)dst_stride_y)</div><div class="line"><a name="l00634"></a><span class="lineno"> 634</span>&#160; + z * (uint)dst_stride_z;</div><div class="line"><a name="l00635"></a><span class="lineno"> 635</span>&#160;</div><div class="line"><a name="l00636"></a><span class="lineno"> 636</span>&#160; <span class="comment">// ---------------------------Load input values --------------------------------</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; <a class="code" href="repeat_8h.xhtml#a4a61a8df72f9bc1b1dcc4cc4ec6ac3b9">REPEAT_VAR_INIT_TO_CONST</a>(K0, <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, N0), a, 0); </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="comment">// Load values from the RHS matrix</span></div><div class="line"><a name="l00641"></a><span class="lineno"> 641</span>&#160; a0 = <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a287e2fc366c312b468382c95bb90f91f">VLOAD</a>(N0)(0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(input_ptr + 0 * src_stride_y));</div><div class="line"><a name="l00642"></a><span class="lineno"> 642</span>&#160;<span class="preprocessor">#if K0 &gt; 1</span></div><div class="line"><a name="l00643"></a><span class="lineno"> 643</span>&#160; <span class="keywordflow">if</span>(y * (uint)K0 + 1 &lt; SRC_HEIGHT)</div><div class="line"><a name="l00644"></a><span class="lineno"> 644</span>&#160; {</div><div class="line"><a name="l00645"></a><span class="lineno"> 645</span>&#160; a1 = <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a287e2fc366c312b468382c95bb90f91f">VLOAD</a>(N0)(0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(input_ptr + 1 * src_stride_y));</div><div class="line"><a name="l00646"></a><span class="lineno"> 646</span>&#160; }</div><div class="line"><a name="l00647"></a><span class="lineno"> 647</span>&#160;<span class="preprocessor">#endif // K0 &gt; 1</span></div><div class="line"><a name="l00648"></a><span class="lineno"> 648</span>&#160;<span class="preprocessor">#if K0 &gt; 2</span></div><div class="line"><a name="l00649"></a><span class="lineno"> 649</span>&#160; <span class="keywordflow">if</span>(y * (uint)K0 + 2 &lt; SRC_HEIGHT)</div><div class="line"><a name="l00650"></a><span class="lineno"> 650</span>&#160; {</div><div class="line"><a name="l00651"></a><span class="lineno"> 651</span>&#160; a2 = <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a287e2fc366c312b468382c95bb90f91f">VLOAD</a>(N0)(0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(input_ptr + 2 * src_stride_y));</div><div class="line"><a name="l00652"></a><span class="lineno"> 652</span>&#160; }</div><div class="line"><a name="l00653"></a><span class="lineno"> 653</span>&#160;<span class="preprocessor">#endif // K0 &gt; 2</span></div><div class="line"><a name="l00654"></a><span class="lineno"> 654</span>&#160;<span class="preprocessor">#if K0 &gt; 3</span></div><div class="line"><a name="l00655"></a><span class="lineno"> 655</span>&#160; <span class="keywordflow">if</span>(y * (uint)K0 + 3 &lt; SRC_HEIGHT)</div><div class="line"><a name="l00656"></a><span class="lineno"> 656</span>&#160; {</div><div class="line"><a name="l00657"></a><span class="lineno"> 657</span>&#160; a3 = <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a287e2fc366c312b468382c95bb90f91f">VLOAD</a>(N0)(0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(input_ptr + 3 * src_stride_y));</div><div class="line"><a name="l00658"></a><span class="lineno"> 658</span>&#160; }</div><div class="line"><a name="l00659"></a><span class="lineno"> 659</span>&#160;<span class="preprocessor">#endif // K0 &gt; 3</span></div><div class="line"><a name="l00660"></a><span class="lineno"> 660</span>&#160;<span class="preprocessor">#if K0 &gt; 4</span></div><div class="line"><a name="l00661"></a><span class="lineno"> 661</span>&#160; <span class="keywordflow">if</span>(y * (uint)K0 + 4 &lt; SRC_HEIGHT)</div><div class="line"><a name="l00662"></a><span class="lineno"> 662</span>&#160; {</div><div class="line"><a name="l00663"></a><span class="lineno"> 663</span>&#160; a4 = <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a287e2fc366c312b468382c95bb90f91f">VLOAD</a>(N0)(0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(input_ptr + 4 * src_stride_y));</div><div class="line"><a name="l00664"></a><span class="lineno"> 664</span>&#160; }</div><div class="line"><a name="l00665"></a><span class="lineno"> 665</span>&#160; <span class="keywordflow">if</span>(y * (uint)K0 + 5 &lt; SRC_HEIGHT)</div><div class="line"><a name="l00666"></a><span class="lineno"> 666</span>&#160; {</div><div class="line"><a name="l00667"></a><span class="lineno"> 667</span>&#160; a5 = <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a287e2fc366c312b468382c95bb90f91f">VLOAD</a>(N0)(0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(input_ptr + 5 * src_stride_y));</div><div class="line"><a name="l00668"></a><span class="lineno"> 668</span>&#160; }</div><div class="line"><a name="l00669"></a><span class="lineno"> 669</span>&#160; <span class="keywordflow">if</span>(y * (uint)K0 + 6 &lt; SRC_HEIGHT)</div><div class="line"><a name="l00670"></a><span class="lineno"> 670</span>&#160; {</div><div class="line"><a name="l00671"></a><span class="lineno"> 671</span>&#160; a6 = <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a287e2fc366c312b468382c95bb90f91f">VLOAD</a>(N0)(0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(input_ptr + 6 * src_stride_y));</div><div class="line"><a name="l00672"></a><span class="lineno"> 672</span>&#160; }</div><div class="line"><a name="l00673"></a><span class="lineno"> 673</span>&#160; <span class="keywordflow">if</span>(y * (uint)K0 + 7 &lt; SRC_HEIGHT)</div><div class="line"><a name="l00674"></a><span class="lineno"> 674</span>&#160; {</div><div class="line"><a name="l00675"></a><span class="lineno"> 675</span>&#160; a7 = <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a287e2fc366c312b468382c95bb90f91f">VLOAD</a>(N0)(0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(input_ptr + 7 * src_stride_y));</div><div class="line"><a name="l00676"></a><span class="lineno"> 676</span>&#160; }</div><div class="line"><a name="l00677"></a><span class="lineno"> 677</span>&#160;<span class="preprocessor">#endif // K0 &gt; 4</span></div><div class="line"><a name="l00678"></a><span class="lineno"> 678</span>&#160;<span class="preprocessor">#if K0 &gt; 8</span></div><div class="line"><a name="l00679"></a><span class="lineno"> 679</span>&#160; <span class="keywordflow">if</span>(y * (uint)K0 + 8 &lt; SRC_HEIGHT)</div><div class="line"><a name="l00680"></a><span class="lineno"> 680</span>&#160; {</div><div class="line"><a name="l00681"></a><span class="lineno"> 681</span>&#160; a8 = <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a287e2fc366c312b468382c95bb90f91f">VLOAD</a>(N0)(0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(input_ptr + 8 * src_stride_y));</div><div class="line"><a name="l00682"></a><span class="lineno"> 682</span>&#160; }</div><div class="line"><a name="l00683"></a><span class="lineno"> 683</span>&#160; <span class="keywordflow">if</span>(y * (uint)K0 + 9 &lt; SRC_HEIGHT)</div><div class="line"><a name="l00684"></a><span class="lineno"> 684</span>&#160; {</div><div class="line"><a name="l00685"></a><span class="lineno"> 685</span>&#160; a9 = <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a287e2fc366c312b468382c95bb90f91f">VLOAD</a>(N0)(0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(input_ptr + 9 * src_stride_y));</div><div class="line"><a name="l00686"></a><span class="lineno"> 686</span>&#160; }</div><div class="line"><a name="l00687"></a><span class="lineno"> 687</span>&#160; <span class="keywordflow">if</span>(y * (uint)K0 + 10 &lt; SRC_HEIGHT)</div><div class="line"><a name="l00688"></a><span class="lineno"> 688</span>&#160; {</div><div class="line"><a name="l00689"></a><span class="lineno"> 689</span>&#160; aA = <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a287e2fc366c312b468382c95bb90f91f">VLOAD</a>(N0)(0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(input_ptr + 10 * src_stride_y));</div><div class="line"><a name="l00690"></a><span class="lineno"> 690</span>&#160; }</div><div class="line"><a name="l00691"></a><span class="lineno"> 691</span>&#160; <span class="keywordflow">if</span>(y * (uint)K0 + 11 &lt; SRC_HEIGHT)</div><div class="line"><a name="l00692"></a><span class="lineno"> 692</span>&#160; {</div><div class="line"><a name="l00693"></a><span class="lineno"> 693</span>&#160; aB = <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a287e2fc366c312b468382c95bb90f91f">VLOAD</a>(N0)(0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(input_ptr + 11 * src_stride_y));</div><div class="line"><a name="l00694"></a><span class="lineno"> 694</span>&#160; }</div><div class="line"><a name="l00695"></a><span class="lineno"> 695</span>&#160; <span class="keywordflow">if</span>(y * (uint)K0 + 12 &lt; SRC_HEIGHT)</div><div class="line"><a name="l00696"></a><span class="lineno"> 696</span>&#160; {</div><div class="line"><a name="l00697"></a><span class="lineno"> 697</span>&#160; aC = <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a287e2fc366c312b468382c95bb90f91f">VLOAD</a>(N0)(0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(input_ptr + 12 * src_stride_y));</div><div class="line"><a name="l00698"></a><span class="lineno"> 698</span>&#160; }</div><div class="line"><a name="l00699"></a><span class="lineno"> 699</span>&#160; <span class="keywordflow">if</span>(y * (uint)K0 + 13 &lt; SRC_HEIGHT)</div><div class="line"><a name="l00700"></a><span class="lineno"> 700</span>&#160; {</div><div class="line"><a name="l00701"></a><span class="lineno"> 701</span>&#160; aD = <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a287e2fc366c312b468382c95bb90f91f">VLOAD</a>(N0)(0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(input_ptr + 13 * src_stride_y));</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; <span class="keywordflow">if</span>(y * (uint)K0 + 14 &lt; SRC_HEIGHT)</div><div class="line"><a name="l00704"></a><span class="lineno"> 704</span>&#160; {</div><div class="line"><a name="l00705"></a><span class="lineno"> 705</span>&#160; aE = <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a287e2fc366c312b468382c95bb90f91f">VLOAD</a>(N0)(0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(input_ptr + 14 * src_stride_y));</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="keywordflow">if</span>(y * (uint)K0 + 15 &lt; SRC_HEIGHT)</div><div class="line"><a name="l00708"></a><span class="lineno"> 708</span>&#160; {</div><div class="line"><a name="l00709"></a><span class="lineno"> 709</span>&#160; aF = <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a287e2fc366c312b468382c95bb90f91f">VLOAD</a>(N0)(0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(input_ptr + 15 * src_stride_y));</div><div class="line"><a name="l00710"></a><span class="lineno"> 710</span>&#160; }</div><div class="line"><a name="l00711"></a><span class="lineno"> 711</span>&#160;<span class="preprocessor">#endif // K0 &gt; 8</span></div><div class="line"><a name="l00712"></a><span class="lineno"> 712</span>&#160;</div><div class="line"><a name="l00713"></a><span class="lineno"> 713</span>&#160; <span class="comment">// ---------------------------Store output values ------------------------------</span></div><div class="line"><a name="l00714"></a><span class="lineno"> 714</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#acb282042d1edeeaa3cc979a206f78b54">VSTORE</a>(N0)</div><div class="line"><a name="l00715"></a><span class="lineno"> 715</span>&#160; (a0, 0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(output_ptr + 0 * OUTPUT_STEP_X * <span class="keyword">sizeof</span>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>)));</div><div class="line"><a name="l00716"></a><span class="lineno"> 716</span>&#160;<span class="preprocessor">#if K0 &gt; 1</span></div><div class="line"><a name="l00717"></a><span class="lineno"> 717</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#acb282042d1edeeaa3cc979a206f78b54">VSTORE</a>(N0)</div><div class="line"><a name="l00718"></a><span class="lineno"> 718</span>&#160; (a1, 0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(output_ptr + 1 * OUTPUT_STEP_X * <span class="keyword">sizeof</span>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>)));</div><div class="line"><a name="l00719"></a><span class="lineno"> 719</span>&#160;<span class="preprocessor">#endif // K0 &gt; 1</span></div><div class="line"><a name="l00720"></a><span class="lineno"> 720</span>&#160;<span class="preprocessor">#if K0 &gt; 2</span></div><div class="line"><a name="l00721"></a><span class="lineno"> 721</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#acb282042d1edeeaa3cc979a206f78b54">VSTORE</a>(N0)</div><div class="line"><a name="l00722"></a><span class="lineno"> 722</span>&#160; (a2, 0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(output_ptr + 2 * OUTPUT_STEP_X * <span class="keyword">sizeof</span>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>)));</div><div class="line"><a name="l00723"></a><span class="lineno"> 723</span>&#160;<span class="preprocessor">#endif // K0 &gt; 2</span></div><div class="line"><a name="l00724"></a><span class="lineno"> 724</span>&#160;<span class="preprocessor">#if K0 &gt; 3</span></div><div class="line"><a name="l00725"></a><span class="lineno"> 725</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#acb282042d1edeeaa3cc979a206f78b54">VSTORE</a>(N0)</div><div class="line"><a name="l00726"></a><span class="lineno"> 726</span>&#160; (a3, 0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(output_ptr + 3 * OUTPUT_STEP_X * <span class="keyword">sizeof</span>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>)));</div><div class="line"><a name="l00727"></a><span class="lineno"> 727</span>&#160;<span class="preprocessor">#endif // K0 &gt; 3</span></div><div class="line"><a name="l00728"></a><span class="lineno"> 728</span>&#160;<span class="preprocessor">#if K0 &gt; 4</span></div><div class="line"><a name="l00729"></a><span class="lineno"> 729</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#acb282042d1edeeaa3cc979a206f78b54">VSTORE</a>(N0)</div><div class="line"><a name="l00730"></a><span class="lineno"> 730</span>&#160; (a4, 0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(output_ptr + 4 * OUTPUT_STEP_X * <span class="keyword">sizeof</span>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>)));</div><div class="line"><a name="l00731"></a><span class="lineno"> 731</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#acb282042d1edeeaa3cc979a206f78b54">VSTORE</a>(N0)</div><div class="line"><a name="l00732"></a><span class="lineno"> 732</span>&#160; (a5, 0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(output_ptr + 5 * OUTPUT_STEP_X * <span class="keyword">sizeof</span>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>)));</div><div class="line"><a name="l00733"></a><span class="lineno"> 733</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#acb282042d1edeeaa3cc979a206f78b54">VSTORE</a>(N0)</div><div class="line"><a name="l00734"></a><span class="lineno"> 734</span>&#160; (a6, 0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(output_ptr + 6 * OUTPUT_STEP_X * <span class="keyword">sizeof</span>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>)));</div><div class="line"><a name="l00735"></a><span class="lineno"> 735</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#acb282042d1edeeaa3cc979a206f78b54">VSTORE</a>(N0)</div><div class="line"><a name="l00736"></a><span class="lineno"> 736</span>&#160; (a7, 0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(output_ptr + 7 * OUTPUT_STEP_X * <span class="keyword">sizeof</span>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>)));</div><div class="line"><a name="l00737"></a><span class="lineno"> 737</span>&#160;<span class="preprocessor">#endif // N0 &gt; 4</span></div><div class="line"><a name="l00738"></a><span class="lineno"> 738</span>&#160;<span class="preprocessor">#if K0 &gt; 8</span></div><div class="line"><a name="l00739"></a><span class="lineno"> 739</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#acb282042d1edeeaa3cc979a206f78b54">VSTORE</a>(N0)</div><div class="line"><a name="l00740"></a><span class="lineno"> 740</span>&#160; (a8, 0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(output_ptr + 8 * OUTPUT_STEP_X * <span class="keyword">sizeof</span>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>)));</div><div class="line"><a name="l00741"></a><span class="lineno"> 741</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#acb282042d1edeeaa3cc979a206f78b54">VSTORE</a>(N0)</div><div class="line"><a name="l00742"></a><span class="lineno"> 742</span>&#160; (a9, 0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(output_ptr + 9 * OUTPUT_STEP_X * <span class="keyword">sizeof</span>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>)));</div><div class="line"><a name="l00743"></a><span class="lineno"> 743</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#acb282042d1edeeaa3cc979a206f78b54">VSTORE</a>(N0)</div><div class="line"><a name="l00744"></a><span class="lineno"> 744</span>&#160; (aA, 0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(output_ptr + 10 * OUTPUT_STEP_X * <span class="keyword">sizeof</span>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>)));</div><div class="line"><a name="l00745"></a><span class="lineno"> 745</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#acb282042d1edeeaa3cc979a206f78b54">VSTORE</a>(N0)</div><div class="line"><a name="l00746"></a><span class="lineno"> 746</span>&#160; (aB, 0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(output_ptr + 11 * OUTPUT_STEP_X * <span class="keyword">sizeof</span>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>)));</div><div class="line"><a name="l00747"></a><span class="lineno"> 747</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#acb282042d1edeeaa3cc979a206f78b54">VSTORE</a>(N0)</div><div class="line"><a name="l00748"></a><span class="lineno"> 748</span>&#160; (aC, 0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(output_ptr + 12 * OUTPUT_STEP_X * <span class="keyword">sizeof</span>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>)));</div><div class="line"><a name="l00749"></a><span class="lineno"> 749</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#acb282042d1edeeaa3cc979a206f78b54">VSTORE</a>(N0)</div><div class="line"><a name="l00750"></a><span class="lineno"> 750</span>&#160; (aD, 0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(output_ptr + 13 * OUTPUT_STEP_X * <span class="keyword">sizeof</span>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>)));</div><div class="line"><a name="l00751"></a><span class="lineno"> 751</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#acb282042d1edeeaa3cc979a206f78b54">VSTORE</a>(N0)</div><div class="line"><a name="l00752"></a><span class="lineno"> 752</span>&#160; (aE, 0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(output_ptr + 14 * OUTPUT_STEP_X * <span class="keyword">sizeof</span>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>)));</div><div class="line"><a name="l00753"></a><span class="lineno"> 753</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#acb282042d1edeeaa3cc979a206f78b54">VSTORE</a>(N0)</div><div class="line"><a name="l00754"></a><span class="lineno"> 754</span>&#160; (aF, 0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(output_ptr + 15 * OUTPUT_STEP_X * <span class="keyword">sizeof</span>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>)));</div><div class="line"><a name="l00755"></a><span class="lineno"> 755</span>&#160;<span class="preprocessor">#endif // N0 &gt; 8</span></div><div class="line"><a name="l00756"></a><span class="lineno"> 756</span>&#160;</div><div class="line"><a name="l00757"></a><span class="lineno"> 757</span>&#160;<span class="preprocessor">#undef BLOCK_SIZE</span></div><div class="line"><a name="l00758"></a><span class="lineno"> 758</span>&#160;<span class="preprocessor">#undef OUTPUT_OFFSET_X</span></div><div class="line"><a name="l00759"></a><span class="lineno"> 759</span>&#160;<span class="preprocessor">#undef OUTPUT_STEP_X</span></div><div class="line"><a name="l00760"></a><span class="lineno"> 760</span>&#160;}</div><div class="line"><a name="l00761"></a><span class="lineno"> 761</span>&#160;</div><div class="line"><a name="l00762"></a><span class="lineno"> 762</span>&#160;<span class="preprocessor">#if defined(TRANSPOSE)</span></div><div class="line"><a name="l00763"></a><span class="lineno"> 763</span>&#160;</div><div class="line"><a name="l00794"></a><span class="lineno"> 794</span>&#160;__kernel <span class="keywordtype">void</span> gemm_reshape_rhs_matrix_t(<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#ae9d2dc29c2789c253406f9b304cc75a8">src</a>),</div><div class="line"><a name="l00795"></a><span class="lineno"> 795</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="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">// Block size</span></div><div class="line"><a name="l00798"></a><span class="lineno"> 798</span>&#160;<span class="preprocessor">#define BLOCK_SIZE ((K0) * (N0))</span></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="comment">// Output offset X</span></div><div class="line"><a name="l00801"></a><span class="lineno"> 801</span>&#160;<span class="preprocessor">#if defined(INTERLEAVE)</span></div><div class="line"><a name="l00802"></a><span class="lineno"> 802</span>&#160;<span class="preprocessor">#define OUTPUT_OFFSET_X (K0)</span></div><div class="line"><a name="l00803"></a><span class="lineno"> 803</span>&#160;<span class="preprocessor">#else // defined(INTERLEAVE)</span></div><div class="line"><a name="l00804"></a><span class="lineno"> 804</span>&#160;<span class="preprocessor">#define OUTPUT_OFFSET_X (BLOCK_SIZE)</span></div><div class="line"><a name="l00805"></a><span class="lineno"> 805</span>&#160;<span class="preprocessor">#endif // defined(INTERLEAVE)</span></div><div class="line"><a name="l00806"></a><span class="lineno"> 806</span>&#160;</div><div class="line"><a name="l00807"></a><span class="lineno"> 807</span>&#160; <span class="comment">// Output step X</span></div><div class="line"><a name="l00808"></a><span class="lineno"> 808</span>&#160;<span class="preprocessor">#if defined(INTERLEAVE)</span></div><div class="line"><a name="l00809"></a><span class="lineno"> 809</span>&#160;<span class="preprocessor">#define OUTPUT_STEP_X (K0) * (H0)</span></div><div class="line"><a name="l00810"></a><span class="lineno"> 810</span>&#160;<span class="preprocessor">#else // Do not interleave</span></div><div class="line"><a name="l00811"></a><span class="lineno"> 811</span>&#160;<span class="preprocessor">#define OUTPUT_STEP_X (K0)</span></div><div class="line"><a name="l00812"></a><span class="lineno"> 812</span>&#160;<span class="preprocessor">#endif // defined(INTERLEAVE)</span></div><div class="line"><a name="l00813"></a><span class="lineno"> 813</span>&#160;</div><div class="line"><a name="l00814"></a><span class="lineno"> 814</span>&#160; <span class="comment">// Compute source and destination addresses</span></div><div class="line"><a name="l00815"></a><span class="lineno"> 815</span>&#160; uint x = get_global_id(0);</div><div class="line"><a name="l00816"></a><span class="lineno"> 816</span>&#160; uint y = get_global_id(1);</div><div class="line"><a name="l00817"></a><span class="lineno"> 817</span>&#160; uint z = get_global_id(2);</div><div class="line"><a name="l00818"></a><span class="lineno"> 818</span>&#160;</div><div class="line"><a name="l00819"></a><span class="lineno"> 819</span>&#160; <span class="comment">// ------------------ Compute input/output addresses ---------------------------</span></div><div class="line"><a name="l00820"></a><span class="lineno"> 820</span>&#160;</div><div class="line"><a name="l00821"></a><span class="lineno"> 821</span>&#160; <span class="comment">// Compute the input address</span></div><div class="line"><a name="l00822"></a><span class="lineno"> 822</span>&#160; __global uchar *input_ptr = src_ptr + src_offset_first_element_in_bytes + x * (uint)N0 * <span class="keyword">sizeof</span>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>) + y * (uint)K0 * src_stride_y + z * (uint)src_stride_z;</div><div class="line"><a name="l00823"></a><span class="lineno"> 823</span>&#160;</div><div class="line"><a name="l00824"></a><span class="lineno"> 824</span>&#160; <span class="comment">// Compute the output address</span></div><div class="line"><a name="l00825"></a><span class="lineno"> 825</span>&#160; __global uchar *output_ptr = dst_ptr + dst_offset_first_element_in_bytes + (y * (uint)BLOCK_SIZE * (uint)H0 * <span class="keyword">sizeof</span>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>)) + ((x % H0) * (uint)OUTPUT_OFFSET_X * <span class="keyword">sizeof</span>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>)) + ((x /</div><div class="line"><a name="l00826"></a><span class="lineno"> 826</span>&#160; (uint)H0) * (uint)dst_stride_y) + z * (uint)dst_stride_z;</div><div class="line"><a name="l00827"></a><span class="lineno"> 827</span>&#160;</div><div class="line"><a name="l00828"></a><span class="lineno"> 828</span>&#160; <span class="comment">// ---------------------------Load input values --------------------------------</span></div><div class="line"><a name="l00829"></a><span class="lineno"> 829</span>&#160; <a class="code" href="repeat_8h.xhtml#a4a61a8df72f9bc1b1dcc4cc4ec6ac3b9">REPEAT_VAR_INIT_TO_CONST</a>(K0, <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, N0), a, 0); <span class="comment">//VEC_DATA_TYPE(DATA_TYPE, N0) a0=0, a1=0, ... a(K0-1)=0;</span></div><div class="line"><a name="l00830"></a><span class="lineno"> 830</span>&#160;</div><div class="line"><a name="l00831"></a><span class="lineno"> 831</span>&#160; <span class="comment">// Load values from the RHS matrix</span></div><div class="line"><a name="l00832"></a><span class="lineno"> 832</span>&#160; a0 = <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a287e2fc366c312b468382c95bb90f91f">VLOAD</a>(N0)(0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(input_ptr + 0 * src_stride_y));</div><div class="line"><a name="l00833"></a><span class="lineno"> 833</span>&#160; <span class="keywordflow">if</span>(y * (uint)K0 + 1 &lt; SRC_HEIGHT)</div><div class="line"><a name="l00834"></a><span class="lineno"> 834</span>&#160; {</div><div class="line"><a name="l00835"></a><span class="lineno"> 835</span>&#160; a1 = <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a287e2fc366c312b468382c95bb90f91f">VLOAD</a>(N0)(0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(input_ptr + 1 * src_stride_y));</div><div class="line"><a name="l00836"></a><span class="lineno"> 836</span>&#160; }</div><div class="line"><a name="l00837"></a><span class="lineno"> 837</span>&#160;<span class="preprocessor">#if K0 &gt; 2</span></div><div class="line"><a name="l00838"></a><span class="lineno"> 838</span>&#160; <span class="keywordflow">if</span>(y * (uint)K0 + 2 &lt; SRC_HEIGHT)</div><div class="line"><a name="l00839"></a><span class="lineno"> 839</span>&#160; {</div><div class="line"><a name="l00840"></a><span class="lineno"> 840</span>&#160; a2 = <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a287e2fc366c312b468382c95bb90f91f">VLOAD</a>(N0)(0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(input_ptr + 2 * src_stride_y));</div><div class="line"><a name="l00841"></a><span class="lineno"> 841</span>&#160; }</div><div class="line"><a name="l00842"></a><span class="lineno"> 842</span>&#160;<span class="preprocessor">#endif // K0 &gt; 2</span></div><div class="line"><a name="l00843"></a><span class="lineno"> 843</span>&#160;<span class="preprocessor">#if K0 &gt; 3</span></div><div class="line"><a name="l00844"></a><span class="lineno"> 844</span>&#160; <span class="keywordflow">if</span>(y * (uint)K0 + 3 &lt; SRC_HEIGHT)</div><div class="line"><a name="l00845"></a><span class="lineno"> 845</span>&#160; {</div><div class="line"><a name="l00846"></a><span class="lineno"> 846</span>&#160; a3 = <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a287e2fc366c312b468382c95bb90f91f">VLOAD</a>(N0)(0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(input_ptr + 3 * src_stride_y));</div><div class="line"><a name="l00847"></a><span class="lineno"> 847</span>&#160; }</div><div class="line"><a name="l00848"></a><span class="lineno"> 848</span>&#160;<span class="preprocessor">#endif // K0 &gt; 3</span></div><div class="line"><a name="l00849"></a><span class="lineno"> 849</span>&#160;<span class="preprocessor">#if K0 &gt; 4</span></div><div class="line"><a name="l00850"></a><span class="lineno"> 850</span>&#160; <span class="keywordflow">if</span>(y * (uint)K0 + 4 &lt; SRC_HEIGHT)</div><div class="line"><a name="l00851"></a><span class="lineno"> 851</span>&#160; {</div><div class="line"><a name="l00852"></a><span class="lineno"> 852</span>&#160; a4 = <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a287e2fc366c312b468382c95bb90f91f">VLOAD</a>(N0)(0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(input_ptr + 4 * src_stride_y));</div><div class="line"><a name="l00853"></a><span class="lineno"> 853</span>&#160; }</div><div class="line"><a name="l00854"></a><span class="lineno"> 854</span>&#160; <span class="keywordflow">if</span>(y * (uint)K0 + 5 &lt; SRC_HEIGHT)</div><div class="line"><a name="l00855"></a><span class="lineno"> 855</span>&#160; {</div><div class="line"><a name="l00856"></a><span class="lineno"> 856</span>&#160; a5 = <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a287e2fc366c312b468382c95bb90f91f">VLOAD</a>(N0)(0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(input_ptr + 5 * src_stride_y));</div><div class="line"><a name="l00857"></a><span class="lineno"> 857</span>&#160; }</div><div class="line"><a name="l00858"></a><span class="lineno"> 858</span>&#160; <span class="keywordflow">if</span>(y * (uint)K0 + 6 &lt; SRC_HEIGHT)</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; a6 = <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a287e2fc366c312b468382c95bb90f91f">VLOAD</a>(N0)(0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(input_ptr + 6 * src_stride_y));</div><div class="line"><a name="l00861"></a><span class="lineno"> 861</span>&#160; }</div><div class="line"><a name="l00862"></a><span class="lineno"> 862</span>&#160; <span class="keywordflow">if</span>(y * (uint)K0 + 7 &lt; SRC_HEIGHT)</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; a7 = <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a287e2fc366c312b468382c95bb90f91f">VLOAD</a>(N0)(0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(input_ptr + 7 * src_stride_y));</div><div class="line"><a name="l00865"></a><span class="lineno"> 865</span>&#160; }</div><div class="line"><a name="l00866"></a><span class="lineno"> 866</span>&#160;<span class="preprocessor">#endif // K0 &gt; 4</span></div><div class="line"><a name="l00867"></a><span class="lineno"> 867</span>&#160;<span class="preprocessor">#if K0 &gt; 8</span></div><div class="line"><a name="l00868"></a><span class="lineno"> 868</span>&#160; <span class="keywordflow">if</span>(y * (uint)K0 + 8 &lt; SRC_HEIGHT)</div><div class="line"><a name="l00869"></a><span class="lineno"> 869</span>&#160; {</div><div class="line"><a name="l00870"></a><span class="lineno"> 870</span>&#160; a8 = <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a287e2fc366c312b468382c95bb90f91f">VLOAD</a>(N0)(0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(input_ptr + 8 * src_stride_y));</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="keywordflow">if</span>(y * (uint)K0 + 9 &lt; SRC_HEIGHT)</div><div class="line"><a name="l00873"></a><span class="lineno"> 873</span>&#160; {</div><div class="line"><a name="l00874"></a><span class="lineno"> 874</span>&#160; a9 = <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a287e2fc366c312b468382c95bb90f91f">VLOAD</a>(N0)(0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(input_ptr + 9 * src_stride_y));</div><div class="line"><a name="l00875"></a><span class="lineno"> 875</span>&#160; }</div><div class="line"><a name="l00876"></a><span class="lineno"> 876</span>&#160; <span class="keywordflow">if</span>(y * (uint)K0 + 10 &lt; SRC_HEIGHT)</div><div class="line"><a name="l00877"></a><span class="lineno"> 877</span>&#160; {</div><div class="line"><a name="l00878"></a><span class="lineno"> 878</span>&#160; aA = <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a287e2fc366c312b468382c95bb90f91f">VLOAD</a>(N0)(0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(input_ptr + 10 * src_stride_y));</div><div class="line"><a name="l00879"></a><span class="lineno"> 879</span>&#160; }</div><div class="line"><a name="l00880"></a><span class="lineno"> 880</span>&#160; <span class="keywordflow">if</span>(y * (uint)K0 + 11 &lt; SRC_HEIGHT)</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; aB = <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a287e2fc366c312b468382c95bb90f91f">VLOAD</a>(N0)(0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(input_ptr + 11 * src_stride_y));</div><div class="line"><a name="l00883"></a><span class="lineno"> 883</span>&#160; }</div><div class="line"><a name="l00884"></a><span class="lineno"> 884</span>&#160; <span class="keywordflow">if</span>(y * (uint)K0 + 12 &lt; SRC_HEIGHT)</div><div class="line"><a name="l00885"></a><span class="lineno"> 885</span>&#160; {</div><div class="line"><a name="l00886"></a><span class="lineno"> 886</span>&#160; aC = <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a287e2fc366c312b468382c95bb90f91f">VLOAD</a>(N0)(0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(input_ptr + 12 * src_stride_y));</div><div class="line"><a name="l00887"></a><span class="lineno"> 887</span>&#160; }</div><div class="line"><a name="l00888"></a><span class="lineno"> 888</span>&#160; <span class="keywordflow">if</span>(y * (uint)K0 + 13 &lt; SRC_HEIGHT)</div><div class="line"><a name="l00889"></a><span class="lineno"> 889</span>&#160; {</div><div class="line"><a name="l00890"></a><span class="lineno"> 890</span>&#160; aD = <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a287e2fc366c312b468382c95bb90f91f">VLOAD</a>(N0)(0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(input_ptr + 13 * src_stride_y));</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; <span class="keywordflow">if</span>(y * (uint)K0 + 14 &lt; SRC_HEIGHT)</div><div class="line"><a name="l00893"></a><span class="lineno"> 893</span>&#160; {</div><div class="line"><a name="l00894"></a><span class="lineno"> 894</span>&#160; aE = <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a287e2fc366c312b468382c95bb90f91f">VLOAD</a>(N0)(0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(input_ptr + 14 * src_stride_y));</div><div class="line"><a name="l00895"></a><span class="lineno"> 895</span>&#160; }</div><div class="line"><a name="l00896"></a><span class="lineno"> 896</span>&#160; <span class="keywordflow">if</span>(y * (uint)K0 + 15 &lt; SRC_HEIGHT)</div><div class="line"><a name="l00897"></a><span class="lineno"> 897</span>&#160; {</div><div class="line"><a name="l00898"></a><span class="lineno"> 898</span>&#160; aF = <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a287e2fc366c312b468382c95bb90f91f">VLOAD</a>(N0)(0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(input_ptr + 15 * src_stride_y));</div><div class="line"><a name="l00899"></a><span class="lineno"> 899</span>&#160; }</div><div class="line"><a name="l00900"></a><span class="lineno"> 900</span>&#160;<span class="preprocessor">#endif // K0 &gt; 8</span></div><div class="line"><a name="l00901"></a><span class="lineno"> 901</span>&#160;</div><div class="line"><a name="l00902"></a><span class="lineno"> 902</span>&#160; <span class="comment">// ---------------------------Transpose the block ------------------------------</span></div><div class="line"><a name="l00903"></a><span class="lineno"> 903</span>&#160; <a class="code" href="repeat_8h.xhtml#a4a61a8df72f9bc1b1dcc4cc4ec6ac3b9">REPEAT_VAR_INIT_TO_CONST</a>(N0, <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, K0), res, 0); <span class="comment">//VEC_DATA_TYPE(DATA_TYPE, K0) res0=0, res1=0, res2=0,... res(N0-1)=0;</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="preprocessor">#if K0 == 2</span></div><div class="line"><a name="l00906"></a><span class="lineno"> 906</span>&#160; <span class="comment">// This part computes the following transpositions:</span></div><div class="line"><a name="l00907"></a><span class="lineno"> 907</span>&#160; <span class="comment">// 2x2 -&gt; 2x2</span></div><div class="line"><a name="l00908"></a><span class="lineno"> 908</span>&#160; <span class="comment">// 2x4 -&gt; 4x2</span></div><div class="line"><a name="l00909"></a><span class="lineno"> 909</span>&#160; <span class="comment">// 2x8 -&gt; 8x2</span></div><div class="line"><a name="l00910"></a><span class="lineno"> 910</span>&#160; <span class="comment">// 2x16 -&gt; 16x2</span></div><div class="line"><a name="l00911"></a><span class="lineno"> 911</span>&#160; res0 = (<a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, K0))(a0.s0, a1.s0);</div><div class="line"><a name="l00912"></a><span class="lineno"> 912</span>&#160; res1 = (<a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, K0))(a0.s1, a1.s1);</div><div class="line"><a name="l00913"></a><span class="lineno"> 913</span>&#160;<span class="preprocessor">#if N0 &gt; 2</span></div><div class="line"><a name="l00914"></a><span class="lineno"> 914</span>&#160; res2 = (<a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, K0))(a0.s2, a1.s2);</div><div class="line"><a name="l00915"></a><span class="lineno"> 915</span>&#160;<span class="preprocessor">#endif // N0 &gt; 2</span></div><div class="line"><a name="l00916"></a><span class="lineno"> 916</span>&#160;<span class="preprocessor">#if N0 &gt; 3</span></div><div class="line"><a name="l00917"></a><span class="lineno"> 917</span>&#160; res3 = (<a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, K0))(a0.s3, a1.s3);</div><div class="line"><a name="l00918"></a><span class="lineno"> 918</span>&#160;<span class="preprocessor">#endif // N0 &gt; 3</span></div><div class="line"><a name="l00919"></a><span class="lineno"> 919</span>&#160;<span class="preprocessor">#if N0 &gt; 4</span></div><div class="line"><a name="l00920"></a><span class="lineno"> 920</span>&#160; res4 = (<a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, K0))(a0.s4, a1.s4);</div><div class="line"><a name="l00921"></a><span class="lineno"> 921</span>&#160; res5 = (<a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, K0))(a0.s5, a1.s5);</div><div class="line"><a name="l00922"></a><span class="lineno"> 922</span>&#160; res6 = (<a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, K0))(a0.s6, a1.s6);</div><div class="line"><a name="l00923"></a><span class="lineno"> 923</span>&#160; res7 = (<a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, K0))(a0.s7, a1.s7);</div><div class="line"><a name="l00924"></a><span class="lineno"> 924</span>&#160;<span class="preprocessor">#endif // N0 &gt; 4</span></div><div class="line"><a name="l00925"></a><span class="lineno"> 925</span>&#160;<span class="preprocessor">#if N0 &gt; 8</span></div><div class="line"><a name="l00926"></a><span class="lineno"> 926</span>&#160; res8 = (<a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, K0))(a0.s8, a1.s8);</div><div class="line"><a name="l00927"></a><span class="lineno"> 927</span>&#160; res9 = (<a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, K0))(a0.s9, a1.s9);</div><div class="line"><a name="l00928"></a><span class="lineno"> 928</span>&#160; resA = (<a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, K0))(a0.sA, a1.sA);</div><div class="line"><a name="l00929"></a><span class="lineno"> 929</span>&#160; resB = (<a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, K0))(a0.sB, a1.sB);</div><div class="line"><a name="l00930"></a><span class="lineno"> 930</span>&#160; resC = (<a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, K0))(a0.sC, a1.sC);</div><div class="line"><a name="l00931"></a><span class="lineno"> 931</span>&#160; resD = (<a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, K0))(a0.sD, a1.sD);</div><div class="line"><a name="l00932"></a><span class="lineno"> 932</span>&#160; resE = (<a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, K0))(a0.sE, a1.sE);</div><div class="line"><a name="l00933"></a><span class="lineno"> 933</span>&#160; resF = (<a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, K0))(a0.sF, a1.sF);</div><div class="line"><a name="l00934"></a><span class="lineno"> 934</span>&#160;<span class="preprocessor">#endif // N0 &gt; 8</span></div><div class="line"><a name="l00935"></a><span class="lineno"> 935</span>&#160;</div><div class="line"><a name="l00936"></a><span class="lineno"> 936</span>&#160;<span class="preprocessor">#elif K0 == 3 // K0 == 2</span></div><div class="line"><a name="l00937"></a><span class="lineno"> 937</span>&#160; <span class="comment">// This part computes the following transpositions:</span></div><div class="line"><a name="l00938"></a><span class="lineno"> 938</span>&#160; <span class="comment">// 3x2 -&gt; 2x3</span></div><div class="line"><a name="l00939"></a><span class="lineno"> 939</span>&#160; <span class="comment">// 3x4 -&gt; 4x3</span></div><div class="line"><a name="l00940"></a><span class="lineno"> 940</span>&#160; <span class="comment">// 3x8 -&gt; 8x3</span></div><div class="line"><a name="l00941"></a><span class="lineno"> 941</span>&#160; <span class="comment">// 3x16 -&gt; 16x3</span></div><div class="line"><a name="l00942"></a><span class="lineno"> 942</span>&#160; res0 = (<a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, K0))(a0.s0, a1.s0, a2.s0);</div><div class="line"><a name="l00943"></a><span class="lineno"> 943</span>&#160; res1 = (<a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, K0))(a0.s1, a1.s1, a2.s1);</div><div class="line"><a name="l00944"></a><span class="lineno"> 944</span>&#160;<span class="preprocessor">#if N0 &gt; 2</span></div><div class="line"><a name="l00945"></a><span class="lineno"> 945</span>&#160; res2 = (<a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, K0))(a0.s2, a1.s2, a2.s2);</div><div class="line"><a name="l00946"></a><span class="lineno"> 946</span>&#160;<span class="preprocessor">#endif // N0 &gt; 2</span></div><div class="line"><a name="l00947"></a><span class="lineno"> 947</span>&#160;<span class="preprocessor">#if N0 &gt; 3</span></div><div class="line"><a name="l00948"></a><span class="lineno"> 948</span>&#160; res3 = (<a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, K0))(a0.s3, a1.s3, a2.s3);</div><div class="line"><a name="l00949"></a><span class="lineno"> 949</span>&#160;<span class="preprocessor">#endif // N0 &gt; 3</span></div><div class="line"><a name="l00950"></a><span class="lineno"> 950</span>&#160;<span class="preprocessor">#if N0 &gt; 4</span></div><div class="line"><a name="l00951"></a><span class="lineno"> 951</span>&#160; res4 = (<a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, K0))(a0.s4, a1.s4, a2.s4);</div><div class="line"><a name="l00952"></a><span class="lineno"> 952</span>&#160; res5 = (<a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, K0))(a0.s5, a1.s5, a2.s5);</div><div class="line"><a name="l00953"></a><span class="lineno"> 953</span>&#160; res6 = (<a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, K0))(a0.s6, a1.s6, a2.s6);</div><div class="line"><a name="l00954"></a><span class="lineno"> 954</span>&#160; res7 = (<a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, K0))(a0.s7, a1.s7, a2.s7);</div><div class="line"><a name="l00955"></a><span class="lineno"> 955</span>&#160;<span class="preprocessor">#endif // N0 &gt; 4</span></div><div class="line"><a name="l00956"></a><span class="lineno"> 956</span>&#160;<span class="preprocessor">#if N0 &gt; 8</span></div><div class="line"><a name="l00957"></a><span class="lineno"> 957</span>&#160; res8 = (<a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, K0))(a0.s8, a1.s8, a2.s8);</div><div class="line"><a name="l00958"></a><span class="lineno"> 958</span>&#160; res9 = (<a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, K0))(a0.s9, a1.s9, a2.s9);</div><div class="line"><a name="l00959"></a><span class="lineno"> 959</span>&#160; resA = (<a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, K0))(a0.sA, a1.sA, a2.sA);</div><div class="line"><a name="l00960"></a><span class="lineno"> 960</span>&#160; resB = (<a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, K0))(a0.sB, a1.sB, a2.sB);</div><div class="line"><a name="l00961"></a><span class="lineno"> 961</span>&#160; resC = (<a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, K0))(a0.sC, a1.sC, a2.sC);</div><div class="line"><a name="l00962"></a><span class="lineno"> 962</span>&#160; resD = (<a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, K0))(a0.sD, a1.sD, a2.sD);</div><div class="line"><a name="l00963"></a><span class="lineno"> 963</span>&#160; resE = (<a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, K0))(a0.sE, a1.sE, a2.sE);</div><div class="line"><a name="l00964"></a><span class="lineno"> 964</span>&#160; resF = (<a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, K0))(a0.sF, a1.sF, a2.sF);</div><div class="line"><a name="l00965"></a><span class="lineno"> 965</span>&#160;<span class="preprocessor">#endif // N0 &gt; 8</span></div><div class="line"><a name="l00966"></a><span class="lineno"> 966</span>&#160;</div><div class="line"><a name="l00967"></a><span class="lineno"> 967</span>&#160;<span class="preprocessor">#elif K0 == 4 // K0 == 4</span></div><div class="line"><a name="l00968"></a><span class="lineno"> 968</span>&#160; <span class="comment">// This part computes the following transpositions:</span></div><div class="line"><a name="l00969"></a><span class="lineno"> 969</span>&#160; <span class="comment">// 4x2 -&gt; 2x4</span></div><div class="line"><a name="l00970"></a><span class="lineno"> 970</span>&#160; <span class="comment">// 4x4 -&gt; 4x4</span></div><div class="line"><a name="l00971"></a><span class="lineno"> 971</span>&#160; <span class="comment">// 4x8 -&gt; 8x4</span></div><div class="line"><a name="l00972"></a><span class="lineno"> 972</span>&#160; <span class="comment">// 4x16 -&gt; 16x4</span></div><div class="line"><a name="l00973"></a><span class="lineno"> 973</span>&#160; res0 = (<a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, K0))(a0.s0, a1.s0, a2.s0, a3.s0);</div><div class="line"><a name="l00974"></a><span class="lineno"> 974</span>&#160; res1 = (<a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, K0))(a0.s1, a1.s1, a2.s1, a3.s1);</div><div class="line"><a name="l00975"></a><span class="lineno"> 975</span>&#160;<span class="preprocessor">#if N0 &gt; 2</span></div><div class="line"><a name="l00976"></a><span class="lineno"> 976</span>&#160; res2 = (<a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, K0))(a0.s2, a1.s2, a2.s2, a3.s2);</div><div class="line"><a name="l00977"></a><span class="lineno"> 977</span>&#160;<span class="preprocessor">#endif // N0 &gt; 2</span></div><div class="line"><a name="l00978"></a><span class="lineno"> 978</span>&#160;<span class="preprocessor">#if N0 &gt; 3</span></div><div class="line"><a name="l00979"></a><span class="lineno"> 979</span>&#160; res3 = (<a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, K0))(a0.s3, a1.s3, a2.s3, a3.s3);</div><div class="line"><a name="l00980"></a><span class="lineno"> 980</span>&#160;<span class="preprocessor">#endif // N0 &gt; 3</span></div><div class="line"><a name="l00981"></a><span class="lineno"> 981</span>&#160;<span class="preprocessor">#if N0 &gt; 4</span></div><div class="line"><a name="l00982"></a><span class="lineno"> 982</span>&#160; res4 = (<a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, K0))(a0.s4, a1.s4, a2.s4, a3.s4);</div><div class="line"><a name="l00983"></a><span class="lineno"> 983</span>&#160; res5 = (<a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, K0))(a0.s5, a1.s5, a2.s5, a3.s5);</div><div class="line"><a name="l00984"></a><span class="lineno"> 984</span>&#160; res6 = (<a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, K0))(a0.s6, a1.s6, a2.s6, a3.s6);</div><div class="line"><a name="l00985"></a><span class="lineno"> 985</span>&#160; res7 = (<a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, K0))(a0.s7, a1.s7, a2.s7, a3.s7);</div><div class="line"><a name="l00986"></a><span class="lineno"> 986</span>&#160;<span class="preprocessor">#endif // N0 &gt; 4</span></div><div class="line"><a name="l00987"></a><span class="lineno"> 987</span>&#160;<span class="preprocessor">#if N0 &gt; 8</span></div><div class="line"><a name="l00988"></a><span class="lineno"> 988</span>&#160; res8 = (<a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, K0))(a0.s8, a1.s8, a2.s8, a3.s8);</div><div class="line"><a name="l00989"></a><span class="lineno"> 989</span>&#160; res9 = (<a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, K0))(a0.s9, a1.s9, a2.s9, a3.s9);</div><div class="line"><a name="l00990"></a><span class="lineno"> 990</span>&#160; resA = (<a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, K0))(a0.sA, a1.sA, a2.sA, a3.sA);</div><div class="line"><a name="l00991"></a><span class="lineno"> 991</span>&#160; resB = (<a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, K0))(a0.sB, a1.sB, a2.sB, a3.sB);</div><div class="line"><a name="l00992"></a><span class="lineno"> 992</span>&#160; resC = (<a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, K0))(a0.sC, a1.sC, a2.sC, a3.sC);</div><div class="line"><a name="l00993"></a><span class="lineno"> 993</span>&#160; resD = (<a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, K0))(a0.sD, a1.sD, a2.sD, a3.sD);</div><div class="line"><a name="l00994"></a><span class="lineno"> 994</span>&#160; resE = (<a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, K0))(a0.sE, a1.sE, a2.sE, a3.sE);</div><div class="line"><a name="l00995"></a><span class="lineno"> 995</span>&#160; resF = (<a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, K0))(a0.sF, a1.sF, a2.sF, a3.sF);</div><div class="line"><a name="l00996"></a><span class="lineno"> 996</span>&#160;<span class="preprocessor">#endif // N0 &gt; 8</span></div><div class="line"><a name="l00997"></a><span class="lineno"> 997</span>&#160;</div><div class="line"><a name="l00998"></a><span class="lineno"> 998</span>&#160;<span class="preprocessor">#elif K0 == 8 // K0 == 8</span></div><div class="line"><a name="l00999"></a><span class="lineno"> 999</span>&#160; <span class="comment">// This part computes the following transpositions:</span></div><div class="line"><a name="l01000"></a><span class="lineno"> 1000</span>&#160; <span class="comment">// 8x2 -&gt; 2x8</span></div><div class="line"><a name="l01001"></a><span class="lineno"> 1001</span>&#160; <span class="comment">// 8x4 -&gt; 4x8</span></div><div class="line"><a name="l01002"></a><span class="lineno"> 1002</span>&#160; <span class="comment">// 8x8 -&gt; 8x8</span></div><div class="line"><a name="l01003"></a><span class="lineno"> 1003</span>&#160; <span class="comment">// 8x16 -&gt; 16x8</span></div><div class="line"><a name="l01004"></a><span class="lineno"> 1004</span>&#160; res0 = (<a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, K0))(a0.s0, a1.s0, a2.s0, a3.s0, a4.s0, a5.s0, a6.s0, a7.s0);</div><div class="line"><a name="l01005"></a><span class="lineno"> 1005</span>&#160; res1 = (<a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, K0))(a0.s1, a1.s1, a2.s1, a3.s1, a4.s1, a5.s1, a6.s1, a7.s1);</div><div class="line"><a name="l01006"></a><span class="lineno"> 1006</span>&#160;<span class="preprocessor">#if N0 &gt; 2</span></div><div class="line"><a name="l01007"></a><span class="lineno"> 1007</span>&#160; res2 = (<a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, K0))(a0.s2, a1.s2, a2.s2, a3.s2, a4.s2, a5.s2, a6.s2, a7.s2);</div><div class="line"><a name="l01008"></a><span class="lineno"> 1008</span>&#160;<span class="preprocessor">#endif // N0 &gt; 2</span></div><div class="line"><a name="l01009"></a><span class="lineno"> 1009</span>&#160;<span class="preprocessor">#if N0 &gt; 3</span></div><div class="line"><a name="l01010"></a><span class="lineno"> 1010</span>&#160; res3 = (<a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, K0))(a0.s3, a1.s3, a2.s3, a3.s3, a4.s3, a5.s3, a6.s3, a7.s3);</div><div class="line"><a name="l01011"></a><span class="lineno"> 1011</span>&#160;<span class="preprocessor">#endif // N0 &gt; 3</span></div><div class="line"><a name="l01012"></a><span class="lineno"> 1012</span>&#160;<span class="preprocessor">#if N0 &gt; 4</span></div><div class="line"><a name="l01013"></a><span class="lineno"> 1013</span>&#160; res4 = (<a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, K0))(a0.s4, a1.s4, a2.s4, a3.s4, a4.s4, a5.s4, a6.s4, a7.s4);</div><div class="line"><a name="l01014"></a><span class="lineno"> 1014</span>&#160; res5 = (<a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, K0))(a0.s5, a1.s5, a2.s5, a3.s5, a4.s5, a5.s5, a6.s5, a7.s5);</div><div class="line"><a name="l01015"></a><span class="lineno"> 1015</span>&#160; res6 = (<a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, K0))(a0.s6, a1.s6, a2.s6, a3.s6, a4.s6, a5.s6, a6.s6, a7.s6);</div><div class="line"><a name="l01016"></a><span class="lineno"> 1016</span>&#160; res7 = (<a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, K0))(a0.s7, a1.s7, a2.s7, a3.s7, a4.s7, a5.s7, a6.s7, a7.s7);</div><div class="line"><a name="l01017"></a><span class="lineno"> 1017</span>&#160;<span class="preprocessor">#endif // N0 &gt; 4</span></div><div class="line"><a name="l01018"></a><span class="lineno"> 1018</span>&#160;<span class="preprocessor">#if N0 &gt; 8</span></div><div class="line"><a name="l01019"></a><span class="lineno"> 1019</span>&#160; res8 = (<a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, K0))(a0.s8, a1.s8, a2.s8, a3.s8, a4.s8, a5.s8, a6.s8, a7.s8);</div><div class="line"><a name="l01020"></a><span class="lineno"> 1020</span>&#160; res9 = (<a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, K0))(a0.s9, a1.s9, a2.s9, a3.s9, a4.s9, a5.s9, a6.s9, a7.s9);</div><div class="line"><a name="l01021"></a><span class="lineno"> 1021</span>&#160; resA = (<a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, K0))(a0.sA, a1.sA, a2.sA, a3.sA, a4.sA, a5.sA, a6.sA, a7.sA);</div><div class="line"><a name="l01022"></a><span class="lineno"> 1022</span>&#160; resB = (<a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, K0))(a0.sB, a1.sB, a2.sB, a3.sB, a4.sB, a5.sB, a6.sB, a7.sB);</div><div class="line"><a name="l01023"></a><span class="lineno"> 1023</span>&#160; resC = (<a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, K0))(a0.sC, a1.sC, a2.sC, a3.sC, a4.sC, a5.sC, a6.sC, a7.sC);</div><div class="line"><a name="l01024"></a><span class="lineno"> 1024</span>&#160; resD = (<a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, K0))(a0.sD, a1.sD, a2.sD, a3.sD, a4.sD, a5.sD, a6.sD, a7.sD);</div><div class="line"><a name="l01025"></a><span class="lineno"> 1025</span>&#160; resE = (<a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, K0))(a0.sE, a1.sE, a2.sE, a3.sE, a4.sE, a5.sE, a6.sE, a7.sE);</div><div class="line"><a name="l01026"></a><span class="lineno"> 1026</span>&#160; resF = (<a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, K0))(a0.sF, a1.sF, a2.sF, a3.sF, a4.sF, a5.sF, a6.sF, a7.sF);</div><div class="line"><a name="l01027"></a><span class="lineno"> 1027</span>&#160;<span class="preprocessor">#endif // N0 &gt; 8</span></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;<span class="preprocessor">#elif K0 == 16 // K0 == 16</span></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; <span class="comment">// This part computes the following transpositions:</span></div><div class="line"><a name="l01032"></a><span class="lineno"> 1032</span>&#160; <span class="comment">// 16x2 -&gt; 2x16</span></div><div class="line"><a name="l01033"></a><span class="lineno"> 1033</span>&#160; <span class="comment">// 16x4 -&gt; 4x16</span></div><div class="line"><a name="l01034"></a><span class="lineno"> 1034</span>&#160; <span class="comment">// 16x8 -&gt; 8x16</span></div><div class="line"><a name="l01035"></a><span class="lineno"> 1035</span>&#160; <span class="comment">// 16x16 -&gt; 16x16</span></div><div class="line"><a name="l01036"></a><span class="lineno"> 1036</span>&#160; res0 = (<a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, K0))(a0.s0, a1.s0, a2.s0, a3.s0, a4.s0, a5.s0, a6.s0, a7.s0,</div><div class="line"><a name="l01037"></a><span class="lineno"> 1037</span>&#160; a8.s0, a9.s0, aA.s0, aB.s0, aC.s0, aD.s0, aE.s0, aF.s0);</div><div class="line"><a name="l01038"></a><span class="lineno"> 1038</span>&#160; res1 = (<a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, K0))(a0.s1, a1.s1, a2.s1, a3.s1, a4.s1, a5.s1, a6.s1, a7.s1,</div><div class="line"><a name="l01039"></a><span class="lineno"> 1039</span>&#160; a8.s1, a9.s1, aA.s1, aB.s1, aC.s1, aD.s1, aE.s1, aF.s1);</div><div class="line"><a name="l01040"></a><span class="lineno"> 1040</span>&#160;<span class="preprocessor">#if N0 &gt; 2</span></div><div class="line"><a name="l01041"></a><span class="lineno"> 1041</span>&#160; res2 = (<a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, K0))(a0.s2, a1.s2, a2.s2, a3.s2, a4.s2, a5.s2, a6.s2, a7.s2,</div><div class="line"><a name="l01042"></a><span class="lineno"> 1042</span>&#160; a8.s2, a9.s2, aA.s2, aB.s2, aC.s2, aD.s2, aE.s2, aF.s2);</div><div class="line"><a name="l01043"></a><span class="lineno"> 1043</span>&#160;<span class="preprocessor">#endif // N0 &gt; 2</span></div><div class="line"><a name="l01044"></a><span class="lineno"> 1044</span>&#160;<span class="preprocessor">#if N0 &gt; 3</span></div><div class="line"><a name="l01045"></a><span class="lineno"> 1045</span>&#160; res3 = (<a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, K0))(a0.s3, a1.s3, a2.s3, a3.s3, a4.s3, a5.s3, a6.s3, a7.s3,</div><div class="line"><a name="l01046"></a><span class="lineno"> 1046</span>&#160; a8.s3, a9.s3, aA.s3, aB.s3, aC.s3, aD.s3, aE.s3, aF.s3);</div><div class="line"><a name="l01047"></a><span class="lineno"> 1047</span>&#160;<span class="preprocessor">#endif // N0 &gt; 3</span></div><div class="line"><a name="l01048"></a><span class="lineno"> 1048</span>&#160;<span class="preprocessor">#if N0 &gt; 4</span></div><div class="line"><a name="l01049"></a><span class="lineno"> 1049</span>&#160; res4 = (<a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, K0))(a0.s4, a1.s4, a2.s4, a3.s4, a4.s4, a5.s4, a6.s4, a7.s4,</div><div class="line"><a name="l01050"></a><span class="lineno"> 1050</span>&#160; a8.s4, a9.s4, aA.s4, aB.s4, aC.s4, aD.s4, aE.s4, aF.s4);</div><div class="line"><a name="l01051"></a><span class="lineno"> 1051</span>&#160; res5 = (<a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, K0))(a0.s5, a1.s5, a2.s5, a3.s5, a4.s5, a5.s5, a6.s5, a7.s5,</div><div class="line"><a name="l01052"></a><span class="lineno"> 1052</span>&#160; a8.s5, a9.s5, aA.s5, aB.s5, aC.s5, aD.s5, aE.s5, aF.s5);</div><div class="line"><a name="l01053"></a><span class="lineno"> 1053</span>&#160; res6 = (<a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, K0))(a0.s6, a1.s6, a2.s6, a3.s6, a4.s6, a5.s6, a6.s6, a7.s6,</div><div class="line"><a name="l01054"></a><span class="lineno"> 1054</span>&#160; a8.s6, a9.s6, aA.s6, aB.s6, aC.s6, aD.s6, aE.s6, aF.s6);</div><div class="line"><a name="l01055"></a><span class="lineno"> 1055</span>&#160; res7 = (<a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, K0))(a0.s7, a1.s7, a2.s7, a3.s7, a4.s7, a5.s7, a6.s7, a7.s7,</div><div class="line"><a name="l01056"></a><span class="lineno"> 1056</span>&#160; a8.s7, a9.s7, aA.s7, aB.s7, aC.s7, aD.s7, aE.s7, aF.s7);</div><div class="line"><a name="l01057"></a><span class="lineno"> 1057</span>&#160;<span class="preprocessor">#endif // N0 &gt; 4</span></div><div class="line"><a name="l01058"></a><span class="lineno"> 1058</span>&#160;<span class="preprocessor">#if N0 &gt; 8</span></div><div class="line"><a name="l01059"></a><span class="lineno"> 1059</span>&#160; res8 = (<a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, K0))(a0.s8, a1.s8, a2.s8, a3.s8, a4.s8, a5.s8, a6.s8, a7.s8,</div><div class="line"><a name="l01060"></a><span class="lineno"> 1060</span>&#160; a8.s8, a9.s8, aA.s8, aB.s8, aC.s8, aD.s8, aE.s8, aF.s8);</div><div class="line"><a name="l01061"></a><span class="lineno"> 1061</span>&#160; res9 = (<a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, K0))(a0.s9, a1.s9, a2.s9, a3.s9, a4.s9, a5.s9, a6.s9, a7.s9,</div><div class="line"><a name="l01062"></a><span class="lineno"> 1062</span>&#160; a8.s9, a9.s9, aA.s9, aB.s9, aC.s9, aD.s9, aE.s9, aF.s9);</div><div class="line"><a name="l01063"></a><span class="lineno"> 1063</span>&#160; resA = (<a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, K0))(a0.sA, a1.sA, a2.sA, a3.sA, a4.sA, a5.sA, a6.sA, a7.sA,</div><div class="line"><a name="l01064"></a><span class="lineno"> 1064</span>&#160; a8.sA, a9.sA, aA.sA, aB.sA, aC.sA, aD.sA, aE.sA, aF.sA);</div><div class="line"><a name="l01065"></a><span class="lineno"> 1065</span>&#160; resB = (<a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, K0))(a0.sB, a1.sB, a2.sB, a3.sB, a4.sB, a5.sB, a6.sB, a7.sB,</div><div class="line"><a name="l01066"></a><span class="lineno"> 1066</span>&#160; a8.sB, a9.sB, aA.sB, aB.sB, aC.sB, aD.sB, aE.sB, aF.sB);</div><div class="line"><a name="l01067"></a><span class="lineno"> 1067</span>&#160; resC = (<a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, K0))(a0.sC, a1.sC, a2.sC, a3.sC, a4.sC, a5.sC, a6.sC, a7.sC,</div><div class="line"><a name="l01068"></a><span class="lineno"> 1068</span>&#160; a8.sC, a9.sC, aA.sC, aB.sC, aC.sC, aD.sC, aE.sC, aF.sC);</div><div class="line"><a name="l01069"></a><span class="lineno"> 1069</span>&#160; resD = (<a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, K0))(a0.sD, a1.sD, a2.sD, a3.sD, a4.sD, a5.sD, a6.sD, a7.sD,</div><div class="line"><a name="l01070"></a><span class="lineno"> 1070</span>&#160; a8.sD, a9.sD, aA.sD, aB.sD, aC.sD, aD.sD, aE.sD, aF.sD);</div><div class="line"><a name="l01071"></a><span class="lineno"> 1071</span>&#160; resE = (<a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, K0))(a0.sE, a1.sE, a2.sE, a3.sE, a4.sE, a5.sE, a6.sE, a7.sE,</div><div class="line"><a name="l01072"></a><span class="lineno"> 1072</span>&#160; a8.sE, a9.sE, aA.sE, aB.sE, aC.sE, aD.sE, aE.sE, aF.sE);</div><div class="line"><a name="l01073"></a><span class="lineno"> 1073</span>&#160; resF = (<a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, K0))(a0.sF, a1.sF, a2.sF, a3.sF, a4.sF, a5.sF, a6.sF, a7.sF,</div><div class="line"><a name="l01074"></a><span class="lineno"> 1074</span>&#160; a8.sF, a9.sF, aA.sF, aB.sF, aC.sF, aD.sF, aE.sF, aF.sF);</div><div class="line"><a name="l01075"></a><span class="lineno"> 1075</span>&#160;<span class="preprocessor">#endif // N0 &gt; 8</span></div><div class="line"><a name="l01076"></a><span class="lineno"> 1076</span>&#160;</div><div class="line"><a name="l01077"></a><span class="lineno"> 1077</span>&#160;<span class="preprocessor">#else // N0 == 16</span></div><div class="line"><a name="l01078"></a><span class="lineno"> 1078</span>&#160;<span class="preprocessor">#error &quot;Not supported N0 value&quot;</span></div><div class="line"><a name="l01079"></a><span class="lineno"> 1079</span>&#160;<span class="preprocessor">#endif // N0 &gt; 2</span></div><div class="line"><a name="l01080"></a><span class="lineno"> 1080</span>&#160;</div><div class="line"><a name="l01081"></a><span class="lineno"> 1081</span>&#160; <span class="comment">// ---------------------------Store the output values ------------------------------</span></div><div class="line"><a name="l01082"></a><span class="lineno"> 1082</span>&#160;</div><div class="line"><a name="l01083"></a><span class="lineno"> 1083</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#acb282042d1edeeaa3cc979a206f78b54">VSTORE</a>(K0)</div><div class="line"><a name="l01084"></a><span class="lineno"> 1084</span>&#160; (res0, 0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(output_ptr + 0 * OUTPUT_STEP_X * <span class="keyword">sizeof</span>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>)));</div><div class="line"><a name="l01085"></a><span class="lineno"> 1085</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#acb282042d1edeeaa3cc979a206f78b54">VSTORE</a>(K0)</div><div class="line"><a name="l01086"></a><span class="lineno"> 1086</span>&#160; (res1, 0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(output_ptr + 1 * OUTPUT_STEP_X * <span class="keyword">sizeof</span>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>)));</div><div class="line"><a name="l01087"></a><span class="lineno"> 1087</span>&#160;<span class="preprocessor">#if N0 &gt; 2</span></div><div class="line"><a name="l01088"></a><span class="lineno"> 1088</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#acb282042d1edeeaa3cc979a206f78b54">VSTORE</a>(K0)</div><div class="line"><a name="l01089"></a><span class="lineno"> 1089</span>&#160; (res2, 0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(output_ptr + 2 * OUTPUT_STEP_X * <span class="keyword">sizeof</span>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>)));</div><div class="line"><a name="l01090"></a><span class="lineno"> 1090</span>&#160;<span class="preprocessor">#endif // N0 &gt; 2</span></div><div class="line"><a name="l01091"></a><span class="lineno"> 1091</span>&#160;<span class="preprocessor">#if N0 &gt; 3</span></div><div class="line"><a name="l01092"></a><span class="lineno"> 1092</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#acb282042d1edeeaa3cc979a206f78b54">VSTORE</a>(K0)</div><div class="line"><a name="l01093"></a><span class="lineno"> 1093</span>&#160; (res3, 0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(output_ptr + 3 * OUTPUT_STEP_X * <span class="keyword">sizeof</span>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>)));</div><div class="line"><a name="l01094"></a><span class="lineno"> 1094</span>&#160;<span class="preprocessor">#endif // N0 &gt; 3</span></div><div class="line"><a name="l01095"></a><span class="lineno"> 1095</span>&#160;<span class="preprocessor">#if N0 &gt; 4</span></div><div class="line"><a name="l01096"></a><span class="lineno"> 1096</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#acb282042d1edeeaa3cc979a206f78b54">VSTORE</a>(K0)</div><div class="line"><a name="l01097"></a><span class="lineno"> 1097</span>&#160; (res4, 0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(output_ptr + 4 * OUTPUT_STEP_X * <span class="keyword">sizeof</span>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>)));</div><div class="line"><a name="l01098"></a><span class="lineno"> 1098</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#acb282042d1edeeaa3cc979a206f78b54">VSTORE</a>(K0)</div><div class="line"><a name="l01099"></a><span class="lineno"> 1099</span>&#160; (res5, 0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(output_ptr + 5 * OUTPUT_STEP_X * <span class="keyword">sizeof</span>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>)));</div><div class="line"><a name="l01100"></a><span class="lineno"> 1100</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#acb282042d1edeeaa3cc979a206f78b54">VSTORE</a>(K0)</div><div class="line"><a name="l01101"></a><span class="lineno"> 1101</span>&#160; (res6, 0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(output_ptr + 6 * OUTPUT_STEP_X * <span class="keyword">sizeof</span>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>)));</div><div class="line"><a name="l01102"></a><span class="lineno"> 1102</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#acb282042d1edeeaa3cc979a206f78b54">VSTORE</a>(K0)</div><div class="line"><a name="l01103"></a><span class="lineno"> 1103</span>&#160; (res7, 0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(output_ptr + 7 * OUTPUT_STEP_X * <span class="keyword">sizeof</span>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>)));</div><div class="line"><a name="l01104"></a><span class="lineno"> 1104</span>&#160;<span class="preprocessor">#endif // N0 &gt; 4</span></div><div class="line"><a name="l01105"></a><span class="lineno"> 1105</span>&#160;<span class="preprocessor">#if N0 &gt; 8</span></div><div class="line"><a name="l01106"></a><span class="lineno"> 1106</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#acb282042d1edeeaa3cc979a206f78b54">VSTORE</a>(K0)</div><div class="line"><a name="l01107"></a><span class="lineno"> 1107</span>&#160; (res8, 0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(output_ptr + 8 * OUTPUT_STEP_X * <span class="keyword">sizeof</span>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>)));</div><div class="line"><a name="l01108"></a><span class="lineno"> 1108</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#acb282042d1edeeaa3cc979a206f78b54">VSTORE</a>(K0)</div><div class="line"><a name="l01109"></a><span class="lineno"> 1109</span>&#160; (res9, 0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(output_ptr + 9 * OUTPUT_STEP_X * <span class="keyword">sizeof</span>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>)));</div><div class="line"><a name="l01110"></a><span class="lineno"> 1110</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#acb282042d1edeeaa3cc979a206f78b54">VSTORE</a>(K0)</div><div class="line"><a name="l01111"></a><span class="lineno"> 1111</span>&#160; (resA, 0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(output_ptr + 10 * OUTPUT_STEP_X * <span class="keyword">sizeof</span>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>)));</div><div class="line"><a name="l01112"></a><span class="lineno"> 1112</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#acb282042d1edeeaa3cc979a206f78b54">VSTORE</a>(K0)</div><div class="line"><a name="l01113"></a><span class="lineno"> 1113</span>&#160; (resB, 0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(output_ptr + 11 * OUTPUT_STEP_X * <span class="keyword">sizeof</span>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>)));</div><div class="line"><a name="l01114"></a><span class="lineno"> 1114</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#acb282042d1edeeaa3cc979a206f78b54">VSTORE</a>(K0)</div><div class="line"><a name="l01115"></a><span class="lineno"> 1115</span>&#160; (resC, 0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(output_ptr + 12 * OUTPUT_STEP_X * <span class="keyword">sizeof</span>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>)));</div><div class="line"><a name="l01116"></a><span class="lineno"> 1116</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#acb282042d1edeeaa3cc979a206f78b54">VSTORE</a>(K0)</div><div class="line"><a name="l01117"></a><span class="lineno"> 1117</span>&#160; (resD, 0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(output_ptr + 13 * OUTPUT_STEP_X * <span class="keyword">sizeof</span>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>)));</div><div class="line"><a name="l01118"></a><span class="lineno"> 1118</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#acb282042d1edeeaa3cc979a206f78b54">VSTORE</a>(K0)</div><div class="line"><a name="l01119"></a><span class="lineno"> 1119</span>&#160; (resE, 0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(output_ptr + 14 * OUTPUT_STEP_X * <span class="keyword">sizeof</span>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>)));</div><div class="line"><a name="l01120"></a><span class="lineno"> 1120</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#acb282042d1edeeaa3cc979a206f78b54">VSTORE</a>(K0)</div><div class="line"><a name="l01121"></a><span class="lineno"> 1121</span>&#160; (resF, 0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(output_ptr + 15 * OUTPUT_STEP_X * <span class="keyword">sizeof</span>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>)));</div><div class="line"><a name="l01122"></a><span class="lineno"> 1122</span>&#160;<span class="preprocessor">#endif // N0 &gt; 8</span></div><div class="line"><a name="l01123"></a><span class="lineno"> 1123</span>&#160;</div><div class="line"><a name="l01124"></a><span class="lineno"> 1124</span>&#160;<span class="preprocessor">#undef BLOCK_SIZE</span></div><div class="line"><a name="l01125"></a><span class="lineno"> 1125</span>&#160;<span class="preprocessor">#undef OUTPUT_OFFSET_X</span></div><div class="line"><a name="l01126"></a><span class="lineno"> 1126</span>&#160;<span class="preprocessor">#undef OUTPUT_STEP_X</span></div><div class="line"><a name="l01127"></a><span class="lineno"> 1127</span>&#160;}</div><div class="line"><a name="l01128"></a><span class="lineno"> 1128</span>&#160;<span class="preprocessor">#endif // defined(TRANSPOSE)</span></div><div class="line"><a name="l01129"></a><span class="lineno"> 1129</span>&#160;<span class="preprocessor">#endif // defined(K0) &amp;&amp; defined(N0) &amp;&amp; defined(H0) &amp;&amp; defined(DATA_TYPE) &amp;&amp; defined(SRC_HEIGHT)</span></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;<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(DATA_TYPE)</span></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="preprocessor">#if K0 == 2</span></div><div class="line"><a name="l01134"></a><span class="lineno"> 1134</span>&#160;<span class="preprocessor">#define ARM_DOT_K0(a, b, c) \</span></div><div class="line"><a name="l01135"></a><span class="lineno"> 1135</span>&#160;<span class="preprocessor"> ({ \</span></div><div class="line"><a name="l01136"></a><span class="lineno"> 1136</span>&#160;<span class="preprocessor"> c = fma(a.s0, b.s0, c); \</span></div><div class="line"><a name="l01137"></a><span class="lineno"> 1137</span>&#160;<span class="preprocessor"> c = fma(a.s1, b.s1, c); \</span></div><div class="line"><a name="l01138"></a><span class="lineno"> 1138</span>&#160;<span class="preprocessor"> })</span></div><div class="line"><a name="l01139"></a><span class="lineno"> 1139</span>&#160;<span class="preprocessor">#elif K0 == 3 // K0 == 3</span></div><div class="line"><a name="l01140"></a><span class="lineno"> 1140</span>&#160;<span class="preprocessor">#define ARM_DOT_K0(a, b, c) \</span></div><div class="line"><a name="l01141"></a><span class="lineno"> 1141</span>&#160;<span class="preprocessor"> ({ \</span></div><div class="line"><a name="l01142"></a><span class="lineno"> 1142</span>&#160;<span class="preprocessor"> c = fma(a.s0, b.s0, c); \</span></div><div class="line"><a name="l01143"></a><span class="lineno"> 1143</span>&#160;<span class="preprocessor"> c = fma(a.s1, b.s1, c); \</span></div><div class="line"><a name="l01144"></a><span class="lineno"> 1144</span>&#160;<span class="preprocessor"> c = fma(a.s2, b.s2, c); \</span></div><div class="line"><a name="l01145"></a><span class="lineno"> 1145</span>&#160;<span class="preprocessor"> })</span></div><div class="line"><a name="l01146"></a><span class="lineno"> 1146</span>&#160;<span class="preprocessor">#elif K0 == 4 // K0 == 4</span></div><div class="line"><a name="l01147"></a><span class="lineno"> 1147</span>&#160;<span class="preprocessor">#define ARM_DOT_K0(a, b, c) \</span></div><div class="line"><a name="l01148"></a><span class="lineno"> 1148</span>&#160;<span class="preprocessor"> ({ \</span></div><div class="line"><a name="l01149"></a><span class="lineno"> 1149</span>&#160;<span class="preprocessor"> c = fma(a.s0, b.s0, c); \</span></div><div class="line"><a name="l01150"></a><span class="lineno"> 1150</span>&#160;<span class="preprocessor"> c = fma(a.s1, b.s1, c); \</span></div><div class="line"><a name="l01151"></a><span class="lineno"> 1151</span>&#160;<span class="preprocessor"> c = fma(a.s2, b.s2, c); \</span></div><div class="line"><a name="l01152"></a><span class="lineno"> 1152</span>&#160;<span class="preprocessor"> c = fma(a.s3, b.s3, c); \</span></div><div class="line"><a name="l01153"></a><span class="lineno"> 1153</span>&#160;<span class="preprocessor"> })</span></div><div class="line"><a name="l01154"></a><span class="lineno"> 1154</span>&#160;<span class="preprocessor">#elif K0 == 8 // K0 == 8</span></div><div class="line"><a name="l01155"></a><span class="lineno"> 1155</span>&#160;<span class="preprocessor">#define ARM_DOT_K0(a, b, c) \</span></div><div class="line"><a name="l01156"></a><span class="lineno"> 1156</span>&#160;<span class="preprocessor"> ({ \</span></div><div class="line"><a name="l01157"></a><span class="lineno"> 1157</span>&#160;<span class="preprocessor"> c = fma(a.s0, b.s0, c); \</span></div><div class="line"><a name="l01158"></a><span class="lineno"> 1158</span>&#160;<span class="preprocessor"> c = fma(a.s1, b.s1, c); \</span></div><div class="line"><a name="l01159"></a><span class="lineno"> 1159</span>&#160;<span class="preprocessor"> c = fma(a.s2, b.s2, c); \</span></div><div class="line"><a name="l01160"></a><span class="lineno"> 1160</span>&#160;<span class="preprocessor"> c = fma(a.s3, b.s3, c); \</span></div><div class="line"><a name="l01161"></a><span class="lineno"> 1161</span>&#160;<span class="preprocessor"> c = fma(a.s4, b.s4, c); \</span></div><div class="line"><a name="l01162"></a><span class="lineno"> 1162</span>&#160;<span class="preprocessor"> c = fma(a.s5, b.s5, c); \</span></div><div class="line"><a name="l01163"></a><span class="lineno"> 1163</span>&#160;<span class="preprocessor"> c = fma(a.s6, b.s6, c); \</span></div><div class="line"><a name="l01164"></a><span class="lineno"> 1164</span>&#160;<span class="preprocessor"> c = fma(a.s7, b.s7, c); \</span></div><div class="line"><a name="l01165"></a><span class="lineno"> 1165</span>&#160;<span class="preprocessor"> })</span></div><div class="line"><a name="l01166"></a><span class="lineno"> 1166</span>&#160;<span class="preprocessor">#elif K0 == 16 // K0 == 16</span></div><div class="line"><a name="l01167"></a><span class="lineno"> 1167</span>&#160;<span class="preprocessor">#define ARM_DOT_K0(a, b, c) \</span></div><div class="line"><a name="l01168"></a><span class="lineno"> 1168</span>&#160;<span class="preprocessor"> ({ \</span></div><div class="line"><a name="l01169"></a><span class="lineno"> 1169</span>&#160;<span class="preprocessor"> c = fma(a.s0, b.s0, c); \</span></div><div class="line"><a name="l01170"></a><span class="lineno"> 1170</span>&#160;<span class="preprocessor"> c = fma(a.s1, b.s1, c); \</span></div><div class="line"><a name="l01171"></a><span class="lineno"> 1171</span>&#160;<span class="preprocessor"> c = fma(a.s2, b.s2, c); \</span></div><div class="line"><a name="l01172"></a><span class="lineno"> 1172</span>&#160;<span class="preprocessor"> c = fma(a.s3, b.s3, c); \</span></div><div class="line"><a name="l01173"></a><span class="lineno"> 1173</span>&#160;<span class="preprocessor"> c = fma(a.s4, b.s4, c); \</span></div><div class="line"><a name="l01174"></a><span class="lineno"> 1174</span>&#160;<span class="preprocessor"> c = fma(a.s5, b.s5, c); \</span></div><div class="line"><a name="l01175"></a><span class="lineno"> 1175</span>&#160;<span class="preprocessor"> c = fma(a.s6, b.s6, c); \</span></div><div class="line"><a name="l01176"></a><span class="lineno"> 1176</span>&#160;<span class="preprocessor"> c = fma(a.s7, b.s7, c); \</span></div><div class="line"><a name="l01177"></a><span class="lineno"> 1177</span>&#160;<span class="preprocessor"> c = fma(a.s8, b.s8, c); \</span></div><div class="line"><a name="l01178"></a><span class="lineno"> 1178</span>&#160;<span class="preprocessor"> c = fma(a.s9, b.s9, c); \</span></div><div class="line"><a name="l01179"></a><span class="lineno"> 1179</span>&#160;<span class="preprocessor"> c = fma(a.sA, b.sA, c); \</span></div><div class="line"><a name="l01180"></a><span class="lineno"> 1180</span>&#160;<span class="preprocessor"> c = fma(a.sB, b.sB, c); \</span></div><div class="line"><a name="l01181"></a><span class="lineno"> 1181</span>&#160;<span class="preprocessor"> c = fma(a.sC, b.sC, c); \</span></div><div class="line"><a name="l01182"></a><span class="lineno"> 1182</span>&#160;<span class="preprocessor"> c = fma(a.sD, b.sD, c); \</span></div><div class="line"><a name="l01183"></a><span class="lineno"> 1183</span>&#160;<span class="preprocessor"> c = fma(a.sE, b.sE, c); \</span></div><div class="line"><a name="l01184"></a><span class="lineno"> 1184</span>&#160;<span class="preprocessor"> c = fma(a.sF, b.sF, c); \</span></div><div class="line"><a name="l01185"></a><span class="lineno"> 1185</span>&#160;<span class="preprocessor"> })</span></div><div class="line"><a name="l01186"></a><span class="lineno"> 1186</span>&#160;<span class="preprocessor">#else // K0 not supported</span></div><div class="line"><a name="l01187"></a><span class="lineno"> 1187</span>&#160;<span class="preprocessor">#error &quot;K0 value not supported&quot;</span></div><div class="line"><a name="l01188"></a><span class="lineno"> 1188</span>&#160;<span class="preprocessor">#endif // K0 conditions</span></div><div class="line"><a name="l01189"></a><span class="lineno"> 1189</span>&#160;</div><div class="line"><a name="l01190"></a><span class="lineno"> 1190</span>&#160;<span class="preprocessor">#if N0 == 2</span></div><div class="line"><a name="l01191"></a><span class="lineno"> 1191</span>&#160;<span class="preprocessor">#define ARM_DOT_K0XN0(a, b, c) \</span></div><div class="line"><a name="l01192"></a><span class="lineno"> 1192</span>&#160;<span class="preprocessor"> ({ \</span></div><div class="line"><a name="l01193"></a><span class="lineno"> 1193</span>&#160;<span class="preprocessor"> ARM_DOT_K0((a), (b##0), (c.s0)); \</span></div><div class="line"><a name="l01194"></a><span class="lineno"> 1194</span>&#160;<span class="preprocessor"> ARM_DOT_K0((a), (b##1), (c.s1)); \</span></div><div class="line"><a name="l01195"></a><span class="lineno"> 1195</span>&#160;<span class="preprocessor"> })</span></div><div class="line"><a name="l01196"></a><span class="lineno"> 1196</span>&#160;<span class="preprocessor">#elif N0 == 3 // N0 == 3</span></div><div class="line"><a name="l01197"></a><span class="lineno"> 1197</span>&#160;<span class="preprocessor">#define ARM_DOT_K0XN0(a, b, c) \</span></div><div class="line"><a name="l01198"></a><span class="lineno"> 1198</span>&#160;<span class="preprocessor"> ({ \</span></div><div class="line"><a name="l01199"></a><span class="lineno"> 1199</span>&#160;<span class="preprocessor"> ARM_DOT_K0((a), (b##0), (c.s0)); \</span></div><div class="line"><a name="l01200"></a><span class="lineno"> 1200</span>&#160;<span class="preprocessor"> ARM_DOT_K0((a), (b##1), (c.s1)); \</span></div><div class="line"><a name="l01201"></a><span class="lineno"> 1201</span>&#160;<span class="preprocessor"> ARM_DOT_K0((a), (b##2), (c.s2)); \</span></div><div class="line"><a name="l01202"></a><span class="lineno"> 1202</span>&#160;<span class="preprocessor"> })</span></div><div class="line"><a name="l01203"></a><span class="lineno"> 1203</span>&#160;<span class="preprocessor">#elif N0 == 4 // N0 == 4</span></div><div class="line"><a name="l01204"></a><span class="lineno"> 1204</span>&#160;<span class="preprocessor">#define ARM_DOT_K0XN0(a, b, c) \</span></div><div class="line"><a name="l01205"></a><span class="lineno"> 1205</span>&#160;<span class="preprocessor"> ({ \</span></div><div class="line"><a name="l01206"></a><span class="lineno"> 1206</span>&#160;<span class="preprocessor"> ARM_DOT_K0((a), (b##0), (c.s0)); \</span></div><div class="line"><a name="l01207"></a><span class="lineno"> 1207</span>&#160;<span class="preprocessor"> ARM_DOT_K0((a), (b##1), (c.s1)); \</span></div><div class="line"><a name="l01208"></a><span class="lineno"> 1208</span>&#160;<span class="preprocessor"> ARM_DOT_K0((a), (b##2), (c.s2)); \</span></div><div class="line"><a name="l01209"></a><span class="lineno"> 1209</span>&#160;<span class="preprocessor"> ARM_DOT_K0((a), (b##3), (c.s3)); \</span></div><div class="line"><a name="l01210"></a><span class="lineno"> 1210</span>&#160;<span class="preprocessor"> })</span></div><div class="line"><a name="l01211"></a><span class="lineno"> 1211</span>&#160;<span class="preprocessor">#elif N0 == 8 // N0 == 8</span></div><div class="line"><a name="l01212"></a><span class="lineno"> 1212</span>&#160;<span class="preprocessor">#define ARM_DOT_K0XN0(a, b, c) \</span></div><div class="line"><a name="l01213"></a><span class="lineno"> 1213</span>&#160;<span class="preprocessor"> ({ \</span></div><div class="line"><a name="l01214"></a><span class="lineno"> 1214</span>&#160;<span class="preprocessor"> ARM_DOT_K0((a), (b##0), (c.s0)); \</span></div><div class="line"><a name="l01215"></a><span class="lineno"> 1215</span>&#160;<span class="preprocessor"> ARM_DOT_K0((a), (b##1), (c.s1)); \</span></div><div class="line"><a name="l01216"></a><span class="lineno"> 1216</span>&#160;<span class="preprocessor"> ARM_DOT_K0((a), (b##2), (c.s2)); \</span></div><div class="line"><a name="l01217"></a><span class="lineno"> 1217</span>&#160;<span class="preprocessor"> ARM_DOT_K0((a), (b##3), (c.s3)); \</span></div><div class="line"><a name="l01218"></a><span class="lineno"> 1218</span>&#160;<span class="preprocessor"> ARM_DOT_K0((a), (b##4), (c.s4)); \</span></div><div class="line"><a name="l01219"></a><span class="lineno"> 1219</span>&#160;<span class="preprocessor"> ARM_DOT_K0((a), (b##5), (c.s5)); \</span></div><div class="line"><a name="l01220"></a><span class="lineno"> 1220</span>&#160;<span class="preprocessor"> ARM_DOT_K0((a), (b##6), (c.s6)); \</span></div><div class="line"><a name="l01221"></a><span class="lineno"> 1221</span>&#160;<span class="preprocessor"> ARM_DOT_K0((a), (b##7), (c.s7)); \</span></div><div class="line"><a name="l01222"></a><span class="lineno"> 1222</span>&#160;<span class="preprocessor"> })</span></div><div class="line"><a name="l01223"></a><span class="lineno"> 1223</span>&#160;<span class="preprocessor">#elif N0 == 16 // N0 == 16</span></div><div class="line"><a name="l01224"></a><span class="lineno"> 1224</span>&#160;<span class="preprocessor">#define ARM_DOT_K0XN0(a, b, c) \</span></div><div class="line"><a name="l01225"></a><span class="lineno"> 1225</span>&#160;<span class="preprocessor"> ({ \</span></div><div class="line"><a name="l01226"></a><span class="lineno"> 1226</span>&#160;<span class="preprocessor"> ARM_DOT_K0((a), (b##0), (c.s0)); \</span></div><div class="line"><a name="l01227"></a><span class="lineno"> 1227</span>&#160;<span class="preprocessor"> ARM_DOT_K0((a), (b##1), (c.s1)); \</span></div><div class="line"><a name="l01228"></a><span class="lineno"> 1228</span>&#160;<span class="preprocessor"> ARM_DOT_K0((a), (b##2), (c.s2)); \</span></div><div class="line"><a name="l01229"></a><span class="lineno"> 1229</span>&#160;<span class="preprocessor"> ARM_DOT_K0((a), (b##3), (c.s3)); \</span></div><div class="line"><a name="l01230"></a><span class="lineno"> 1230</span>&#160;<span class="preprocessor"> ARM_DOT_K0((a), (b##4), (c.s4)); \</span></div><div class="line"><a name="l01231"></a><span class="lineno"> 1231</span>&#160;<span class="preprocessor"> ARM_DOT_K0((a), (b##5), (c.s5)); \</span></div><div class="line"><a name="l01232"></a><span class="lineno"> 1232</span>&#160;<span class="preprocessor"> ARM_DOT_K0((a), (b##6), (c.s6)); \</span></div><div class="line"><a name="l01233"></a><span class="lineno"> 1233</span>&#160;<span class="preprocessor"> ARM_DOT_K0((a), (b##7), (c.s7)); \</span></div><div class="line"><a name="l01234"></a><span class="lineno"> 1234</span>&#160;<span class="preprocessor"> ARM_DOT_K0((a), (b##8), (c.s8)); \</span></div><div class="line"><a name="l01235"></a><span class="lineno"> 1235</span>&#160;<span class="preprocessor"> ARM_DOT_K0((a), (b##9), (c.s9)); \</span></div><div class="line"><a name="l01236"></a><span class="lineno"> 1236</span>&#160;<span class="preprocessor"> ARM_DOT_K0((a), (b##A), (c.sA)); \</span></div><div class="line"><a name="l01237"></a><span class="lineno"> 1237</span>&#160;<span class="preprocessor"> ARM_DOT_K0((a), (b##B), (c.sB)); \</span></div><div class="line"><a name="l01238"></a><span class="lineno"> 1238</span>&#160;<span class="preprocessor"> ARM_DOT_K0((a), (b##C), (c.sC)); \</span></div><div class="line"><a name="l01239"></a><span class="lineno"> 1239</span>&#160;<span class="preprocessor"> ARM_DOT_K0((a), (b##D), (c.sD)); \</span></div><div class="line"><a name="l01240"></a><span class="lineno"> 1240</span>&#160;<span class="preprocessor"> ARM_DOT_K0((a), (b##E), (c.sE)); \</span></div><div class="line"><a name="l01241"></a><span class="lineno"> 1241</span>&#160;<span class="preprocessor"> ARM_DOT_K0((a), (b##F), (c.sF)); \</span></div><div class="line"><a name="l01242"></a><span class="lineno"> 1242</span>&#160;<span class="preprocessor"> })</span></div><div class="line"><a name="l01243"></a><span class="lineno"> 1243</span>&#160;<span class="preprocessor">#else // N0 not supported</span></div><div class="line"><a name="l01244"></a><span class="lineno"> 1244</span>&#160;<span class="preprocessor">#error &quot;N0 value not supported&quot;</span></div><div class="line"><a name="l01245"></a><span class="lineno"> 1245</span>&#160;<span class="preprocessor">#endif // N0 conditions</span></div><div class="line"><a name="l01246"></a><span class="lineno"> 1246</span>&#160;</div><div class="line"><a name="l01291"></a><span class="lineno"> 1291</span>&#160;__kernel <span class="keywordtype">void</span> gemm_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="l01292"></a><span class="lineno"> 1292</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="l01293"></a><span class="lineno"> 1293</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="l01294"></a><span class="lineno"> 1294</span>&#160; uint k,</div><div class="line"><a name="l01295"></a><span class="lineno"> 1295</span>&#160; uint lhs_stride_z,</div><div class="line"><a name="l01296"></a><span class="lineno"> 1296</span>&#160; uint rhs_stride_z,</div><div class="line"><a name="l01297"></a><span class="lineno"> 1297</span>&#160; uint dst_stride_z</div><div class="line"><a name="l01298"></a><span class="lineno"> 1298</span>&#160;#<span class="keywordflow">if</span> defined(REINTERPRET_OUTPUT_AS_3D)</div><div class="line"><a name="l01299"></a><span class="lineno"> 1299</span>&#160; ,</div><div class="line"><a name="l01300"></a><span class="lineno"> 1300</span>&#160; uint dst_cross_plane_pad</div><div class="line"><a name="l01301"></a><span class="lineno"> 1301</span>&#160;#endif <span class="comment">// REINTERPRET_OUTPUT_AS_3D</span></div><div class="line"><a name="l01302"></a><span class="lineno"> 1302</span>&#160; )</div><div class="line"><a name="l01303"></a><span class="lineno"> 1303</span>&#160;{</div><div class="line"><a name="l01304"></a><span class="lineno"> 1304</span>&#160; <span class="comment">// Block size</span></div><div class="line"><a name="l01305"></a><span class="lineno"> 1305</span>&#160;<span class="preprocessor">#define LHS_BLOCK_SIZE ((K0) * (M0))</span></div><div class="line"><a name="l01306"></a><span class="lineno"> 1306</span>&#160;</div><div class="line"><a name="l01307"></a><span class="lineno"> 1307</span>&#160;<span class="preprocessor">#if defined(LHS_INTERLEAVE)</span></div><div class="line"><a name="l01308"></a><span class="lineno"> 1308</span>&#160;<span class="preprocessor">#define LHS_OFFSET_X (K0)</span></div><div class="line"><a name="l01309"></a><span class="lineno"> 1309</span>&#160;<span class="preprocessor">#define LHS_STEP_X ((K0) * (V0))</span></div><div class="line"><a name="l01310"></a><span class="lineno"> 1310</span>&#160;<span class="preprocessor">#define LHS_STEP_LOOP (1)</span></div><div class="line"><a name="l01311"></a><span class="lineno"> 1311</span>&#160;<span class="preprocessor">#else // defined(INTERLEAVE)</span></div><div class="line"><a name="l01312"></a><span class="lineno"> 1312</span>&#160;<span class="preprocessor">#define LHS_OFFSET_X (LHS_BLOCK_SIZE)</span></div><div class="line"><a name="l01313"></a><span class="lineno"> 1313</span>&#160;<span class="preprocessor">#define LHS_STEP_X (K0)</span></div><div class="line"><a name="l01314"></a><span class="lineno"> 1314</span>&#160;<span class="preprocessor">#define LHS_STEP_LOOP (V0)</span></div><div class="line"><a name="l01315"></a><span class="lineno"> 1315</span>&#160;<span class="preprocessor">#endif // defined(INTERLEAVE)</span></div><div class="line"><a name="l01316"></a><span class="lineno"> 1316</span>&#160;</div><div class="line"><a name="l01317"></a><span class="lineno"> 1317</span>&#160; <span class="comment">// Block size</span></div><div class="line"><a name="l01318"></a><span class="lineno"> 1318</span>&#160;<span class="preprocessor">#define RHS_BLOCK_SIZE ((K0) * (N0))</span></div><div class="line"><a name="l01319"></a><span class="lineno"> 1319</span>&#160;</div><div class="line"><a name="l01320"></a><span class="lineno"> 1320</span>&#160; <span class="comment">// RHS offset and step X</span></div><div class="line"><a name="l01321"></a><span class="lineno"> 1321</span>&#160;<span class="preprocessor">#if defined(RHS_INTERLEAVE)</span></div><div class="line"><a name="l01322"></a><span class="lineno"> 1322</span>&#160;<span class="preprocessor">#define RHS_OFFSET_X (K0)</span></div><div class="line"><a name="l01323"></a><span class="lineno"> 1323</span>&#160;<span class="preprocessor">#define RHS_STEP_X ((K0) * (H0))</span></div><div class="line"><a name="l01324"></a><span class="lineno"> 1324</span>&#160;<span class="preprocessor">#define RHS_STEP_LOOP (1)</span></div><div class="line"><a name="l01325"></a><span class="lineno"> 1325</span>&#160;<span class="preprocessor">#else // defined(RHS_INTERLEAVE)</span></div><div class="line"><a name="l01326"></a><span class="lineno"> 1326</span>&#160;<span class="preprocessor">#define RHS_OFFSET_X (RHS_BLOCK_SIZE)</span></div><div class="line"><a name="l01327"></a><span class="lineno"> 1327</span>&#160;<span class="preprocessor">#define RHS_STEP_X (K0)</span></div><div class="line"><a name="l01328"></a><span class="lineno"> 1328</span>&#160;<span class="preprocessor">#define RHS_STEP_LOOP (H0)</span></div><div class="line"><a name="l01329"></a><span class="lineno"> 1329</span>&#160;<span class="preprocessor">#endif // defined(RHS_INTERLEAVE)</span></div><div class="line"><a name="l01330"></a><span class="lineno"> 1330</span>&#160;</div><div class="line"><a name="l01331"></a><span class="lineno"> 1331</span>&#160; <span class="comment">// Compute LHS matrix address</span></div><div class="line"><a name="l01332"></a><span class="lineno"> 1332</span>&#160; __global uchar *lhs_addr = lhs_ptr + lhs_offset_first_element_in_bytes + (get_global_id(1) % V0) * (uint)LHS_OFFSET_X * <span class="keyword">sizeof</span>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>) + (get_global_id(1) / V0) * (uint)lhs_stride_y +</div><div class="line"><a name="l01333"></a><span class="lineno"> 1333</span>&#160; (get_global_id(2) * lhs_stride_z);</div><div class="line"><a name="l01334"></a><span class="lineno"> 1334</span>&#160;</div><div class="line"><a name="l01335"></a><span class="lineno"> 1335</span>&#160; <span class="comment">// Compute RHS matrix address</span></div><div class="line"><a name="l01336"></a><span class="lineno"> 1336</span>&#160; __global uchar *rhs_addr = rhs_ptr + rhs_offset_first_element_in_bytes + (get_global_id(0) % H0) * (uint)RHS_OFFSET_X * <span class="keyword">sizeof</span>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>) + (get_global_id(0) / (uint)H0) * rhs_stride_y;</div><div class="line"><a name="l01337"></a><span class="lineno"> 1337</span>&#160;</div><div class="line"><a name="l01338"></a><span class="lineno"> 1338</span>&#160;<span class="preprocessor">#if defined(MATRIX_B_DEPTH)</span></div><div class="line"><a name="l01339"></a><span class="lineno"> 1339</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="l01340"></a><span class="lineno"> 1340</span>&#160; rhs_addr += (get_global_id(2) % MATRIX_B_DEPTH) * rhs_stride_z;</div><div class="line"><a name="l01341"></a><span class="lineno"> 1341</span>&#160;<span class="preprocessor">#else // defined(MATRIX_B_DEPTH)</span></div><div class="line"><a name="l01342"></a><span class="lineno"> 1342</span>&#160; rhs_addr += get_global_id(2) * rhs_stride_z;</div><div class="line"><a name="l01343"></a><span class="lineno"> 1343</span>&#160;<span class="preprocessor">#endif // defined(MATRIX_B_DEPTH)</span></div><div class="line"><a name="l01344"></a><span class="lineno"> 1344</span>&#160;</div><div class="line"><a name="l01345"></a><span class="lineno"> 1345</span>&#160; <span class="comment">// Initialize the accumulators</span></div><div class="line"><a name="l01346"></a><span class="lineno"> 1346</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>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, N0), c, 0); <span class="comment">//VEC_DATA_TYPE(DATA_TYPE, N0) c0=0,c1=0,c2=0,... c(M0-1)=0;</span></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; <span class="keywordflow">for</span>(<span class="keywordtype">int</span> i = 0; i &lt; k; i += K0)</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; <span class="comment">// Supported cases (M0, K0):</span></div><div class="line"><a name="l01351"></a><span class="lineno"> 1351</span>&#160; <span class="comment">// 2,4 - 2,8 - 2,16</span></div><div class="line"><a name="l01352"></a><span class="lineno"> 1352</span>&#160; <span class="comment">// 3,4 - 3,8 - 3,16</span></div><div class="line"><a name="l01353"></a><span class="lineno"> 1353</span>&#160; <span class="comment">// 4,4 - 4,8 - 4,16</span></div><div class="line"><a name="l01354"></a><span class="lineno"> 1354</span>&#160; <span class="comment">// 5,4 - 5,8 - 5,16</span></div><div class="line"><a name="l01355"></a><span class="lineno"> 1355</span>&#160; <span class="comment">// 6,4 - 6,8 - 6,16</span></div><div class="line"><a name="l01356"></a><span class="lineno"> 1356</span>&#160; <span class="comment">// Load values from LHS matrix</span></div><div class="line"><a name="l01357"></a><span class="lineno"> 1357</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, K0)</div><div class="line"><a name="l01358"></a><span class="lineno"> 1358</span>&#160; a0 = <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a287e2fc366c312b468382c95bb90f91f">VLOAD</a>(K0)(0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(lhs_addr + 0 * LHS_STEP_X * <span class="keyword">sizeof</span>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>)));</div><div class="line"><a name="l01359"></a><span class="lineno"> 1359</span>&#160;<span class="preprocessor">#if M0 &gt; 1</span></div><div class="line"><a name="l01360"></a><span class="lineno"> 1360</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, K0)</div><div class="line"><a name="l01361"></a><span class="lineno"> 1361</span>&#160; a1 = <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a287e2fc366c312b468382c95bb90f91f">VLOAD</a>(K0)(0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(lhs_addr + 1 * LHS_STEP_X * <span class="keyword">sizeof</span>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>)));</div><div class="line"><a name="l01362"></a><span class="lineno"> 1362</span>&#160;<span class="preprocessor">#endif // M0 &gt; 1</span></div><div class="line"><a name="l01363"></a><span class="lineno"> 1363</span>&#160;<span class="preprocessor">#if M0 &gt; 2</span></div><div class="line"><a name="l01364"></a><span class="lineno"> 1364</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, K0)</div><div class="line"><a name="l01365"></a><span class="lineno"> 1365</span>&#160; a2 = <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a287e2fc366c312b468382c95bb90f91f">VLOAD</a>(K0)(0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(lhs_addr + 2 * LHS_STEP_X * <span class="keyword">sizeof</span>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>)));</div><div class="line"><a name="l01366"></a><span class="lineno"> 1366</span>&#160;<span class="preprocessor">#endif // M0 &gt; 2</span></div><div class="line"><a name="l01367"></a><span class="lineno"> 1367</span>&#160;<span class="preprocessor">#if M0 &gt; 3</span></div><div class="line"><a name="l01368"></a><span class="lineno"> 1368</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, K0)</div><div class="line"><a name="l01369"></a><span class="lineno"> 1369</span>&#160; a3 = <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a287e2fc366c312b468382c95bb90f91f">VLOAD</a>(K0)(0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(lhs_addr + 3 * LHS_STEP_X * <span class="keyword">sizeof</span>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>)));</div><div class="line"><a name="l01370"></a><span class="lineno"> 1370</span>&#160;<span class="preprocessor">#endif // M0 &gt; 3</span></div><div class="line"><a name="l01371"></a><span class="lineno"> 1371</span>&#160;<span class="preprocessor">#if M0 &gt; 4</span></div><div class="line"><a name="l01372"></a><span class="lineno"> 1372</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, K0)</div><div class="line"><a name="l01373"></a><span class="lineno"> 1373</span>&#160; a4 = <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a287e2fc366c312b468382c95bb90f91f">VLOAD</a>(K0)(0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(lhs_addr + 4 * LHS_STEP_X * <span class="keyword">sizeof</span>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>)));</div><div class="line"><a name="l01374"></a><span class="lineno"> 1374</span>&#160;<span class="preprocessor">#endif // M0 &gt; 4</span></div><div class="line"><a name="l01375"></a><span class="lineno"> 1375</span>&#160;<span class="preprocessor">#if M0 &gt; 5</span></div><div class="line"><a name="l01376"></a><span class="lineno"> 1376</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, K0)</div><div class="line"><a name="l01377"></a><span class="lineno"> 1377</span>&#160; a5 = <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a287e2fc366c312b468382c95bb90f91f">VLOAD</a>(K0)(0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(lhs_addr + 5 * LHS_STEP_X * <span class="keyword">sizeof</span>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>)));</div><div class="line"><a name="l01378"></a><span class="lineno"> 1378</span>&#160;<span class="preprocessor">#endif // M0 &gt; 5</span></div><div class="line"><a name="l01379"></a><span class="lineno"> 1379</span>&#160;<span class="preprocessor">#if M0 &gt; 6</span></div><div class="line"><a name="l01380"></a><span class="lineno"> 1380</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, K0)</div><div class="line"><a name="l01381"></a><span class="lineno"> 1381</span>&#160; a6 = <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a287e2fc366c312b468382c95bb90f91f">VLOAD</a>(K0)(0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(lhs_addr + 6 * LHS_STEP_X * <span class="keyword">sizeof</span>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>)));</div><div class="line"><a name="l01382"></a><span class="lineno"> 1382</span>&#160;<span class="preprocessor">#endif // M0 &gt; 6</span></div><div class="line"><a name="l01383"></a><span class="lineno"> 1383</span>&#160;<span class="preprocessor">#if M0 &gt; 7</span></div><div class="line"><a name="l01384"></a><span class="lineno"> 1384</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, K0)</div><div class="line"><a name="l01385"></a><span class="lineno"> 1385</span>&#160; a7 = <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a287e2fc366c312b468382c95bb90f91f">VLOAD</a>(K0)(0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(lhs_addr + 7 * LHS_STEP_X * <span class="keyword">sizeof</span>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>)));</div><div class="line"><a name="l01386"></a><span class="lineno"> 1386</span>&#160;<span class="preprocessor">#endif // M0 &gt; 7</span></div><div class="line"><a name="l01387"></a><span class="lineno"> 1387</span>&#160;</div><div class="line"><a name="l01388"></a><span class="lineno"> 1388</span>&#160; <span class="comment">// Load values from RHS matrix</span></div><div class="line"><a name="l01389"></a><span class="lineno"> 1389</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, K0)</div><div class="line"><a name="l01390"></a><span class="lineno"> 1390</span>&#160; b0 = <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a287e2fc366c312b468382c95bb90f91f">VLOAD</a>(K0)(0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(rhs_addr + 0 * RHS_STEP_X * <span class="keyword">sizeof</span>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>)));</div><div class="line"><a name="l01391"></a><span class="lineno"> 1391</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, K0)</div><div class="line"><a name="l01392"></a><span class="lineno"> 1392</span>&#160; b1 = <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a287e2fc366c312b468382c95bb90f91f">VLOAD</a>(K0)(0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(rhs_addr + 1 * RHS_STEP_X * <span class="keyword">sizeof</span>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>)));</div><div class="line"><a name="l01393"></a><span class="lineno"> 1393</span>&#160;<span class="preprocessor">#if N0 &gt; 2</span></div><div class="line"><a name="l01394"></a><span class="lineno"> 1394</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, K0)</div><div class="line"><a name="l01395"></a><span class="lineno"> 1395</span>&#160; b2 = <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a287e2fc366c312b468382c95bb90f91f">VLOAD</a>(K0)(0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(rhs_addr + 2 * RHS_STEP_X * <span class="keyword">sizeof</span>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>)));</div><div class="line"><a name="l01396"></a><span class="lineno"> 1396</span>&#160;<span class="preprocessor">#endif // N0 &gt; 2</span></div><div class="line"><a name="l01397"></a><span class="lineno"> 1397</span>&#160;<span class="preprocessor">#if N0 &gt; 3</span></div><div class="line"><a name="l01398"></a><span class="lineno"> 1398</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, K0)</div><div class="line"><a name="l01399"></a><span class="lineno"> 1399</span>&#160; b3 = <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a287e2fc366c312b468382c95bb90f91f">VLOAD</a>(K0)(0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(rhs_addr + 3 * RHS_STEP_X * <span class="keyword">sizeof</span>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>)));</div><div class="line"><a name="l01400"></a><span class="lineno"> 1400</span>&#160;<span class="preprocessor">#endif // N0 &gt; 3</span></div><div class="line"><a name="l01401"></a><span class="lineno"> 1401</span>&#160;<span class="preprocessor">#if N0 &gt; 4</span></div><div class="line"><a name="l01402"></a><span class="lineno"> 1402</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, K0)</div><div class="line"><a name="l01403"></a><span class="lineno"> 1403</span>&#160; b4 = <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a287e2fc366c312b468382c95bb90f91f">VLOAD</a>(K0)(0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(rhs_addr + 4 * RHS_STEP_X * <span class="keyword">sizeof</span>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>)));</div><div class="line"><a name="l01404"></a><span class="lineno"> 1404</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, K0)</div><div class="line"><a name="l01405"></a><span class="lineno"> 1405</span>&#160; b5 = <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a287e2fc366c312b468382c95bb90f91f">VLOAD</a>(K0)(0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(rhs_addr + 5 * RHS_STEP_X * <span class="keyword">sizeof</span>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>)));</div><div class="line"><a name="l01406"></a><span class="lineno"> 1406</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, K0)</div><div class="line"><a name="l01407"></a><span class="lineno"> 1407</span>&#160; b6 = <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a287e2fc366c312b468382c95bb90f91f">VLOAD</a>(K0)(0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(rhs_addr + 6 * RHS_STEP_X * <span class="keyword">sizeof</span>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>)));</div><div class="line"><a name="l01408"></a><span class="lineno"> 1408</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, K0)</div><div class="line"><a name="l01409"></a><span class="lineno"> 1409</span>&#160; b7 = <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a287e2fc366c312b468382c95bb90f91f">VLOAD</a>(K0)(0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(rhs_addr + 7 * RHS_STEP_X * <span class="keyword">sizeof</span>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>)));</div><div class="line"><a name="l01410"></a><span class="lineno"> 1410</span>&#160;<span class="preprocessor">#endif // N0 &gt; 4</span></div><div class="line"><a name="l01411"></a><span class="lineno"> 1411</span>&#160;<span class="preprocessor">#if N0 &gt; 8</span></div><div class="line"><a name="l01412"></a><span class="lineno"> 1412</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, K0)</div><div class="line"><a name="l01413"></a><span class="lineno"> 1413</span>&#160; b8 = <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a287e2fc366c312b468382c95bb90f91f">VLOAD</a>(K0)(0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(rhs_addr + 8 * RHS_STEP_X * <span class="keyword">sizeof</span>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>)));</div><div class="line"><a name="l01414"></a><span class="lineno"> 1414</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, K0)</div><div class="line"><a name="l01415"></a><span class="lineno"> 1415</span>&#160; b9 = <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a287e2fc366c312b468382c95bb90f91f">VLOAD</a>(K0)(0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(rhs_addr + 9 * RHS_STEP_X * <span class="keyword">sizeof</span>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>)));</div><div class="line"><a name="l01416"></a><span class="lineno"> 1416</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, K0)</div><div class="line"><a name="l01417"></a><span class="lineno"> 1417</span>&#160; bA = <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a287e2fc366c312b468382c95bb90f91f">VLOAD</a>(K0)(0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(rhs_addr + 10 * RHS_STEP_X * <span class="keyword">sizeof</span>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>)));</div><div class="line"><a name="l01418"></a><span class="lineno"> 1418</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, K0)</div><div class="line"><a name="l01419"></a><span class="lineno"> 1419</span>&#160; bB = <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a287e2fc366c312b468382c95bb90f91f">VLOAD</a>(K0)(0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(rhs_addr + 11 * RHS_STEP_X * <span class="keyword">sizeof</span>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>)));</div><div class="line"><a name="l01420"></a><span class="lineno"> 1420</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, K0)</div><div class="line"><a name="l01421"></a><span class="lineno"> 1421</span>&#160; bC = <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a287e2fc366c312b468382c95bb90f91f">VLOAD</a>(K0)(0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(rhs_addr + 12 * RHS_STEP_X * <span class="keyword">sizeof</span>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>)));</div><div class="line"><a name="l01422"></a><span class="lineno"> 1422</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, K0)</div><div class="line"><a name="l01423"></a><span class="lineno"> 1423</span>&#160; bD = <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a287e2fc366c312b468382c95bb90f91f">VLOAD</a>(K0)(0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(rhs_addr + 13 * RHS_STEP_X * <span class="keyword">sizeof</span>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>)));</div><div class="line"><a name="l01424"></a><span class="lineno"> 1424</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, K0)</div><div class="line"><a name="l01425"></a><span class="lineno"> 1425</span>&#160; bE = <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a287e2fc366c312b468382c95bb90f91f">VLOAD</a>(K0)(0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(rhs_addr + 14 * RHS_STEP_X * <span class="keyword">sizeof</span>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>)));</div><div class="line"><a name="l01426"></a><span class="lineno"> 1426</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, K0)</div><div class="line"><a name="l01427"></a><span class="lineno"> 1427</span>&#160; bF = <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a287e2fc366c312b468382c95bb90f91f">VLOAD</a>(K0)(0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(rhs_addr + 15 * RHS_STEP_X * <span class="keyword">sizeof</span>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>)));</div><div class="line"><a name="l01428"></a><span class="lineno"> 1428</span>&#160;<span class="preprocessor">#endif // N0 &gt; 8</span></div><div class="line"><a name="l01429"></a><span class="lineno"> 1429</span>&#160;</div><div class="line"><a name="l01430"></a><span class="lineno"> 1430</span>&#160; <span class="comment">// Accumulate</span></div><div class="line"><a name="l01431"></a><span class="lineno"> 1431</span>&#160; ARM_DOT_K0XN0(a0, b, c0);</div><div class="line"><a name="l01432"></a><span class="lineno"> 1432</span>&#160;<span class="preprocessor">#if M0 &gt; 1</span></div><div class="line"><a name="l01433"></a><span class="lineno"> 1433</span>&#160; ARM_DOT_K0XN0(a1, b, c1);</div><div class="line"><a name="l01434"></a><span class="lineno"> 1434</span>&#160;<span class="preprocessor">#endif // M0 &gt; 1</span></div><div class="line"><a name="l01435"></a><span class="lineno"> 1435</span>&#160;<span class="preprocessor">#if M0 &gt; 2</span></div><div class="line"><a name="l01436"></a><span class="lineno"> 1436</span>&#160; ARM_DOT_K0XN0(a2, b, c2);</div><div class="line"><a name="l01437"></a><span class="lineno"> 1437</span>&#160;<span class="preprocessor">#endif // M0 &gt; 2</span></div><div class="line"><a name="l01438"></a><span class="lineno"> 1438</span>&#160;<span class="preprocessor">#if M0 &gt; 3</span></div><div class="line"><a name="l01439"></a><span class="lineno"> 1439</span>&#160; ARM_DOT_K0XN0(a3, b, c3);</div><div class="line"><a name="l01440"></a><span class="lineno"> 1440</span>&#160;<span class="preprocessor">#endif // M0 &gt; 3</span></div><div class="line"><a name="l01441"></a><span class="lineno"> 1441</span>&#160;<span class="preprocessor">#if M0 &gt; 4</span></div><div class="line"><a name="l01442"></a><span class="lineno"> 1442</span>&#160; ARM_DOT_K0XN0(a4, b, c4);</div><div class="line"><a name="l01443"></a><span class="lineno"> 1443</span>&#160;<span class="preprocessor">#endif // M0 &gt; 4</span></div><div class="line"><a name="l01444"></a><span class="lineno"> 1444</span>&#160;<span class="preprocessor">#if M0 &gt; 5</span></div><div class="line"><a name="l01445"></a><span class="lineno"> 1445</span>&#160; ARM_DOT_K0XN0(a5, b, c5);</div><div class="line"><a name="l01446"></a><span class="lineno"> 1446</span>&#160;<span class="preprocessor">#endif // M0 &gt; 5</span></div><div class="line"><a name="l01447"></a><span class="lineno"> 1447</span>&#160;<span class="preprocessor">#if M0 &gt; 6</span></div><div class="line"><a name="l01448"></a><span class="lineno"> 1448</span>&#160; ARM_DOT_K0XN0(a6, b, c6);</div><div class="line"><a name="l01449"></a><span class="lineno"> 1449</span>&#160;<span class="preprocessor">#endif // M0 &gt; 6</span></div><div class="line"><a name="l01450"></a><span class="lineno"> 1450</span>&#160;<span class="preprocessor">#if M0 &gt; 7</span></div><div class="line"><a name="l01451"></a><span class="lineno"> 1451</span>&#160; ARM_DOT_K0XN0(a7, b, c7);</div><div class="line"><a name="l01452"></a><span class="lineno"> 1452</span>&#160;<span class="preprocessor">#endif // M0 &gt; 7</span></div><div class="line"><a name="l01453"></a><span class="lineno"> 1453</span>&#160;</div><div class="line"><a name="l01454"></a><span class="lineno"> 1454</span>&#160; lhs_addr += (M0 * LHS_STEP_X * LHS_STEP_LOOP) * <span class="keyword">sizeof</span>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>);</div><div class="line"><a name="l01455"></a><span class="lineno"> 1455</span>&#160; rhs_addr += (N0 * RHS_STEP_X * RHS_STEP_LOOP) * <span class="keyword">sizeof</span>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>);</div><div class="line"><a name="l01456"></a><span class="lineno"> 1456</span>&#160; }</div><div class="line"><a name="l01457"></a><span class="lineno"> 1457</span>&#160;</div><div class="line"><a name="l01458"></a><span class="lineno"> 1458</span>&#160; __global uchar *dst_addr = dst_ptr + dst_offset_first_element_in_bytes + (get_global_id(0) * (uint)N0 * <span class="keyword">sizeof</span>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>)) + (get_global_id(1) * (uint)M0 * dst_stride_y);</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; <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="l01461"></a><span class="lineno"> 1461</span>&#160;</div><div class="line"><a name="l01462"></a><span class="lineno"> 1462</span>&#160;<span class="preprocessor">#if defined(REINTERPRET_OUTPUT_AS_3D)</span></div><div class="line"><a name="l01463"></a><span class="lineno"> 1463</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="l01464"></a><span class="lineno"> 1464</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="l01465"></a><span class="lineno"> 1465</span>&#160; <span class="comment">//</span></div><div class="line"><a name="l01466"></a><span class="lineno"> 1466</span>&#160; <span class="comment">// | |</span></div><div class="line"><a name="l01467"></a><span class="lineno"> 1467</span>&#160; <span class="comment">// | plane0 |</span></div><div class="line"><a name="l01468"></a><span class="lineno"> 1468</span>&#160; <span class="comment">// | |</span></div><div class="line"><a name="l01469"></a><span class="lineno"> 1469</span>&#160; <span class="comment">// |__________________|</span></div><div class="line"><a name="l01470"></a><span class="lineno"> 1470</span>&#160; <span class="comment">// |******************|</span></div><div class="line"><a name="l01471"></a><span class="lineno"> 1471</span>&#160; <span class="comment">// | cross_plane_pad |</span></div><div class="line"><a name="l01472"></a><span class="lineno"> 1472</span>&#160; <span class="comment">// |******************|</span></div><div class="line"><a name="l01473"></a><span class="lineno"> 1473</span>&#160; <span class="comment">// | |</span></div><div class="line"><a name="l01474"></a><span class="lineno"> 1474</span>&#160; <span class="comment">// | plane1 |</span></div><div class="line"><a name="l01475"></a><span class="lineno"> 1475</span>&#160; <span class="comment">// | |</span></div><div class="line"><a name="l01476"></a><span class="lineno"> 1476</span>&#160; <span class="comment">// |__________________|</span></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; <span class="comment">// The plane (zin) is calculated dividing M (y * M0) by HEIGHT_GEMM3D</span></div><div class="line"><a name="l01479"></a><span class="lineno"> 1479</span>&#160; zout0 = (0 + (uint)(get_global_id(1) * (uint)M0)) / (uint)HEIGHT_GEMM3D;</div><div class="line"><a name="l01480"></a><span class="lineno"> 1480</span>&#160; zout0 = min((uint)(DEPTH_GEMM3D - 1), zout0);</div><div class="line"><a name="l01481"></a><span class="lineno"> 1481</span>&#160; zout0 *= (dst_cross_plane_pad * dst_stride_y);</div><div class="line"><a name="l01482"></a><span class="lineno"> 1482</span>&#160;<span class="preprocessor">#if M0 &gt; 1</span></div><div class="line"><a name="l01483"></a><span class="lineno"> 1483</span>&#160; zout1 = (1 + (uint)(get_global_id(1) * (uint)M0)) / (uint)HEIGHT_GEMM3D;</div><div class="line"><a name="l01484"></a><span class="lineno"> 1484</span>&#160; zout1 = min((uint)(DEPTH_GEMM3D - 1), zout1);</div><div class="line"><a name="l01485"></a><span class="lineno"> 1485</span>&#160; zout1 *= (dst_cross_plane_pad * dst_stride_y);</div><div class="line"><a name="l01486"></a><span class="lineno"> 1486</span>&#160;<span class="preprocessor">#endif // M0 &gt; 1</span></div><div class="line"><a name="l01487"></a><span class="lineno"> 1487</span>&#160;<span class="preprocessor">#if M0 &gt; 2</span></div><div class="line"><a name="l01488"></a><span class="lineno"> 1488</span>&#160; zout2 = (2 + (uint)(get_global_id(1) * (uint)M0)) / (uint)HEIGHT_GEMM3D;</div><div class="line"><a name="l01489"></a><span class="lineno"> 1489</span>&#160; zout2 = min((uint)(DEPTH_GEMM3D - 1), zout2);</div><div class="line"><a name="l01490"></a><span class="lineno"> 1490</span>&#160; zout2 *= (dst_cross_plane_pad * dst_stride_y);</div><div class="line"><a name="l01491"></a><span class="lineno"> 1491</span>&#160;<span class="preprocessor">#endif // M0 &gt; 2</span></div><div class="line"><a name="l01492"></a><span class="lineno"> 1492</span>&#160;<span class="preprocessor">#if M0 &gt; 3</span></div><div class="line"><a name="l01493"></a><span class="lineno"> 1493</span>&#160; zout3 = (3 + (uint)(get_global_id(1) * (uint)M0)) / (uint)HEIGHT_GEMM3D;</div><div class="line"><a name="l01494"></a><span class="lineno"> 1494</span>&#160; zout3 = min((uint)(DEPTH_GEMM3D - 1), zout3);</div><div class="line"><a name="l01495"></a><span class="lineno"> 1495</span>&#160; zout3 *= (dst_cross_plane_pad * dst_stride_y);</div><div class="line"><a name="l01496"></a><span class="lineno"> 1496</span>&#160;<span class="preprocessor">#endif // M0 &gt; 3</span></div><div class="line"><a name="l01497"></a><span class="lineno"> 1497</span>&#160;<span class="preprocessor">#if M0 &gt; 4</span></div><div class="line"><a name="l01498"></a><span class="lineno"> 1498</span>&#160; zout4 = (4 + (uint)(get_global_id(1) * (uint)M0)) / (uint)HEIGHT_GEMM3D;</div><div class="line"><a name="l01499"></a><span class="lineno"> 1499</span>&#160; zout4 = min((uint)(DEPTH_GEMM3D - 1), zout4);</div><div class="line"><a name="l01500"></a><span class="lineno"> 1500</span>&#160; zout4 *= (dst_cross_plane_pad * dst_stride_y);</div><div class="line"><a name="l01501"></a><span class="lineno"> 1501</span>&#160;<span class="preprocessor">#endif // M0 &gt; 4</span></div><div class="line"><a name="l01502"></a><span class="lineno"> 1502</span>&#160;<span class="preprocessor">#if M0 &gt; 5</span></div><div class="line"><a name="l01503"></a><span class="lineno"> 1503</span>&#160; zout5 = (5 + (uint)(get_global_id(1) * (uint)M0)) / (uint)HEIGHT_GEMM3D;</div><div class="line"><a name="l01504"></a><span class="lineno"> 1504</span>&#160; zout5 = min((uint)(DEPTH_GEMM3D - 1), zout5);</div><div class="line"><a name="l01505"></a><span class="lineno"> 1505</span>&#160; zout5 *= (dst_cross_plane_pad * dst_stride_y);</div><div class="line"><a name="l01506"></a><span class="lineno"> 1506</span>&#160;<span class="preprocessor">#endif // M0 &gt; 5</span></div><div class="line"><a name="l01507"></a><span class="lineno"> 1507</span>&#160;<span class="preprocessor">#if M0 &gt; 6</span></div><div class="line"><a name="l01508"></a><span class="lineno"> 1508</span>&#160; zout6 = (6 + (uint)(get_global_id(1) * (uint)M0)) / (uint)HEIGHT_GEMM3D;</div><div class="line"><a name="l01509"></a><span class="lineno"> 1509</span>&#160; zout6 = min((uint)(DEPTH_GEMM3D - 1), zout6);</div><div class="line"><a name="l01510"></a><span class="lineno"> 1510</span>&#160; zout6 *= (dst_cross_plane_pad * dst_stride_y);</div><div class="line"><a name="l01511"></a><span class="lineno"> 1511</span>&#160;<span class="preprocessor">#endif // M0 &gt; 6</span></div><div class="line"><a name="l01512"></a><span class="lineno"> 1512</span>&#160;<span class="preprocessor">#if M0 &gt; 7</span></div><div class="line"><a name="l01513"></a><span class="lineno"> 1513</span>&#160; zout7 = (7 + (uint)(get_global_id(1) * (uint)M0)) / (uint)HEIGHT_GEMM3D;</div><div class="line"><a name="l01514"></a><span class="lineno"> 1514</span>&#160; zout7 = min((uint)(DEPTH_GEMM3D - 1), zout7);</div><div class="line"><a name="l01515"></a><span class="lineno"> 1515</span>&#160; zout7 *= (dst_cross_plane_pad * dst_stride_y);</div><div class="line"><a name="l01516"></a><span class="lineno"> 1516</span>&#160;<span class="preprocessor">#endif // M0 &gt; 7</span></div><div class="line"><a name="l01517"></a><span class="lineno"> 1517</span>&#160;</div><div class="line"><a name="l01518"></a><span class="lineno"> 1518</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="l01519"></a><span class="lineno"> 1519</span>&#160; <span class="comment">// multiply dst_stride_z by DEPTH_GEMM3D</span></div><div class="line"><a name="l01520"></a><span class="lineno"> 1520</span>&#160; dst_addr += get_global_id(2) * dst_stride_z * DEPTH_GEMM3D;</div><div class="line"><a name="l01521"></a><span class="lineno"> 1521</span>&#160;</div><div class="line"><a name="l01522"></a><span class="lineno"> 1522</span>&#160;<span class="preprocessor">#else // defined(REINTERPRET_OUTPUT_AS_3D)</span></div><div class="line"><a name="l01523"></a><span class="lineno"> 1523</span>&#160;</div><div class="line"><a name="l01524"></a><span class="lineno"> 1524</span>&#160; <span class="comment">// Add offset for batched GEMM</span></div><div class="line"><a name="l01525"></a><span class="lineno"> 1525</span>&#160; dst_addr += get_global_id(2) * dst_stride_z;</div><div class="line"><a name="l01526"></a><span class="lineno"> 1526</span>&#160;</div><div class="line"><a name="l01527"></a><span class="lineno"> 1527</span>&#160;<span class="preprocessor">#endif // defined(REINTERPRET_OUTPUT_AS_3D)</span></div><div class="line"><a name="l01528"></a><span class="lineno"> 1528</span>&#160;</div><div class="line"><a name="l01529"></a><span class="lineno"> 1529</span>&#160; <span class="comment">// Multiply by the weight of matrix-matrix product and store the result</span></div><div class="line"><a name="l01530"></a><span class="lineno"> 1530</span>&#160;<span class="preprocessor">#if defined(ALPHA)</span></div><div class="line"><a name="l01531"></a><span class="lineno"> 1531</span>&#160; c0 = c0 * (<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>)ALPHA;</div><div class="line"><a name="l01532"></a><span class="lineno"> 1532</span>&#160;<span class="preprocessor">#if M0 &gt; 1</span></div><div class="line"><a name="l01533"></a><span class="lineno"> 1533</span>&#160; c1 = c1 * (<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>)ALPHA;</div><div class="line"><a name="l01534"></a><span class="lineno"> 1534</span>&#160;<span class="preprocessor">#endif // M0 &gt; 1</span></div><div class="line"><a name="l01535"></a><span class="lineno"> 1535</span>&#160;<span class="preprocessor">#if M0 &gt; 2</span></div><div class="line"><a name="l01536"></a><span class="lineno"> 1536</span>&#160; c2 = c2 * (<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>)ALPHA;</div><div class="line"><a name="l01537"></a><span class="lineno"> 1537</span>&#160;<span class="preprocessor">#endif // M0 &gt; 2</span></div><div class="line"><a name="l01538"></a><span class="lineno"> 1538</span>&#160;<span class="preprocessor">#if M0 &gt; 3</span></div><div class="line"><a name="l01539"></a><span class="lineno"> 1539</span>&#160; c3 = c3 * (<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>)ALPHA;</div><div class="line"><a name="l01540"></a><span class="lineno"> 1540</span>&#160;<span class="preprocessor">#endif // M0 &gt; 3</span></div><div class="line"><a name="l01541"></a><span class="lineno"> 1541</span>&#160;<span class="preprocessor">#if M0 &gt; 4</span></div><div class="line"><a name="l01542"></a><span class="lineno"> 1542</span>&#160; c4 = c4 * (<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>)ALPHA;</div><div class="line"><a name="l01543"></a><span class="lineno"> 1543</span>&#160;<span class="preprocessor">#endif // M0 &gt; 4</span></div><div class="line"><a name="l01544"></a><span class="lineno"> 1544</span>&#160;<span class="preprocessor">#if M0 &gt; 5</span></div><div class="line"><a name="l01545"></a><span class="lineno"> 1545</span>&#160; c5 = c5 * (<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>)ALPHA;</div><div class="line"><a name="l01546"></a><span class="lineno"> 1546</span>&#160;<span class="preprocessor">#endif // M0 &gt; 5</span></div><div class="line"><a name="l01547"></a><span class="lineno"> 1547</span>&#160;<span class="preprocessor">#if M0 &gt; 6</span></div><div class="line"><a name="l01548"></a><span class="lineno"> 1548</span>&#160; c6 = c6 * (<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>)ALPHA;</div><div class="line"><a name="l01549"></a><span class="lineno"> 1549</span>&#160;<span class="preprocessor">#endif // M0 &gt; 5</span></div><div class="line"><a name="l01550"></a><span class="lineno"> 1550</span>&#160;<span class="preprocessor">#if M0 &gt; 7</span></div><div class="line"><a name="l01551"></a><span class="lineno"> 1551</span>&#160; c7 = c7 * (<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>)ALPHA;</div><div class="line"><a name="l01552"></a><span class="lineno"> 1552</span>&#160;<span class="preprocessor">#endif // M0 &gt; 7</span></div><div class="line"><a name="l01553"></a><span class="lineno"> 1553</span>&#160;<span class="preprocessor">#endif // defined(ALPHA)</span></div><div class="line"><a name="l01554"></a><span class="lineno"> 1554</span>&#160;</div><div class="line"><a name="l01555"></a><span class="lineno"> 1555</span>&#160; <span class="comment">// Store output block</span></div><div class="line"><a name="l01556"></a><span class="lineno"> 1556</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#acb282042d1edeeaa3cc979a206f78b54">VSTORE</a>(N0)</div><div class="line"><a name="l01557"></a><span class="lineno"> 1557</span>&#160; (c0, 0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 0 * dst_stride_y + zout0));</div><div class="line"><a name="l01558"></a><span class="lineno"> 1558</span>&#160;<span class="preprocessor">#if M0 &gt; 1</span></div><div class="line"><a name="l01559"></a><span class="lineno"> 1559</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#acb282042d1edeeaa3cc979a206f78b54">VSTORE</a>(N0)</div><div class="line"><a name="l01560"></a><span class="lineno"> 1560</span>&#160; (c1, 0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 1 * dst_stride_y + zout1));</div><div class="line"><a name="l01561"></a><span class="lineno"> 1561</span>&#160;<span class="preprocessor">#endif // M0 &gt; 1</span></div><div class="line"><a name="l01562"></a><span class="lineno"> 1562</span>&#160;<span class="preprocessor">#if M0 &gt; 2</span></div><div class="line"><a name="l01563"></a><span class="lineno"> 1563</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#acb282042d1edeeaa3cc979a206f78b54">VSTORE</a>(N0)</div><div class="line"><a name="l01564"></a><span class="lineno"> 1564</span>&#160; (c2, 0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 2 * dst_stride_y + zout2));</div><div class="line"><a name="l01565"></a><span class="lineno"> 1565</span>&#160;<span class="preprocessor">#endif // M0 &gt; 2</span></div><div class="line"><a name="l01566"></a><span class="lineno"> 1566</span>&#160;<span class="preprocessor">#if M0 &gt; 3</span></div><div class="line"><a name="l01567"></a><span class="lineno"> 1567</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#acb282042d1edeeaa3cc979a206f78b54">VSTORE</a>(N0)</div><div class="line"><a name="l01568"></a><span class="lineno"> 1568</span>&#160; (c3, 0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 3 * dst_stride_y + zout3));</div><div class="line"><a name="l01569"></a><span class="lineno"> 1569</span>&#160;<span class="preprocessor">#endif // M0 &gt; 3</span></div><div class="line"><a name="l01570"></a><span class="lineno"> 1570</span>&#160;<span class="preprocessor">#if M0 &gt; 4</span></div><div class="line"><a name="l01571"></a><span class="lineno"> 1571</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#acb282042d1edeeaa3cc979a206f78b54">VSTORE</a>(N0)</div><div class="line"><a name="l01572"></a><span class="lineno"> 1572</span>&#160; (c4, 0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 4 * dst_stride_y + zout4));</div><div class="line"><a name="l01573"></a><span class="lineno"> 1573</span>&#160;<span class="preprocessor">#endif // M0 &gt; 4</span></div><div class="line"><a name="l01574"></a><span class="lineno"> 1574</span>&#160;<span class="preprocessor">#if M0 &gt; 5</span></div><div class="line"><a name="l01575"></a><span class="lineno"> 1575</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#acb282042d1edeeaa3cc979a206f78b54">VSTORE</a>(N0)</div><div class="line"><a name="l01576"></a><span class="lineno"> 1576</span>&#160; (c5, 0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 5 * dst_stride_y + zout5));</div><div class="line"><a name="l01577"></a><span class="lineno"> 1577</span>&#160;<span class="preprocessor">#endif // M0 &gt; 5</span></div><div class="line"><a name="l01578"></a><span class="lineno"> 1578</span>&#160;<span class="preprocessor">#if M0 &gt; 6</span></div><div class="line"><a name="l01579"></a><span class="lineno"> 1579</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#acb282042d1edeeaa3cc979a206f78b54">VSTORE</a>(N0)</div><div class="line"><a name="l01580"></a><span class="lineno"> 1580</span>&#160; (c6, 0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 6 * dst_stride_y + zout6));</div><div class="line"><a name="l01581"></a><span class="lineno"> 1581</span>&#160;<span class="preprocessor">#endif // M0 &gt; 6</span></div><div class="line"><a name="l01582"></a><span class="lineno"> 1582</span>&#160;<span class="preprocessor">#if M0 &gt; 7</span></div><div class="line"><a name="l01583"></a><span class="lineno"> 1583</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#acb282042d1edeeaa3cc979a206f78b54">VSTORE</a>(N0)</div><div class="line"><a name="l01584"></a><span class="lineno"> 1584</span>&#160; (c7, 0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 7 * dst_stride_y + zout7));</div><div class="line"><a name="l01585"></a><span class="lineno"> 1585</span>&#160;<span class="preprocessor">#endif // M0 &gt; 7</span></div><div class="line"><a name="l01586"></a><span class="lineno"> 1586</span>&#160;</div><div class="line"><a name="l01587"></a><span class="lineno"> 1587</span>&#160;<span class="preprocessor">#undef LHS_BLOCK_SIZE</span></div><div class="line"><a name="l01588"></a><span class="lineno"> 1588</span>&#160;<span class="preprocessor">#undef LHS_OFFSET_X</span></div><div class="line"><a name="l01589"></a><span class="lineno"> 1589</span>&#160;<span class="preprocessor">#undef LHS_STEP_X</span></div><div class="line"><a name="l01590"></a><span class="lineno"> 1590</span>&#160;<span class="preprocessor">#undef RHS_BLOCK_SIZE</span></div><div class="line"><a name="l01591"></a><span class="lineno"> 1591</span>&#160;<span class="preprocessor">#undef RHS_OFFSET_X</span></div><div class="line"><a name="l01592"></a><span class="lineno"> 1592</span>&#160;<span class="preprocessor">#undef RHS_STEP_X</span></div><div class="line"><a name="l01593"></a><span class="lineno"> 1593</span>&#160;}</div><div class="line"><a name="l01594"></a><span class="lineno"> 1594</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) &amp;&amp; defined(DATA_TYPE)</span></div><div class="line"><a name="l01595"></a><span class="lineno"> 1595</span>&#160;</div><div class="line"><a name="l01596"></a><span class="lineno"> 1596</span>&#160;<span class="preprocessor">#if defined(TRANSPOSE_W) &amp;&amp; defined(MULT_TRANSPOSE1XW_WIDTH)</span></div><div class="line"><a name="l01597"></a><span class="lineno"> 1597</span>&#160;</div><div class="line"><a name="l01598"></a><span class="lineno"> 1598</span>&#160;<span class="preprocessor">#if ELEMENT_SIZE == 1</span></div><div class="line"><a name="l01599"></a><span class="lineno"> 1599</span>&#160;<span class="preprocessor">#define DATA_TYPE uchar</span></div><div class="line"><a name="l01600"></a><span class="lineno"> 1600</span>&#160;<span class="preprocessor">#elif ELEMENT_SIZE == 2</span></div><div class="line"><a name="l01601"></a><span class="lineno"> 1601</span>&#160;<span class="preprocessor">#define DATA_TYPE ushort</span></div><div class="line"><a name="l01602"></a><span class="lineno"> 1602</span>&#160;<span class="preprocessor">#elif ELEMENT_SIZE == 4</span></div><div class="line"><a name="l01603"></a><span class="lineno"> 1603</span>&#160;<span class="preprocessor">#define DATA_TYPE uint</span></div><div class="line"><a name="l01604"></a><span class="lineno"> 1604</span>&#160;<span class="preprocessor">#else // ELEMENT_SIZE == 1</span></div><div class="line"><a name="l01605"></a><span class="lineno"> 1605</span>&#160;<span class="preprocessor">#error &quot;Element size not supported&quot;</span></div><div class="line"><a name="l01606"></a><span class="lineno"> 1606</span>&#160;<span class="preprocessor">#endif // ELEMENT_SIZE</span></div><div class="line"><a name="l01607"></a><span class="lineno"> 1607</span>&#160;</div><div class="line"><a name="l01630"></a><span class="lineno"> 1630</span>&#160;__kernel <span class="keywordtype">void</span> gemm_transpose1xW(<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#ae9d2dc29c2789c253406f9b304cc75a8">src</a>),</div><div class="line"><a name="l01631"></a><span class="lineno"> 1631</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="l01632"></a><span class="lineno"> 1632</span>&#160;{</div><div class="line"><a name="l01633"></a><span class="lineno"> 1633</span>&#160; uint x = get_global_id(0);</div><div class="line"><a name="l01634"></a><span class="lineno"> 1634</span>&#160; uint y = get_global_id(1);</div><div class="line"><a name="l01635"></a><span class="lineno"> 1635</span>&#160; uint z = get_global_id(2);</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; <span class="comment">// Compute address for Matrix B - source</span></div><div class="line"><a name="l01638"></a><span class="lineno"> 1638</span>&#160; <a class="code" href="struct_tensor3_d.xhtml">Tensor3D</a> <a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#ae9d2dc29c2789c253406f9b304cc75a8">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#ae9d2dc29c2789c253406f9b304cc75a8">src</a>);</div><div class="line"><a name="l01639"></a><span class="lineno"> 1639</span>&#160;</div><div class="line"><a name="l01640"></a><span class="lineno"> 1640</span>&#160; <span class="comment">// Compute address for Matrix B transposed - destination. X and Y are swapped</span></div><div class="line"><a name="l01641"></a><span class="lineno"> 1641</span>&#160; uint dst_addr_in_bytes = dst_offset_first_element_in_bytes + y * TRANSPOSE_W * <span class="keyword">sizeof</span>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>) * MULT_TRANSPOSE1XW_WIDTH + (x / MULT_TRANSPOSE1XW_WIDTH) * dst_stride_y +</div><div class="line"><a name="l01642"></a><span class="lineno"> 1642</span>&#160; (x % MULT_TRANSPOSE1XW_WIDTH) * TRANSPOSE_W * <span class="keyword">sizeof</span>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>);</div><div class="line"><a name="l01643"></a><span class="lineno"> 1643</span>&#160;</div><div class="line"><a name="l01644"></a><span class="lineno"> 1644</span>&#160; <span class="comment">// Add offset for batched GEMM</span></div><div class="line"><a name="l01645"></a><span class="lineno"> 1645</span>&#160; dst_addr_in_bytes += z * dst_stride_z;</div><div class="line"><a name="l01646"></a><span class="lineno"> 1646</span>&#160;</div><div class="line"><a name="l01647"></a><span class="lineno"> 1647</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, TRANSPOSE_W)</div><div class="line"><a name="l01648"></a><span class="lineno"> 1648</span>&#160; b0 = <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a287e2fc366c312b468382c95bb90f91f">VLOAD</a>(TRANSPOSE_W)(0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)<a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#ae9d2dc29c2789c253406f9b304cc75a8">src</a>.ptr);</div><div class="line"><a name="l01649"></a><span class="lineno"> 1649</span>&#160;</div><div class="line"><a name="l01650"></a><span class="lineno"> 1650</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#acb282042d1edeeaa3cc979a206f78b54">VSTORE</a>(TRANSPOSE_W)</div><div class="line"><a name="l01651"></a><span class="lineno"> 1651</span>&#160; (b0, 0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_ptr + dst_addr_in_bytes));</div><div class="line"><a name="l01652"></a><span class="lineno"> 1652</span>&#160;}</div><div class="line"><a name="l01653"></a><span class="lineno"> 1653</span>&#160;<span class="preprocessor">#endif // defined(TRANSPOSE_W) &amp;&amp; defined(MULT_TRANSPOSE1XW_WIDTH)</span></div><div class="line"><a name="l01654"></a><span class="lineno"> 1654</span>&#160;</div><div class="line"><a name="l01655"></a><span class="lineno"> 1655</span>&#160;<span class="preprocessor">#if defined(MULT_INTERLEAVE4X4_HEIGHT) &amp;&amp; defined(DATA_TYPE)</span></div><div class="line"><a name="l01656"></a><span class="lineno"> 1656</span>&#160;</div><div class="line"><a name="l01686"></a><span class="lineno"> 1686</span>&#160;__kernel <span class="keywordtype">void</span> gemm_interleave4x4(<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#ae9d2dc29c2789c253406f9b304cc75a8">src</a>),</div><div class="line"><a name="l01687"></a><span class="lineno"> 1687</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="l01688"></a><span class="lineno"> 1688</span>&#160;#<span class="keywordflow">if</span> defined(REINTERPRET_INPUT_AS_3D)</div><div class="line"><a name="l01689"></a><span class="lineno"> 1689</span>&#160; ,</div><div class="line"><a name="l01690"></a><span class="lineno"> 1690</span>&#160; uint cross_plane_pad</div><div class="line"><a name="l01691"></a><span class="lineno"> 1691</span>&#160;#endif <span class="comment">// REINTERPRET_INPUT_AS_3D</span></div><div class="line"><a name="l01692"></a><span class="lineno"> 1692</span>&#160; )</div><div class="line"><a name="l01693"></a><span class="lineno"> 1693</span>&#160;{</div><div class="line"><a name="l01694"></a><span class="lineno"> 1694</span>&#160; <span class="comment">// Compute source and destination addresses</span></div><div class="line"><a name="l01695"></a><span class="lineno"> 1695</span>&#160; uint x = get_global_id(0);</div><div class="line"><a name="l01696"></a><span class="lineno"> 1696</span>&#160; uint y = get_global_id(1);</div><div class="line"><a name="l01697"></a><span class="lineno"> 1697</span>&#160; uint z = get_global_id(2);</div><div class="line"><a name="l01698"></a><span class="lineno"> 1698</span>&#160;</div><div class="line"><a name="l01699"></a><span class="lineno"> 1699</span>&#160; <span class="comment">// Compute address for source tensor</span></div><div class="line"><a name="l01700"></a><span class="lineno"> 1700</span>&#160; <a class="code" href="struct_tensor3_d.xhtml">Tensor3D</a> <a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#ae9d2dc29c2789c253406f9b304cc75a8">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#ae9d2dc29c2789c253406f9b304cc75a8">src</a>);</div><div class="line"><a name="l01701"></a><span class="lineno"> 1701</span>&#160;</div><div class="line"><a name="l01702"></a><span class="lineno"> 1702</span>&#160; <span class="comment">// Compute address for Matrix B transposed - destination. X and Y are swapped</span></div><div class="line"><a name="l01703"></a><span class="lineno"> 1703</span>&#160; uint dst_addr_in_bytes = dst_offset_first_element_in_bytes + x * <span class="keyword">sizeof</span>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>) * 16 * MULT_INTERLEAVE4X4_HEIGHT + (y / MULT_INTERLEAVE4X4_HEIGHT) * dst_stride_y +</div><div class="line"><a name="l01704"></a><span class="lineno"> 1704</span>&#160; (y % MULT_INTERLEAVE4X4_HEIGHT) * 4 * <span class="keyword">sizeof</span>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>);</div><div class="line"><a name="l01705"></a><span class="lineno"> 1705</span>&#160;</div><div class="line"><a name="l01706"></a><span class="lineno"> 1706</span>&#160; <span class="comment">// Add offset for batched GEMM</span></div><div class="line"><a name="l01707"></a><span class="lineno"> 1707</span>&#160; dst_addr_in_bytes += z * dst_stride_z;</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(REINTERPRET_INPUT_AS_3D)</span></div><div class="line"><a name="l01710"></a><span class="lineno"> 1710</span>&#160; __global uchar *input_ptr = src_ptr + src_offset_first_element_in_bytes + x * 4 * <span class="keyword">sizeof</span>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>) + y * 4 * src_stride_y;</div><div class="line"><a name="l01711"></a><span class="lineno"> 1711</span>&#160;</div><div class="line"><a name="l01712"></a><span class="lineno"> 1712</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="l01713"></a><span class="lineno"> 1713</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="l01714"></a><span class="lineno"> 1714</span>&#160; <span class="comment">//</span></div><div class="line"><a name="l01715"></a><span class="lineno"> 1715</span>&#160; <span class="comment">// | |</span></div><div class="line"><a name="l01716"></a><span class="lineno"> 1716</span>&#160; <span class="comment">// | plane0 |</span></div><div class="line"><a name="l01717"></a><span class="lineno"> 1717</span>&#160; <span class="comment">// | |</span></div><div class="line"><a name="l01718"></a><span class="lineno"> 1718</span>&#160; <span class="comment">// |__________________|</span></div><div class="line"><a name="l01719"></a><span class="lineno"> 1719</span>&#160; <span class="comment">// |******************|</span></div><div class="line"><a name="l01720"></a><span class="lineno"> 1720</span>&#160; <span class="comment">// | cross_plane_pad |</span></div><div class="line"><a name="l01721"></a><span class="lineno"> 1721</span>&#160; <span class="comment">// |******************|</span></div><div class="line"><a name="l01722"></a><span class="lineno"> 1722</span>&#160; <span class="comment">// | |</span></div><div class="line"><a name="l01723"></a><span class="lineno"> 1723</span>&#160; <span class="comment">// | plane1 |</span></div><div class="line"><a name="l01724"></a><span class="lineno"> 1724</span>&#160; <span class="comment">// | |</span></div><div class="line"><a name="l01725"></a><span class="lineno"> 1725</span>&#160; <span class="comment">// |__________________|</span></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="comment">// The plane (zin) is calculated dividing M (y * 4) by HEIGHT_GEMM3D</span></div><div class="line"><a name="l01728"></a><span class="lineno"> 1728</span>&#160; uint4 zin = ((uint4)(0, 1, 2, 3) + (uint4)(y * 4)) / (uint4)HEIGHT_GEMM3D;</div><div class="line"><a name="l01729"></a><span class="lineno"> 1729</span>&#160; zin = min(DEPTH_GEMM3D - 1, zin);</div><div class="line"><a name="l01730"></a><span class="lineno"> 1730</span>&#160;</div><div class="line"><a name="l01731"></a><span class="lineno"> 1731</span>&#160; <span class="comment">// Add offset due to the cross plane paddings</span></div><div class="line"><a name="l01732"></a><span class="lineno"> 1732</span>&#160; zin *= (cross_plane_pad * src_stride_y);</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">// 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="l01735"></a><span class="lineno"> 1735</span>&#160; <span class="comment">// multiply src_stride_z by DEPTH_GEMM3D</span></div><div class="line"><a name="l01736"></a><span class="lineno"> 1736</span>&#160; input_ptr += z * src_stride_z * DEPTH_GEMM3D;</div><div class="line"><a name="l01737"></a><span class="lineno"> 1737</span>&#160;</div><div class="line"><a name="l01738"></a><span class="lineno"> 1738</span>&#160; <span class="comment">// Load values from Matrix A</span></div><div class="line"><a name="l01739"></a><span class="lineno"> 1739</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, 4)</div><div class="line"><a name="l01740"></a><span class="lineno"> 1740</span>&#160; a0 = vload4(0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(input_ptr + 0 * src_stride_y + zin.s0));</div><div class="line"><a name="l01741"></a><span class="lineno"> 1741</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, 4)</div><div class="line"><a name="l01742"></a><span class="lineno"> 1742</span>&#160; a1 = vload4(0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(input_ptr + 1 * src_stride_y + zin.s1));</div><div class="line"><a name="l01743"></a><span class="lineno"> 1743</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, 4)</div><div class="line"><a name="l01744"></a><span class="lineno"> 1744</span>&#160; a2 = vload4(0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(input_ptr + 2 * src_stride_y + zin.s2));</div><div class="line"><a name="l01745"></a><span class="lineno"> 1745</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, 4)</div><div class="line"><a name="l01746"></a><span class="lineno"> 1746</span>&#160; a3 = vload4(0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(input_ptr + 3 * src_stride_y + zin.s3));</div><div class="line"><a name="l01747"></a><span class="lineno"> 1747</span>&#160;<span class="preprocessor">#else // defined(REINTERPRET_INPUT_AS_3D)</span></div><div class="line"><a name="l01748"></a><span class="lineno"> 1748</span>&#160; __global uchar *input_ptr = <a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#ae9d2dc29c2789c253406f9b304cc75a8">src</a>.ptr;</div><div class="line"><a name="l01749"></a><span class="lineno"> 1749</span>&#160;</div><div class="line"><a name="l01750"></a><span class="lineno"> 1750</span>&#160; <span class="comment">// Load values from Matrix A</span></div><div class="line"><a name="l01751"></a><span class="lineno"> 1751</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, 4)</div><div class="line"><a name="l01752"></a><span class="lineno"> 1752</span>&#160; a0 = vload4(0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(input_ptr + 0 * src_stride_y));</div><div class="line"><a name="l01753"></a><span class="lineno"> 1753</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, 4)</div><div class="line"><a name="l01754"></a><span class="lineno"> 1754</span>&#160; a1 = vload4(0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(input_ptr + 1 * src_stride_y));</div><div class="line"><a name="l01755"></a><span class="lineno"> 1755</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, 4)</div><div class="line"><a name="l01756"></a><span class="lineno"> 1756</span>&#160; a2 = vload4(0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(input_ptr + 2 * src_stride_y));</div><div class="line"><a name="l01757"></a><span class="lineno"> 1757</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, 4)</div><div class="line"><a name="l01758"></a><span class="lineno"> 1758</span>&#160; a3 = vload4(0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(input_ptr + 3 * src_stride_y));</div><div class="line"><a name="l01759"></a><span class="lineno"> 1759</span>&#160;<span class="preprocessor">#endif // defined(REINTERPRET_INPUT_AS_3D)</span></div><div class="line"><a name="l01760"></a><span class="lineno"> 1760</span>&#160;</div><div class="line"><a name="l01761"></a><span class="lineno"> 1761</span>&#160;<span class="preprocessor">#if defined(UNROLL_BLOCK)</span></div><div class="line"><a name="l01762"></a><span class="lineno"> 1762</span>&#160; vstore4(a0, 0, ((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_ptr + dst_addr_in_bytes) + 0 * MULT_INTERLEAVE4X4_HEIGHT));</div><div class="line"><a name="l01763"></a><span class="lineno"> 1763</span>&#160; vstore4(a1, 0, ((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_ptr + dst_addr_in_bytes) + 4 * MULT_INTERLEAVE4X4_HEIGHT));</div><div class="line"><a name="l01764"></a><span class="lineno"> 1764</span>&#160; vstore4(a2, 0, ((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_ptr + dst_addr_in_bytes) + 8 * MULT_INTERLEAVE4X4_HEIGHT));</div><div class="line"><a name="l01765"></a><span class="lineno"> 1765</span>&#160; vstore4(a3, 0, ((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_ptr + dst_addr_in_bytes) + 12 * MULT_INTERLEAVE4X4_HEIGHT));</div><div class="line"><a name="l01766"></a><span class="lineno"> 1766</span>&#160;<span class="preprocessor">#else // defined(UNROLL_BLOCK)</span></div><div class="line"><a name="l01767"></a><span class="lineno"> 1767</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, 4)</div><div class="line"><a name="l01768"></a><span class="lineno"> 1768</span>&#160; val0 = (<a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, 4))(a0.s0, a1.s0, a2.s0, a3.s0);</div><div class="line"><a name="l01769"></a><span class="lineno"> 1769</span>&#160; vstore4(val0, 0, ((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_ptr + dst_addr_in_bytes) + 0 * MULT_INTERLEAVE4X4_HEIGHT));</div><div class="line"><a name="l01770"></a><span class="lineno"> 1770</span>&#160;</div><div class="line"><a name="l01771"></a><span class="lineno"> 1771</span>&#160; val0 = (<a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, 4))(a0.s1, a1.s1, a2.s1, a3.s1);</div><div class="line"><a name="l01772"></a><span class="lineno"> 1772</span>&#160; vstore4(val0, 0, ((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_ptr + dst_addr_in_bytes) + 4 * MULT_INTERLEAVE4X4_HEIGHT));</div><div class="line"><a name="l01773"></a><span class="lineno"> 1773</span>&#160;</div><div class="line"><a name="l01774"></a><span class="lineno"> 1774</span>&#160; val0 = (<a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, 4))(a0.s2, a1.s2, a2.s2, a3.s2);</div><div class="line"><a name="l01775"></a><span class="lineno"> 1775</span>&#160; vstore4(val0, 0, ((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_ptr + dst_addr_in_bytes) + 8 * MULT_INTERLEAVE4X4_HEIGHT));</div><div class="line"><a name="l01776"></a><span class="lineno"> 1776</span>&#160;</div><div class="line"><a name="l01777"></a><span class="lineno"> 1777</span>&#160; val0 = (<a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, 4))(a0.s3, a1.s3, a2.s3, a3.s3);</div><div class="line"><a name="l01778"></a><span class="lineno"> 1778</span>&#160; vstore4(val0, 0, ((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_ptr + dst_addr_in_bytes) + 12 * MULT_INTERLEAVE4X4_HEIGHT));</div><div class="line"><a name="l01779"></a><span class="lineno"> 1779</span>&#160;<span class="preprocessor">#endif // defined(UNROLL_BLOCK)</span></div><div class="line"><a name="l01780"></a><span class="lineno"> 1780</span>&#160;}</div><div class="line"><a name="l01781"></a><span class="lineno"> 1781</span>&#160;<span class="preprocessor">#endif // defined(MULT_INTERLEAVE4X4_HEIGHT) &amp;&amp; defined(DATA_TYPE)</span></div><div class="line"><a name="l01782"></a><span class="lineno"> 1782</span>&#160;</div><div class="line"><a name="l01783"></a><span class="lineno"> 1783</span>&#160;<span class="preprocessor">#if defined(COLS_B) &amp;&amp; defined(MULT_TRANSPOSE1XW_WIDTH) &amp;&amp; defined(MULT_INTERLEAVE4X4_HEIGHT)</span></div><div class="line"><a name="l01784"></a><span class="lineno"> 1784</span>&#160;</div><div class="line"><a name="l01830"></a><span class="lineno"> 1830</span>&#160;__kernel <span class="keywordtype">void</span> gemm_mm_interleaved_transposed_f32(<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="l01831"></a><span class="lineno"> 1831</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="l01832"></a><span class="lineno"> 1832</span>&#160;#<span class="keywordflow">if</span> defined(ADD_VEC_C)</div><div class="line"><a name="l01833"></a><span class="lineno"> 1833</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a40a6eb9f2a7712f08d6bb8ff6c9e6ca7">VECTOR_DECLARATION</a>(src2),</div><div class="line"><a name="l01834"></a><span class="lineno"> 1834</span>&#160;#endif <span class="comment">/* defined(ADD_VEC_C) */</span></div><div class="line"><a name="l01835"></a><span class="lineno"> 1835</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="l01836"></a><span class="lineno"> 1836</span>&#160; uint src0_stride_z,</div><div class="line"><a name="l01837"></a><span class="lineno"> 1837</span>&#160; uint src1_stride_z,</div><div class="line"><a name="l01838"></a><span class="lineno"> 1838</span>&#160; uint dst_stride_z</div><div class="line"><a name="l01839"></a><span class="lineno"> 1839</span>&#160;#<span class="keywordflow">if</span> defined(REINTERPRET_OUTPUT_AS_3D)</div><div class="line"><a name="l01840"></a><span class="lineno"> 1840</span>&#160; ,</div><div class="line"><a name="l01841"></a><span class="lineno"> 1841</span>&#160; uint cross_plane_pad</div><div class="line"><a name="l01842"></a><span class="lineno"> 1842</span>&#160;#endif <span class="comment">// REINTERPRET_OUTPUT_AS_3D</span></div><div class="line"><a name="l01843"></a><span class="lineno"> 1843</span>&#160; )</div><div class="line"><a name="l01844"></a><span class="lineno"> 1844</span>&#160;{</div><div class="line"><a name="l01845"></a><span class="lineno"> 1845</span>&#160; <span class="keywordtype">int</span> x = get_global_id(0) / MULT_TRANSPOSE1XW_WIDTH;</div><div class="line"><a name="l01846"></a><span class="lineno"> 1846</span>&#160; <span class="keywordtype">int</span> y = get_global_id(1) / MULT_INTERLEAVE4X4_HEIGHT;</div><div class="line"><a name="l01847"></a><span class="lineno"> 1847</span>&#160; <span class="keywordtype">int</span> z = get_global_id(2);</div><div class="line"><a name="l01848"></a><span class="lineno"> 1848</span>&#160;</div><div class="line"><a name="l01849"></a><span class="lineno"> 1849</span>&#160; <span class="comment">// Offset</span></div><div class="line"><a name="l01850"></a><span class="lineno"> 1850</span>&#160; <span class="keyword">const</span> <span class="keywordtype">int</span> offset_row_a = (get_global_id(1) % MULT_INTERLEAVE4X4_HEIGHT) * 4;</div><div class="line"><a name="l01851"></a><span class="lineno"> 1851</span>&#160; <span class="keyword">const</span> <span class="keywordtype">int</span> offset_row_b = (get_global_id(0) % MULT_TRANSPOSE1XW_WIDTH) * 4;</div><div class="line"><a name="l01852"></a><span class="lineno"> 1852</span>&#160;</div><div class="line"><a name="l01853"></a><span class="lineno"> 1853</span>&#160; <span class="comment">// src_addr_a = address of matrix A</span></div><div class="line"><a name="l01854"></a><span class="lineno"> 1854</span>&#160; <span class="comment">// src_addr_b = address of matrix B</span></div><div class="line"><a name="l01855"></a><span class="lineno"> 1855</span>&#160; <span class="keywordtype">int</span> src0_addr_in_bytes = z * src0_stride_z + y * src0_stride_y + src0_offset_first_element_in_bytes;</div><div class="line"><a name="l01856"></a><span class="lineno"> 1856</span>&#160; <span class="keywordtype">int</span> src1_addr_in_bytes = x * src1_stride_y + src1_offset_first_element_in_bytes;</div><div class="line"><a name="l01857"></a><span class="lineno"> 1857</span>&#160;</div><div class="line"><a name="l01858"></a><span class="lineno"> 1858</span>&#160;<span class="preprocessor">#if defined(MATRIX_B_DEPTH)</span></div><div class="line"><a name="l01859"></a><span class="lineno"> 1859</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="l01860"></a><span class="lineno"> 1860</span>&#160; src1_addr_in_bytes += (z % MATRIX_B_DEPTH) * src1_stride_z;</div><div class="line"><a name="l01861"></a><span class="lineno"> 1861</span>&#160;<span class="preprocessor">#else // defined(MATRIX_B_DEPTH)</span></div><div class="line"><a name="l01862"></a><span class="lineno"> 1862</span>&#160; src1_addr_in_bytes += z * src1_stride_z;</div><div class="line"><a name="l01863"></a><span class="lineno"> 1863</span>&#160;<span class="preprocessor">#endif // defined(MATRIX_B_DEPTH)</span></div><div class="line"><a name="l01864"></a><span class="lineno"> 1864</span>&#160;</div><div class="line"><a name="l01865"></a><span class="lineno"> 1865</span>&#160; __global <span class="keywordtype">float</span> *src_addr_a = (__global <span class="keywordtype">float</span> *)(src0_ptr + src0_addr_in_bytes);</div><div class="line"><a name="l01866"></a><span class="lineno"> 1866</span>&#160; __global <span class="keywordtype">float</span> *src_addr_b = (__global <span class="keywordtype">float</span> *)(src1_ptr + src1_addr_in_bytes);</div><div class="line"><a name="l01867"></a><span class="lineno"> 1867</span>&#160;</div><div class="line"><a name="l01868"></a><span class="lineno"> 1868</span>&#160; <span class="comment">// Compute end row address for matrix B</span></div><div class="line"><a name="l01869"></a><span class="lineno"> 1869</span>&#160; __global <span class="keywordtype">float</span> *src_end_addr_b = src_addr_b + COLS_B;</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; src_addr_a += offset_row_a;</div><div class="line"><a name="l01872"></a><span class="lineno"> 1872</span>&#160; src_addr_b += offset_row_b;</div><div class="line"><a name="l01873"></a><span class="lineno"> 1873</span>&#160;</div><div class="line"><a name="l01874"></a><span class="lineno"> 1874</span>&#160; <span class="comment">// Reset accumulators</span></div><div class="line"><a name="l01875"></a><span class="lineno"> 1875</span>&#160; float4 c00 = 0.0f;</div><div class="line"><a name="l01876"></a><span class="lineno"> 1876</span>&#160; float4 c10 = 0.0f;</div><div class="line"><a name="l01877"></a><span class="lineno"> 1877</span>&#160; float4 c20 = 0.0f;</div><div class="line"><a name="l01878"></a><span class="lineno"> 1878</span>&#160; float4 c30 = 0.0f;</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; <span class="keywordflow">for</span>(; src_addr_b &lt;= (src_end_addr_b - (int)(8 * MULT_TRANSPOSE1XW_WIDTH)); src_addr_a += 8 * MULT_INTERLEAVE4X4_HEIGHT, src_addr_b += 8 * MULT_TRANSPOSE1XW_WIDTH)</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="comment">// Load values from matrix A (interleaved) and matrix B (transposed)</span></div><div class="line"><a name="l01883"></a><span class="lineno"> 1883</span>&#160; float4 a0 = vload4(0, src_addr_a);</div><div class="line"><a name="l01884"></a><span class="lineno"> 1884</span>&#160; float4 b0 = vload4(0, src_addr_b);</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; c00 += (float4)a0.s0 * b0;</div><div class="line"><a name="l01887"></a><span class="lineno"> 1887</span>&#160; c10 += (float4)a0.s1 * b0;</div><div class="line"><a name="l01888"></a><span class="lineno"> 1888</span>&#160; c20 += (float4)a0.s2 * b0;</div><div class="line"><a name="l01889"></a><span class="lineno"> 1889</span>&#160; c30 += (float4)a0.s3 * b0;</div><div class="line"><a name="l01890"></a><span class="lineno"> 1890</span>&#160;</div><div class="line"><a name="l01891"></a><span class="lineno"> 1891</span>&#160; <span class="comment">// Load values from matrix A (interleaved) and matrix B (transposed)</span></div><div class="line"><a name="l01892"></a><span class="lineno"> 1892</span>&#160; a0 = vload4(0, src_addr_a + 4 * MULT_INTERLEAVE4X4_HEIGHT);</div><div class="line"><a name="l01893"></a><span class="lineno"> 1893</span>&#160; b0 = vload4(0, src_addr_b + 4 * MULT_TRANSPOSE1XW_WIDTH);</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; c00 += (float4)a0.s0 * b0;</div><div class="line"><a name="l01896"></a><span class="lineno"> 1896</span>&#160; c10 += (float4)a0.s1 * b0;</div><div class="line"><a name="l01897"></a><span class="lineno"> 1897</span>&#160; c20 += (float4)a0.s2 * b0;</div><div class="line"><a name="l01898"></a><span class="lineno"> 1898</span>&#160; c30 += (float4)a0.s3 * b0;</div><div class="line"><a name="l01899"></a><span class="lineno"> 1899</span>&#160; }</div><div class="line"><a name="l01900"></a><span class="lineno"> 1900</span>&#160;</div><div class="line"><a name="l01901"></a><span class="lineno"> 1901</span>&#160; <span class="keywordflow">for</span>(; src_addr_b &lt; src_end_addr_b; src_addr_a += 4 * MULT_INTERLEAVE4X4_HEIGHT, src_addr_b += 4 * MULT_TRANSPOSE1XW_WIDTH)</div><div class="line"><a name="l01902"></a><span class="lineno"> 1902</span>&#160; {</div><div class="line"><a name="l01903"></a><span class="lineno"> 1903</span>&#160; <span class="comment">// Load values from matrix A (interleaved) and matrix B (transposed)</span></div><div class="line"><a name="l01904"></a><span class="lineno"> 1904</span>&#160; float4 a0 = vload4(0, src_addr_a);</div><div class="line"><a name="l01905"></a><span class="lineno"> 1905</span>&#160; float4 b0 = vload4(0, src_addr_b);</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; c00 += (float4)a0.s0 * b0;</div><div class="line"><a name="l01908"></a><span class="lineno"> 1908</span>&#160; c10 += (float4)a0.s1 * b0;</div><div class="line"><a name="l01909"></a><span class="lineno"> 1909</span>&#160; c20 += (float4)a0.s2 * b0;</div><div class="line"><a name="l01910"></a><span class="lineno"> 1910</span>&#160; c30 += (float4)a0.s3 * b0;</div><div class="line"><a name="l01911"></a><span class="lineno"> 1911</span>&#160; }</div><div class="line"><a name="l01912"></a><span class="lineno"> 1912</span>&#160;</div><div class="line"><a name="l01913"></a><span class="lineno"> 1913</span>&#160; <span class="comment">// Compute destination address</span></div><div class="line"><a name="l01914"></a><span class="lineno"> 1914</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="l01915"></a><span class="lineno"> 1915</span>&#160;</div><div class="line"><a name="l01916"></a><span class="lineno"> 1916</span>&#160;<span class="preprocessor">#if defined(ALPHA)</span></div><div class="line"><a name="l01917"></a><span class="lineno"> 1917</span>&#160; <span class="comment">// Multiply by the weight of matrix product</span></div><div class="line"><a name="l01918"></a><span class="lineno"> 1918</span>&#160; c00 = c00 * (float4)ALPHA;</div><div class="line"><a name="l01919"></a><span class="lineno"> 1919</span>&#160; c10 = c10 * (float4)ALPHA;</div><div class="line"><a name="l01920"></a><span class="lineno"> 1920</span>&#160; c20 = c20 * (float4)ALPHA;</div><div class="line"><a name="l01921"></a><span class="lineno"> 1921</span>&#160; c30 = c30 * (float4)ALPHA;</div><div class="line"><a name="l01922"></a><span class="lineno"> 1922</span>&#160;<span class="preprocessor">#endif // defined(ALPHA)</span></div><div class="line"><a name="l01923"></a><span class="lineno"> 1923</span>&#160;</div><div class="line"><a name="l01924"></a><span class="lineno"> 1924</span>&#160;<span class="preprocessor">#if defined(ADD_VEC_C)</span></div><div class="line"><a name="l01925"></a><span class="lineno"> 1925</span>&#160; __global <span class="keywordtype">float</span> *src2_addr = (__global <span class="keywordtype">float</span> *)(src2_ptr + src2_offset_first_element_in_bytes + get_global_id(0) * src2_step_x);</div><div class="line"><a name="l01926"></a><span class="lineno"> 1926</span>&#160; float4 c0 = vload4(0, src2_addr);</div><div class="line"><a name="l01927"></a><span class="lineno"> 1927</span>&#160;</div><div class="line"><a name="l01928"></a><span class="lineno"> 1928</span>&#160; c00 += c0;</div><div class="line"><a name="l01929"></a><span class="lineno"> 1929</span>&#160; c10 += c0;</div><div class="line"><a name="l01930"></a><span class="lineno"> 1930</span>&#160; c20 += c0;</div><div class="line"><a name="l01931"></a><span class="lineno"> 1931</span>&#160; c30 += c0;</div><div class="line"><a name="l01932"></a><span class="lineno"> 1932</span>&#160;<span class="preprocessor">#endif </span><span class="comment">/* defined(ADD_VEC_C) */</span><span class="preprocessor"></span></div><div class="line"><a name="l01933"></a><span class="lineno"> 1933</span>&#160;</div><div class="line"><a name="l01934"></a><span class="lineno"> 1934</span>&#160; <span class="comment">// Compute dst address</span></div><div class="line"><a name="l01935"></a><span class="lineno"> 1935</span>&#160; __global uchar *dst_addr = <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>(&amp;<a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#adbf67dcee294e673cf796f1ed8aeb6a4">dst</a>, 0, 0);</div><div class="line"><a name="l01936"></a><span class="lineno"> 1936</span>&#160;</div><div class="line"><a name="l01937"></a><span class="lineno"> 1937</span>&#160;<span class="preprocessor">#if defined(REINTERPRET_OUTPUT_AS_3D)</span></div><div class="line"><a name="l01938"></a><span class="lineno"> 1938</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="l01939"></a><span class="lineno"> 1939</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="l01940"></a><span class="lineno"> 1940</span>&#160; <span class="comment">//</span></div><div class="line"><a name="l01941"></a><span class="lineno"> 1941</span>&#160; <span class="comment">// | |</span></div><div class="line"><a name="l01942"></a><span class="lineno"> 1942</span>&#160; <span class="comment">// | plane0 |</span></div><div class="line"><a name="l01943"></a><span class="lineno"> 1943</span>&#160; <span class="comment">// | |</span></div><div class="line"><a name="l01944"></a><span class="lineno"> 1944</span>&#160; <span class="comment">// |__________________|</span></div><div class="line"><a name="l01945"></a><span class="lineno"> 1945</span>&#160; <span class="comment">// |******************|</span></div><div class="line"><a name="l01946"></a><span class="lineno"> 1946</span>&#160; <span class="comment">// | cross_plane_pad |</span></div><div class="line"><a name="l01947"></a><span class="lineno"> 1947</span>&#160; <span class="comment">// |******************|</span></div><div class="line"><a name="l01948"></a><span class="lineno"> 1948</span>&#160; <span class="comment">// | |</span></div><div class="line"><a name="l01949"></a><span class="lineno"> 1949</span>&#160; <span class="comment">// | plane1 |</span></div><div class="line"><a name="l01950"></a><span class="lineno"> 1950</span>&#160; <span class="comment">// | |</span></div><div class="line"><a name="l01951"></a><span class="lineno"> 1951</span>&#160; <span class="comment">// |__________________|</span></div><div class="line"><a name="l01952"></a><span class="lineno"> 1952</span>&#160;</div><div class="line"><a name="l01953"></a><span class="lineno"> 1953</span>&#160; <span class="comment">// The plane (zout) is calculated dividing M (get_global_id(1) * 4) by HEIGHT_GEMM3D</span></div><div class="line"><a name="l01954"></a><span class="lineno"> 1954</span>&#160; uint4 zout = ((uint4)(0, 1, 2, 3) + (uint4)(get_global_id(1) * 4)) / (uint4)HEIGHT_GEMM3D;</div><div class="line"><a name="l01955"></a><span class="lineno"> 1955</span>&#160; zout = min(DEPTH_GEMM3D - 1, zout);</div><div class="line"><a name="l01956"></a><span class="lineno"> 1956</span>&#160;</div><div class="line"><a name="l01957"></a><span class="lineno"> 1957</span>&#160; <span class="comment">// Add offset due to the cross plane paddings</span></div><div class="line"><a name="l01958"></a><span class="lineno"> 1958</span>&#160; zout *= (cross_plane_pad * dst_stride_y);</div><div class="line"><a name="l01959"></a><span class="lineno"> 1959</span>&#160;</div><div class="line"><a name="l01960"></a><span class="lineno"> 1960</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="l01961"></a><span class="lineno"> 1961</span>&#160; <span class="comment">// multiply dst_stride_z by DEPTH_GEMM3D</span></div><div class="line"><a name="l01962"></a><span class="lineno"> 1962</span>&#160; dst_addr += z * dst_stride_z * DEPTH_GEMM3D;</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; <span class="comment">// Store 4x4 block</span></div><div class="line"><a name="l01965"></a><span class="lineno"> 1965</span>&#160; vstore4(c00, 0, (__global <span class="keywordtype">float</span> *)(dst_addr + 0 * dst_stride_y + zout.s0));</div><div class="line"><a name="l01966"></a><span class="lineno"> 1966</span>&#160; vstore4(c10, 0, (__global <span class="keywordtype">float</span> *)(dst_addr + 1 * dst_stride_y + zout.s1));</div><div class="line"><a name="l01967"></a><span class="lineno"> 1967</span>&#160; vstore4(c20, 0, (__global <span class="keywordtype">float</span> *)(dst_addr + 2 * dst_stride_y + zout.s2));</div><div class="line"><a name="l01968"></a><span class="lineno"> 1968</span>&#160; vstore4(c30, 0, (__global <span class="keywordtype">float</span> *)(dst_addr + 3 * dst_stride_y + zout.s3));</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">#else // defined(REINTERPRET_OUTPUT_AS_3D)</span></div><div class="line"><a name="l01971"></a><span class="lineno"> 1971</span>&#160; <span class="comment">// Add offset for batched GEMM</span></div><div class="line"><a name="l01972"></a><span class="lineno"> 1972</span>&#160; dst_addr += z * dst_stride_z;</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; <span class="comment">// Store 4x4 block</span></div><div class="line"><a name="l01975"></a><span class="lineno"> 1975</span>&#160; vstore4(c00, 0, (__global <span class="keywordtype">float</span> *)(dst_addr + 0 * dst_stride_y));</div><div class="line"><a name="l01976"></a><span class="lineno"> 1976</span>&#160; vstore4(c10, 0, (__global <span class="keywordtype">float</span> *)(dst_addr + 1 * dst_stride_y));</div><div class="line"><a name="l01977"></a><span class="lineno"> 1977</span>&#160; vstore4(c20, 0, (__global <span class="keywordtype">float</span> *)(dst_addr + 2 * dst_stride_y));</div><div class="line"><a name="l01978"></a><span class="lineno"> 1978</span>&#160; vstore4(c30, 0, (__global <span class="keywordtype">float</span> *)(dst_addr + 3 * dst_stride_y));</div><div class="line"><a name="l01979"></a><span class="lineno"> 1979</span>&#160;<span class="preprocessor">#endif // defined(REINTERPRET_OUTPUT_AS_3D)</span></div><div class="line"><a name="l01980"></a><span class="lineno"> 1980</span>&#160;}</div><div class="line"><a name="l01981"></a><span class="lineno"> 1981</span>&#160;</div><div class="line"><a name="l02029"></a><span class="lineno"> 2029</span>&#160;__kernel <span class="keywordtype">void</span> gemm_mm_interleaved_transposed_f32_bifrost(<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="l02030"></a><span class="lineno"> 2030</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="l02031"></a><span class="lineno"> 2031</span>&#160;#<span class="keywordflow">if</span> defined(ADD_VEC_C)</div><div class="line"><a name="l02032"></a><span class="lineno"> 2032</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a40a6eb9f2a7712f08d6bb8ff6c9e6ca7">VECTOR_DECLARATION</a>(src2),</div><div class="line"><a name="l02033"></a><span class="lineno"> 2033</span>&#160;#endif <span class="comment">/* defined(ADD_VEC_C) */</span></div><div class="line"><a name="l02034"></a><span class="lineno"> 2034</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="l02035"></a><span class="lineno"> 2035</span>&#160; uint src0_stride_z,</div><div class="line"><a name="l02036"></a><span class="lineno"> 2036</span>&#160; uint src1_stride_z,</div><div class="line"><a name="l02037"></a><span class="lineno"> 2037</span>&#160; uint dst_stride_z</div><div class="line"><a name="l02038"></a><span class="lineno"> 2038</span>&#160;#<span class="keywordflow">if</span> defined(REINTERPRET_OUTPUT_AS_3D)</div><div class="line"><a name="l02039"></a><span class="lineno"> 2039</span>&#160; ,</div><div class="line"><a name="l02040"></a><span class="lineno"> 2040</span>&#160; uint cross_plane_pad</div><div class="line"><a name="l02041"></a><span class="lineno"> 2041</span>&#160;#endif <span class="comment">// REINTERPRET_OUTPUT_AS_3D</span></div><div class="line"><a name="l02042"></a><span class="lineno"> 2042</span>&#160; )</div><div class="line"><a name="l02043"></a><span class="lineno"> 2043</span>&#160;{</div><div class="line"><a name="l02044"></a><span class="lineno"> 2044</span>&#160; <span class="keywordtype">int</span> x = get_global_id(0) / MULT_TRANSPOSE1XW_WIDTH;</div><div class="line"><a name="l02045"></a><span class="lineno"> 2045</span>&#160; <span class="keywordtype">int</span> y = get_global_id(1) / MULT_INTERLEAVE4X4_HEIGHT;</div><div class="line"><a name="l02046"></a><span class="lineno"> 2046</span>&#160; <span class="keywordtype">int</span> z = get_global_id(2);</div><div class="line"><a name="l02047"></a><span class="lineno"> 2047</span>&#160;</div><div class="line"><a name="l02048"></a><span class="lineno"> 2048</span>&#160; <span class="comment">// Offset</span></div><div class="line"><a name="l02049"></a><span class="lineno"> 2049</span>&#160; <span class="keyword">const</span> <span class="keywordtype">int</span> offset_row_a = (get_global_id(1) % MULT_INTERLEAVE4X4_HEIGHT) * 4;</div><div class="line"><a name="l02050"></a><span class="lineno"> 2050</span>&#160; <span class="keyword">const</span> <span class="keywordtype">int</span> offset_row_b = (get_global_id(0) % MULT_TRANSPOSE1XW_WIDTH) * 4;</div><div class="line"><a name="l02051"></a><span class="lineno"> 2051</span>&#160;</div><div class="line"><a name="l02052"></a><span class="lineno"> 2052</span>&#160; <span class="comment">// src_addr_a = address of matrix A</span></div><div class="line"><a name="l02053"></a><span class="lineno"> 2053</span>&#160; <span class="comment">// src_addr_b = address of matrix B</span></div><div class="line"><a name="l02054"></a><span class="lineno"> 2054</span>&#160; <span class="keywordtype">int</span> src0_addr_in_bytes = z * src0_stride_z + y * src0_stride_y + src0_offset_first_element_in_bytes;</div><div class="line"><a name="l02055"></a><span class="lineno"> 2055</span>&#160; <span class="keywordtype">int</span> src1_addr_in_bytes = x * src1_stride_y + src1_offset_first_element_in_bytes;</div><div class="line"><a name="l02056"></a><span class="lineno"> 2056</span>&#160;</div><div class="line"><a name="l02057"></a><span class="lineno"> 2057</span>&#160;<span class="preprocessor">#if defined(MATRIX_B_DEPTH)</span></div><div class="line"><a name="l02058"></a><span class="lineno"> 2058</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="l02059"></a><span class="lineno"> 2059</span>&#160; src1_addr_in_bytes += (z % MATRIX_B_DEPTH) * src1_stride_z;</div><div class="line"><a name="l02060"></a><span class="lineno"> 2060</span>&#160;<span class="preprocessor">#else // defined(MATRIX_B_DEPTH)</span></div><div class="line"><a name="l02061"></a><span class="lineno"> 2061</span>&#160; src1_addr_in_bytes += z * src1_stride_z;</div><div class="line"><a name="l02062"></a><span class="lineno"> 2062</span>&#160;<span class="preprocessor">#endif // defined(MATRIX_B_DEPTH)</span></div><div class="line"><a name="l02063"></a><span class="lineno"> 2063</span>&#160;</div><div class="line"><a name="l02064"></a><span class="lineno"> 2064</span>&#160; __global <span class="keywordtype">float</span> *src_addr_a = (__global <span class="keywordtype">float</span> *)(src0_ptr + src0_addr_in_bytes);</div><div class="line"><a name="l02065"></a><span class="lineno"> 2065</span>&#160; __global <span class="keywordtype">float</span> *src_addr_b = (__global <span class="keywordtype">float</span> *)(src1_ptr + src1_addr_in_bytes);</div><div class="line"><a name="l02066"></a><span class="lineno"> 2066</span>&#160;</div><div class="line"><a name="l02067"></a><span class="lineno"> 2067</span>&#160; src_addr_a += offset_row_a;</div><div class="line"><a name="l02068"></a><span class="lineno"> 2068</span>&#160; src_addr_b += offset_row_b;</div><div class="line"><a name="l02069"></a><span class="lineno"> 2069</span>&#160;</div><div class="line"><a name="l02070"></a><span class="lineno"> 2070</span>&#160; <span class="comment">// Reset accumulators</span></div><div class="line"><a name="l02071"></a><span class="lineno"> 2071</span>&#160; <span class="keywordtype">float</span> c00 = 0.0f;</div><div class="line"><a name="l02072"></a><span class="lineno"> 2072</span>&#160; <span class="keywordtype">float</span> c01 = 0.0f;</div><div class="line"><a name="l02073"></a><span class="lineno"> 2073</span>&#160; <span class="keywordtype">float</span> c02 = 0.0f;</div><div class="line"><a name="l02074"></a><span class="lineno"> 2074</span>&#160; <span class="keywordtype">float</span> c03 = 0.0f;</div><div class="line"><a name="l02075"></a><span class="lineno"> 2075</span>&#160; <span class="keywordtype">float</span> c10 = 0.0f;</div><div class="line"><a name="l02076"></a><span class="lineno"> 2076</span>&#160; <span class="keywordtype">float</span> c11 = 0.0f;</div><div class="line"><a name="l02077"></a><span class="lineno"> 2077</span>&#160; <span class="keywordtype">float</span> c12 = 0.0f;</div><div class="line"><a name="l02078"></a><span class="lineno"> 2078</span>&#160; <span class="keywordtype">float</span> c13 = 0.0f;</div><div class="line"><a name="l02079"></a><span class="lineno"> 2079</span>&#160; <span class="keywordtype">float</span> c20 = 0.0f;</div><div class="line"><a name="l02080"></a><span class="lineno"> 2080</span>&#160; <span class="keywordtype">float</span> c21 = 0.0f;</div><div class="line"><a name="l02081"></a><span class="lineno"> 2081</span>&#160; <span class="keywordtype">float</span> c22 = 0.0f;</div><div class="line"><a name="l02082"></a><span class="lineno"> 2082</span>&#160; <span class="keywordtype">float</span> c23 = 0.0f;</div><div class="line"><a name="l02083"></a><span class="lineno"> 2083</span>&#160; <span class="keywordtype">float</span> c30 = 0.0f;</div><div class="line"><a name="l02084"></a><span class="lineno"> 2084</span>&#160; <span class="keywordtype">float</span> c31 = 0.0f;</div><div class="line"><a name="l02085"></a><span class="lineno"> 2085</span>&#160; <span class="keywordtype">float</span> c32 = 0.0f;</div><div class="line"><a name="l02086"></a><span class="lineno"> 2086</span>&#160; <span class="keywordtype">float</span> c33 = 0.0f;</div><div class="line"><a name="l02087"></a><span class="lineno"> 2087</span>&#160;</div><div class="line"><a name="l02088"></a><span class="lineno"> 2088</span>&#160;<span class="preprocessor">#define COLS_MTX_B (COLS_B / (4 * MULT_TRANSPOSE1XW_WIDTH))</span></div><div class="line"><a name="l02089"></a><span class="lineno"> 2089</span>&#160;</div><div class="line"><a name="l02090"></a><span class="lineno"> 2090</span>&#160; <span class="keywordtype">int</span> i = 0;</div><div class="line"><a name="l02091"></a><span class="lineno"> 2091</span>&#160; <span class="keywordflow">for</span>(; i &lt;= (int)(COLS_MTX_B - 4); i += 4)</div><div class="line"><a name="l02092"></a><span class="lineno"> 2092</span>&#160; {</div><div class="line"><a name="l02093"></a><span class="lineno"> 2093</span>&#160; <span class="comment">// Load values from matrix A (interleaved) and matrix B (transposed)</span></div><div class="line"><a name="l02094"></a><span class="lineno"> 2094</span>&#160; float4 a0 = vload4(0, src_addr_a);</div><div class="line"><a name="l02095"></a><span class="lineno"> 2095</span>&#160; float4 b0 = vload4(0, src_addr_b);</div><div class="line"><a name="l02096"></a><span class="lineno"> 2096</span>&#160;</div><div class="line"><a name="l02097"></a><span class="lineno"> 2097</span>&#160; src_addr_a += 4 * MULT_INTERLEAVE4X4_HEIGHT;</div><div class="line"><a name="l02098"></a><span class="lineno"> 2098</span>&#160; src_addr_b += 4 * MULT_TRANSPOSE1XW_WIDTH;</div><div class="line"><a name="l02099"></a><span class="lineno"> 2099</span>&#160;</div><div class="line"><a name="l02100"></a><span class="lineno"> 2100</span>&#160; c00 = fma(a0.s0, b0.s0, c00);</div><div class="line"><a name="l02101"></a><span class="lineno"> 2101</span>&#160; c01 = fma(a0.s0, b0.s1, c01);</div><div class="line"><a name="l02102"></a><span class="lineno"> 2102</span>&#160; c02 = fma(a0.s0, b0.s2, c02);</div><div class="line"><a name="l02103"></a><span class="lineno"> 2103</span>&#160; c03 = fma(a0.s0, b0.s3, c03);</div><div class="line"><a name="l02104"></a><span class="lineno"> 2104</span>&#160;</div><div class="line"><a name="l02105"></a><span class="lineno"> 2105</span>&#160; c10 = fma(a0.s1, b0.s0, c10);</div><div class="line"><a name="l02106"></a><span class="lineno"> 2106</span>&#160; c11 = fma(a0.s1, b0.s1, c11);</div><div class="line"><a name="l02107"></a><span class="lineno"> 2107</span>&#160; c12 = fma(a0.s1, b0.s2, c12);</div><div class="line"><a name="l02108"></a><span class="lineno"> 2108</span>&#160; c13 = fma(a0.s1, b0.s3, c13);</div><div class="line"><a name="l02109"></a><span class="lineno"> 2109</span>&#160;</div><div class="line"><a name="l02110"></a><span class="lineno"> 2110</span>&#160; c20 = fma(a0.s2, b0.s0, c20);</div><div class="line"><a name="l02111"></a><span class="lineno"> 2111</span>&#160; c21 = fma(a0.s2, b0.s1, c21);</div><div class="line"><a name="l02112"></a><span class="lineno"> 2112</span>&#160; c22 = fma(a0.s2, b0.s2, c22);</div><div class="line"><a name="l02113"></a><span class="lineno"> 2113</span>&#160; c23 = fma(a0.s2, b0.s3, c23);</div><div class="line"><a name="l02114"></a><span class="lineno"> 2114</span>&#160;</div><div class="line"><a name="l02115"></a><span class="lineno"> 2115</span>&#160; c30 = fma(a0.s3, b0.s0, c30);</div><div class="line"><a name="l02116"></a><span class="lineno"> 2116</span>&#160; c31 = fma(a0.s3, b0.s1, c31);</div><div class="line"><a name="l02117"></a><span class="lineno"> 2117</span>&#160; c32 = fma(a0.s3, b0.s2, c32);</div><div class="line"><a name="l02118"></a><span class="lineno"> 2118</span>&#160; c33 = fma(a0.s3, b0.s3, c33);</div><div class="line"><a name="l02119"></a><span class="lineno"> 2119</span>&#160;</div><div class="line"><a name="l02120"></a><span class="lineno"> 2120</span>&#160; <span class="comment">// Load values from matrix A (interleaved) and matrix B (transposed)</span></div><div class="line"><a name="l02121"></a><span class="lineno"> 2121</span>&#160; a0 = vload4(0, src_addr_a);</div><div class="line"><a name="l02122"></a><span class="lineno"> 2122</span>&#160; b0 = vload4(0, src_addr_b);</div><div class="line"><a name="l02123"></a><span class="lineno"> 2123</span>&#160;</div><div class="line"><a name="l02124"></a><span class="lineno"> 2124</span>&#160; src_addr_a += 4 * MULT_INTERLEAVE4X4_HEIGHT;</div><div class="line"><a name="l02125"></a><span class="lineno"> 2125</span>&#160; src_addr_b += 4 * MULT_TRANSPOSE1XW_WIDTH;</div><div class="line"><a name="l02126"></a><span class="lineno"> 2126</span>&#160;</div><div class="line"><a name="l02127"></a><span class="lineno"> 2127</span>&#160; c00 = fma(a0.s0, b0.s0, c00);</div><div class="line"><a name="l02128"></a><span class="lineno"> 2128</span>&#160; c01 = fma(a0.s0, b0.s1, c01);</div><div class="line"><a name="l02129"></a><span class="lineno"> 2129</span>&#160; c02 = fma(a0.s0, b0.s2, c02);</div><div class="line"><a name="l02130"></a><span class="lineno"> 2130</span>&#160; c03 = fma(a0.s0, b0.s3, c03);</div><div class="line"><a name="l02131"></a><span class="lineno"> 2131</span>&#160;</div><div class="line"><a name="l02132"></a><span class="lineno"> 2132</span>&#160; c10 = fma(a0.s1, b0.s0, c10);</div><div class="line"><a name="l02133"></a><span class="lineno"> 2133</span>&#160; c11 = fma(a0.s1, b0.s1, c11);</div><div class="line"><a name="l02134"></a><span class="lineno"> 2134</span>&#160; c12 = fma(a0.s1, b0.s2, c12);</div><div class="line"><a name="l02135"></a><span class="lineno"> 2135</span>&#160; c13 = fma(a0.s1, b0.s3, c13);</div><div class="line"><a name="l02136"></a><span class="lineno"> 2136</span>&#160;</div><div class="line"><a name="l02137"></a><span class="lineno"> 2137</span>&#160; c20 = fma(a0.s2, b0.s0, c20);</div><div class="line"><a name="l02138"></a><span class="lineno"> 2138</span>&#160; c21 = fma(a0.s2, b0.s1, c21);</div><div class="line"><a name="l02139"></a><span class="lineno"> 2139</span>&#160; c22 = fma(a0.s2, b0.s2, c22);</div><div class="line"><a name="l02140"></a><span class="lineno"> 2140</span>&#160; c23 = fma(a0.s2, b0.s3, c23);</div><div class="line"><a name="l02141"></a><span class="lineno"> 2141</span>&#160;</div><div class="line"><a name="l02142"></a><span class="lineno"> 2142</span>&#160; c30 = fma(a0.s3, b0.s0, c30);</div><div class="line"><a name="l02143"></a><span class="lineno"> 2143</span>&#160; c31 = fma(a0.s3, b0.s1, c31);</div><div class="line"><a name="l02144"></a><span class="lineno"> 2144</span>&#160; c32 = fma(a0.s3, b0.s2, c32);</div><div class="line"><a name="l02145"></a><span class="lineno"> 2145</span>&#160; c33 = fma(a0.s3, b0.s3, c33);</div><div class="line"><a name="l02146"></a><span class="lineno"> 2146</span>&#160;</div><div class="line"><a name="l02147"></a><span class="lineno"> 2147</span>&#160; <span class="comment">// Load values from matrix A (interleaved) and matrix B (transposed)</span></div><div class="line"><a name="l02148"></a><span class="lineno"> 2148</span>&#160; a0 = vload4(0, src_addr_a);</div><div class="line"><a name="l02149"></a><span class="lineno"> 2149</span>&#160; b0 = vload4(0, src_addr_b);</div><div class="line"><a name="l02150"></a><span class="lineno"> 2150</span>&#160;</div><div class="line"><a name="l02151"></a><span class="lineno"> 2151</span>&#160; src_addr_a += 4 * MULT_INTERLEAVE4X4_HEIGHT;</div><div class="line"><a name="l02152"></a><span class="lineno"> 2152</span>&#160; src_addr_b += 4 * MULT_TRANSPOSE1XW_WIDTH;</div><div class="line"><a name="l02153"></a><span class="lineno"> 2153</span>&#160;</div><div class="line"><a name="l02154"></a><span class="lineno"> 2154</span>&#160; c00 = fma(a0.s0, b0.s0, c00);</div><div class="line"><a name="l02155"></a><span class="lineno"> 2155</span>&#160; c01 = fma(a0.s0, b0.s1, c01);</div><div class="line"><a name="l02156"></a><span class="lineno"> 2156</span>&#160; c02 = fma(a0.s0, b0.s2, c02);</div><div class="line"><a name="l02157"></a><span class="lineno"> 2157</span>&#160; c03 = fma(a0.s0, b0.s3, c03);</div><div class="line"><a name="l02158"></a><span class="lineno"> 2158</span>&#160;</div><div class="line"><a name="l02159"></a><span class="lineno"> 2159</span>&#160; c10 = fma(a0.s1, b0.s0, c10);</div><div class="line"><a name="l02160"></a><span class="lineno"> 2160</span>&#160; c11 = fma(a0.s1, b0.s1, c11);</div><div class="line"><a name="l02161"></a><span class="lineno"> 2161</span>&#160; c12 = fma(a0.s1, b0.s2, c12);</div><div class="line"><a name="l02162"></a><span class="lineno"> 2162</span>&#160; c13 = fma(a0.s1, b0.s3, c13);</div><div class="line"><a name="l02163"></a><span class="lineno"> 2163</span>&#160;</div><div class="line"><a name="l02164"></a><span class="lineno"> 2164</span>&#160; c20 = fma(a0.s2, b0.s0, c20);</div><div class="line"><a name="l02165"></a><span class="lineno"> 2165</span>&#160; c21 = fma(a0.s2, b0.s1, c21);</div><div class="line"><a name="l02166"></a><span class="lineno"> 2166</span>&#160; c22 = fma(a0.s2, b0.s2, c22);</div><div class="line"><a name="l02167"></a><span class="lineno"> 2167</span>&#160; c23 = fma(a0.s2, b0.s3, c23);</div><div class="line"><a name="l02168"></a><span class="lineno"> 2168</span>&#160;</div><div class="line"><a name="l02169"></a><span class="lineno"> 2169</span>&#160; c30 = fma(a0.s3, b0.s0, c30);</div><div class="line"><a name="l02170"></a><span class="lineno"> 2170</span>&#160; c31 = fma(a0.s3, b0.s1, c31);</div><div class="line"><a name="l02171"></a><span class="lineno"> 2171</span>&#160; c32 = fma(a0.s3, b0.s2, c32);</div><div class="line"><a name="l02172"></a><span class="lineno"> 2172</span>&#160; c33 = fma(a0.s3, b0.s3, c33);</div><div class="line"><a name="l02173"></a><span class="lineno"> 2173</span>&#160;</div><div class="line"><a name="l02174"></a><span class="lineno"> 2174</span>&#160; <span class="comment">// Load values from matrix A (interleaved) and matrix B (transposed)</span></div><div class="line"><a name="l02175"></a><span class="lineno"> 2175</span>&#160; a0 = vload4(0, src_addr_a);</div><div class="line"><a name="l02176"></a><span class="lineno"> 2176</span>&#160; b0 = vload4(0, src_addr_b);</div><div class="line"><a name="l02177"></a><span class="lineno"> 2177</span>&#160;</div><div class="line"><a name="l02178"></a><span class="lineno"> 2178</span>&#160; src_addr_a += 4 * MULT_INTERLEAVE4X4_HEIGHT;</div><div class="line"><a name="l02179"></a><span class="lineno"> 2179</span>&#160; src_addr_b += 4 * MULT_TRANSPOSE1XW_WIDTH;</div><div class="line"><a name="l02180"></a><span class="lineno"> 2180</span>&#160;</div><div class="line"><a name="l02181"></a><span class="lineno"> 2181</span>&#160; c00 = fma(a0.s0, b0.s0, c00);</div><div class="line"><a name="l02182"></a><span class="lineno"> 2182</span>&#160; c01 = fma(a0.s0, b0.s1, c01);</div><div class="line"><a name="l02183"></a><span class="lineno"> 2183</span>&#160; c02 = fma(a0.s0, b0.s2, c02);</div><div class="line"><a name="l02184"></a><span class="lineno"> 2184</span>&#160; c03 = fma(a0.s0, b0.s3, c03);</div><div class="line"><a name="l02185"></a><span class="lineno"> 2185</span>&#160;</div><div class="line"><a name="l02186"></a><span class="lineno"> 2186</span>&#160; c10 = fma(a0.s1, b0.s0, c10);</div><div class="line"><a name="l02187"></a><span class="lineno"> 2187</span>&#160; c11 = fma(a0.s1, b0.s1, c11);</div><div class="line"><a name="l02188"></a><span class="lineno"> 2188</span>&#160; c12 = fma(a0.s1, b0.s2, c12);</div><div class="line"><a name="l02189"></a><span class="lineno"> 2189</span>&#160; c13 = fma(a0.s1, b0.s3, c13);</div><div class="line"><a name="l02190"></a><span class="lineno"> 2190</span>&#160;</div><div class="line"><a name="l02191"></a><span class="lineno"> 2191</span>&#160; c20 = fma(a0.s2, b0.s0, c20);</div><div class="line"><a name="l02192"></a><span class="lineno"> 2192</span>&#160; c21 = fma(a0.s2, b0.s1, c21);</div><div class="line"><a name="l02193"></a><span class="lineno"> 2193</span>&#160; c22 = fma(a0.s2, b0.s2, c22);</div><div class="line"><a name="l02194"></a><span class="lineno"> 2194</span>&#160; c23 = fma(a0.s2, b0.s3, c23);</div><div class="line"><a name="l02195"></a><span class="lineno"> 2195</span>&#160;</div><div class="line"><a name="l02196"></a><span class="lineno"> 2196</span>&#160; c30 = fma(a0.s3, b0.s0, c30);</div><div class="line"><a name="l02197"></a><span class="lineno"> 2197</span>&#160; c31 = fma(a0.s3, b0.s1, c31);</div><div class="line"><a name="l02198"></a><span class="lineno"> 2198</span>&#160; c32 = fma(a0.s3, b0.s2, c32);</div><div class="line"><a name="l02199"></a><span class="lineno"> 2199</span>&#160; c33 = fma(a0.s3, b0.s3, c33);</div><div class="line"><a name="l02200"></a><span class="lineno"> 2200</span>&#160; }</div><div class="line"><a name="l02201"></a><span class="lineno"> 2201</span>&#160;</div><div class="line"><a name="l02202"></a><span class="lineno"> 2202</span>&#160; <span class="keywordflow">for</span>(; i &lt; (int)(COLS_MTX_B); ++i)</div><div class="line"><a name="l02203"></a><span class="lineno"> 2203</span>&#160; {</div><div class="line"><a name="l02204"></a><span class="lineno"> 2204</span>&#160; <span class="comment">// Load values from matrix A (interleaved) and matrix B (transposed)</span></div><div class="line"><a name="l02205"></a><span class="lineno"> 2205</span>&#160; float4 a0 = vload4(0, src_addr_a);</div><div class="line"><a name="l02206"></a><span class="lineno"> 2206</span>&#160; float4 b0 = vload4(0, src_addr_b);</div><div class="line"><a name="l02207"></a><span class="lineno"> 2207</span>&#160;</div><div class="line"><a name="l02208"></a><span class="lineno"> 2208</span>&#160; src_addr_a += 4 * MULT_INTERLEAVE4X4_HEIGHT;</div><div class="line"><a name="l02209"></a><span class="lineno"> 2209</span>&#160; src_addr_b += 4 * MULT_TRANSPOSE1XW_WIDTH;</div><div class="line"><a name="l02210"></a><span class="lineno"> 2210</span>&#160;</div><div class="line"><a name="l02211"></a><span class="lineno"> 2211</span>&#160; c00 = fma(a0.s0, b0.s0, c00);</div><div class="line"><a name="l02212"></a><span class="lineno"> 2212</span>&#160; c01 = fma(a0.s0, b0.s1, c01);</div><div class="line"><a name="l02213"></a><span class="lineno"> 2213</span>&#160; c02 = fma(a0.s0, b0.s2, c02);</div><div class="line"><a name="l02214"></a><span class="lineno"> 2214</span>&#160; c03 = fma(a0.s0, b0.s3, c03);</div><div class="line"><a name="l02215"></a><span class="lineno"> 2215</span>&#160;</div><div class="line"><a name="l02216"></a><span class="lineno"> 2216</span>&#160; c10 = fma(a0.s1, b0.s0, c10);</div><div class="line"><a name="l02217"></a><span class="lineno"> 2217</span>&#160; c11 = fma(a0.s1, b0.s1, c11);</div><div class="line"><a name="l02218"></a><span class="lineno"> 2218</span>&#160; c12 = fma(a0.s1, b0.s2, c12);</div><div class="line"><a name="l02219"></a><span class="lineno"> 2219</span>&#160; c13 = fma(a0.s1, b0.s3, c13);</div><div class="line"><a name="l02220"></a><span class="lineno"> 2220</span>&#160;</div><div class="line"><a name="l02221"></a><span class="lineno"> 2221</span>&#160; c20 = fma(a0.s2, b0.s0, c20);</div><div class="line"><a name="l02222"></a><span class="lineno"> 2222</span>&#160; c21 = fma(a0.s2, b0.s1, c21);</div><div class="line"><a name="l02223"></a><span class="lineno"> 2223</span>&#160; c22 = fma(a0.s2, b0.s2, c22);</div><div class="line"><a name="l02224"></a><span class="lineno"> 2224</span>&#160; c23 = fma(a0.s2, b0.s3, c23);</div><div class="line"><a name="l02225"></a><span class="lineno"> 2225</span>&#160;</div><div class="line"><a name="l02226"></a><span class="lineno"> 2226</span>&#160; c30 = fma(a0.s3, b0.s0, c30);</div><div class="line"><a name="l02227"></a><span class="lineno"> 2227</span>&#160; c31 = fma(a0.s3, b0.s1, c31);</div><div class="line"><a name="l02228"></a><span class="lineno"> 2228</span>&#160; c32 = fma(a0.s3, b0.s2, c32);</div><div class="line"><a name="l02229"></a><span class="lineno"> 2229</span>&#160; c33 = fma(a0.s3, b0.s3, c33);</div><div class="line"><a name="l02230"></a><span class="lineno"> 2230</span>&#160; }</div><div class="line"><a name="l02231"></a><span class="lineno"> 2231</span>&#160;</div><div class="line"><a name="l02232"></a><span class="lineno"> 2232</span>&#160; <span class="comment">// Compute destination address</span></div><div class="line"><a name="l02233"></a><span class="lineno"> 2233</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="l02234"></a><span class="lineno"> 2234</span>&#160;</div><div class="line"><a name="l02235"></a><span class="lineno"> 2235</span>&#160;<span class="preprocessor">#if defined(ALPHA)</span></div><div class="line"><a name="l02236"></a><span class="lineno"> 2236</span>&#160; <span class="comment">// Multiply by the weight of matrix product</span></div><div class="line"><a name="l02237"></a><span class="lineno"> 2237</span>&#160; c00 = c00 * ALPHA;</div><div class="line"><a name="l02238"></a><span class="lineno"> 2238</span>&#160; c01 = c01 * ALPHA;</div><div class="line"><a name="l02239"></a><span class="lineno"> 2239</span>&#160; c02 = c02 * ALPHA;</div><div class="line"><a name="l02240"></a><span class="lineno"> 2240</span>&#160; c03 = c03 * ALPHA;</div><div class="line"><a name="l02241"></a><span class="lineno"> 2241</span>&#160; c10 = c10 * ALPHA;</div><div class="line"><a name="l02242"></a><span class="lineno"> 2242</span>&#160; c11 = c11 * ALPHA;</div><div class="line"><a name="l02243"></a><span class="lineno"> 2243</span>&#160; c12 = c12 * ALPHA;</div><div class="line"><a name="l02244"></a><span class="lineno"> 2244</span>&#160; c13 = c13 * ALPHA;</div><div class="line"><a name="l02245"></a><span class="lineno"> 2245</span>&#160; c20 = c20 * ALPHA;</div><div class="line"><a name="l02246"></a><span class="lineno"> 2246</span>&#160; c21 = c21 * ALPHA;</div><div class="line"><a name="l02247"></a><span class="lineno"> 2247</span>&#160; c22 = c22 * ALPHA;</div><div class="line"><a name="l02248"></a><span class="lineno"> 2248</span>&#160; c23 = c23 * ALPHA;</div><div class="line"><a name="l02249"></a><span class="lineno"> 2249</span>&#160; c30 = c30 * ALPHA;</div><div class="line"><a name="l02250"></a><span class="lineno"> 2250</span>&#160; c31 = c31 * ALPHA;</div><div class="line"><a name="l02251"></a><span class="lineno"> 2251</span>&#160; c32 = c32 * ALPHA;</div><div class="line"><a name="l02252"></a><span class="lineno"> 2252</span>&#160; c33 = c33 * ALPHA;</div><div class="line"><a name="l02253"></a><span class="lineno"> 2253</span>&#160;<span class="preprocessor">#endif // defined(ALPHA)</span></div><div class="line"><a name="l02254"></a><span class="lineno"> 2254</span>&#160;</div><div class="line"><a name="l02255"></a><span class="lineno"> 2255</span>&#160; <span class="comment">// Compute dst address</span></div><div class="line"><a name="l02256"></a><span class="lineno"> 2256</span>&#160; __global uchar *dst_addr = <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>(&amp;<a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#adbf67dcee294e673cf796f1ed8aeb6a4">dst</a>, 0, 0);</div><div class="line"><a name="l02257"></a><span class="lineno"> 2257</span>&#160;</div><div class="line"><a name="l02258"></a><span class="lineno"> 2258</span>&#160;<span class="preprocessor">#if defined(ADD_VEC_C)</span></div><div class="line"><a name="l02259"></a><span class="lineno"> 2259</span>&#160; __global <span class="keywordtype">float</span> *src2_addr = (__global <span class="keywordtype">float</span> *)(src2_ptr + src2_offset_first_element_in_bytes + get_global_id(0) * src2_step_x);</div><div class="line"><a name="l02260"></a><span class="lineno"> 2260</span>&#160; float4 c0 = vload4(0, src2_addr);</div><div class="line"><a name="l02261"></a><span class="lineno"> 2261</span>&#160;</div><div class="line"><a name="l02262"></a><span class="lineno"> 2262</span>&#160; c00 += c0.s0;</div><div class="line"><a name="l02263"></a><span class="lineno"> 2263</span>&#160; c01 += c0.s1;</div><div class="line"><a name="l02264"></a><span class="lineno"> 2264</span>&#160; c02 += c0.s2;</div><div class="line"><a name="l02265"></a><span class="lineno"> 2265</span>&#160; c03 += c0.s3;</div><div class="line"><a name="l02266"></a><span class="lineno"> 2266</span>&#160; c10 += c0.s0;</div><div class="line"><a name="l02267"></a><span class="lineno"> 2267</span>&#160; c11 += c0.s1;</div><div class="line"><a name="l02268"></a><span class="lineno"> 2268</span>&#160; c12 += c0.s2;</div><div class="line"><a name="l02269"></a><span class="lineno"> 2269</span>&#160; c13 += c0.s3;</div><div class="line"><a name="l02270"></a><span class="lineno"> 2270</span>&#160; c20 += c0.s0;</div><div class="line"><a name="l02271"></a><span class="lineno"> 2271</span>&#160; c21 += c0.s1;</div><div class="line"><a name="l02272"></a><span class="lineno"> 2272</span>&#160; c22 += c0.s2;</div><div class="line"><a name="l02273"></a><span class="lineno"> 2273</span>&#160; c23 += c0.s3;</div><div class="line"><a name="l02274"></a><span class="lineno"> 2274</span>&#160; c30 += c0.s0;</div><div class="line"><a name="l02275"></a><span class="lineno"> 2275</span>&#160; c31 += c0.s1;</div><div class="line"><a name="l02276"></a><span class="lineno"> 2276</span>&#160; c32 += c0.s2;</div><div class="line"><a name="l02277"></a><span class="lineno"> 2277</span>&#160; c33 += c0.s3;</div><div class="line"><a name="l02278"></a><span class="lineno"> 2278</span>&#160;<span class="preprocessor">#endif </span><span class="comment">/* defined(ADD_VEC_C) */</span><span class="preprocessor"></span></div><div class="line"><a name="l02279"></a><span class="lineno"> 2279</span>&#160;</div><div class="line"><a name="l02280"></a><span class="lineno"> 2280</span>&#160;<span class="preprocessor">#if defined(REINTERPRET_OUTPUT_AS_3D)</span></div><div class="line"><a name="l02281"></a><span class="lineno"> 2281</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="l02282"></a><span class="lineno"> 2282</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="l02283"></a><span class="lineno"> 2283</span>&#160; <span class="comment">//</span></div><div class="line"><a name="l02284"></a><span class="lineno"> 2284</span>&#160; <span class="comment">// | |</span></div><div class="line"><a name="l02285"></a><span class="lineno"> 2285</span>&#160; <span class="comment">// | plane0 |</span></div><div class="line"><a name="l02286"></a><span class="lineno"> 2286</span>&#160; <span class="comment">// | |</span></div><div class="line"><a name="l02287"></a><span class="lineno"> 2287</span>&#160; <span class="comment">// |__________________|</span></div><div class="line"><a name="l02288"></a><span class="lineno"> 2288</span>&#160; <span class="comment">// |******************|</span></div><div class="line"><a name="l02289"></a><span class="lineno"> 2289</span>&#160; <span class="comment">// | cross_plane_pad |</span></div><div class="line"><a name="l02290"></a><span class="lineno"> 2290</span>&#160; <span class="comment">// |******************|</span></div><div class="line"><a name="l02291"></a><span class="lineno"> 2291</span>&#160; <span class="comment">// | |</span></div><div class="line"><a name="l02292"></a><span class="lineno"> 2292</span>&#160; <span class="comment">// | plane1 |</span></div><div class="line"><a name="l02293"></a><span class="lineno"> 2293</span>&#160; <span class="comment">// | |</span></div><div class="line"><a name="l02294"></a><span class="lineno"> 2294</span>&#160; <span class="comment">// |__________________|</span></div><div class="line"><a name="l02295"></a><span class="lineno"> 2295</span>&#160;</div><div class="line"><a name="l02296"></a><span class="lineno"> 2296</span>&#160; <span class="comment">// The plane (zout) is calculated dividing M (get_global_id(1) * 4) by HEIGHT_GEMM3D</span></div><div class="line"><a name="l02297"></a><span class="lineno"> 2297</span>&#160; uint4 zout = ((uint4)(0, 1, 2, 3) + (uint4)(get_global_id(1) * 4)) / (uint4)HEIGHT_GEMM3D;</div><div class="line"><a name="l02298"></a><span class="lineno"> 2298</span>&#160; zout = min(DEPTH_GEMM3D - 1, zout);</div><div class="line"><a name="l02299"></a><span class="lineno"> 2299</span>&#160;</div><div class="line"><a name="l02300"></a><span class="lineno"> 2300</span>&#160; <span class="comment">// Add offset due to the cross plane paddings</span></div><div class="line"><a name="l02301"></a><span class="lineno"> 2301</span>&#160; zout *= (cross_plane_pad * dst_stride_y);</div><div class="line"><a name="l02302"></a><span class="lineno"> 2302</span>&#160;</div><div class="line"><a name="l02303"></a><span class="lineno"> 2303</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="l02304"></a><span class="lineno"> 2304</span>&#160; <span class="comment">// multiply dst_stride_z by DEPTH_GEMM3D</span></div><div class="line"><a name="l02305"></a><span class="lineno"> 2305</span>&#160; dst_addr += z * dst_stride_z * DEPTH_GEMM3D;</div><div class="line"><a name="l02306"></a><span class="lineno"> 2306</span>&#160;</div><div class="line"><a name="l02307"></a><span class="lineno"> 2307</span>&#160; <span class="comment">// Store 4x4 block</span></div><div class="line"><a name="l02308"></a><span class="lineno"> 2308</span>&#160; vstore4((float4)(c00, c01, c02, c03), 0, (__global <span class="keywordtype">float</span> *)(dst_addr + 0 * dst_stride_y + zout.s0));</div><div class="line"><a name="l02309"></a><span class="lineno"> 2309</span>&#160; vstore4((float4)(c10, c11, c12, c13), 0, (__global <span class="keywordtype">float</span> *)(dst_addr + 1 * dst_stride_y + zout.s1));</div><div class="line"><a name="l02310"></a><span class="lineno"> 2310</span>&#160; vstore4((float4)(c20, c21, c22, c23), 0, (__global <span class="keywordtype">float</span> *)(dst_addr + 2 * dst_stride_y + zout.s2));</div><div class="line"><a name="l02311"></a><span class="lineno"> 2311</span>&#160; vstore4((float4)(c30, c31, c32, c33), 0, (__global <span class="keywordtype">float</span> *)(dst_addr + 3 * dst_stride_y + zout.s3));</div><div class="line"><a name="l02312"></a><span class="lineno"> 2312</span>&#160;</div><div class="line"><a name="l02313"></a><span class="lineno"> 2313</span>&#160;<span class="preprocessor">#else // defined(REINTERPRET_OUTPUT_AS_3D)</span></div><div class="line"><a name="l02314"></a><span class="lineno"> 2314</span>&#160; <span class="comment">// Add offset for batched GEMM</span></div><div class="line"><a name="l02315"></a><span class="lineno"> 2315</span>&#160; dst_addr += z * dst_stride_z;</div><div class="line"><a name="l02316"></a><span class="lineno"> 2316</span>&#160;</div><div class="line"><a name="l02317"></a><span class="lineno"> 2317</span>&#160; <span class="comment">// Store 4x4 block</span></div><div class="line"><a name="l02318"></a><span class="lineno"> 2318</span>&#160; vstore4((float4)(c00, c01, c02, c03), 0, (__global <span class="keywordtype">float</span> *)(dst_addr + 0 * dst_stride_y));</div><div class="line"><a name="l02319"></a><span class="lineno"> 2319</span>&#160; vstore4((float4)(c10, c11, c12, c13), 0, (__global <span class="keywordtype">float</span> *)(dst_addr + 1 * dst_stride_y));</div><div class="line"><a name="l02320"></a><span class="lineno"> 2320</span>&#160; vstore4((float4)(c20, c21, c22, c23), 0, (__global <span class="keywordtype">float</span> *)(dst_addr + 2 * dst_stride_y));</div><div class="line"><a name="l02321"></a><span class="lineno"> 2321</span>&#160; vstore4((float4)(c30, c31, c32, c33), 0, (__global <span class="keywordtype">float</span> *)(dst_addr + 3 * dst_stride_y));</div><div class="line"><a name="l02322"></a><span class="lineno"> 2322</span>&#160;<span class="preprocessor">#endif // defined(REINTERPRET_OUTPUT_AS_3D)</span></div><div class="line"><a name="l02323"></a><span class="lineno"> 2323</span>&#160;}</div><div class="line"><a name="l02324"></a><span class="lineno"> 2324</span>&#160;</div><div class="line"><a name="l02325"></a><span class="lineno"> 2325</span>&#160;<span class="comment">// Undefine local defines</span></div><div class="line"><a name="l02326"></a><span class="lineno"> 2326</span>&#160;<span class="preprocessor">#undef COLS_MTX_B</span></div><div class="line"><a name="l02327"></a><span class="lineno"> 2327</span>&#160;</div><div class="line"><a name="l02328"></a><span class="lineno"> 2328</span>&#160;<span class="preprocessor">#if defined(ARM_COMPUTE_OPENCL_FP16_ENABLED)</span></div><div class="line"><a name="l02329"></a><span class="lineno"> 2329</span>&#160;</div><div class="line"><a name="l02375"></a><span class="lineno"> 2375</span>&#160;__kernel <span class="keywordtype">void</span> gemm_mm_interleaved_transposed_f16(<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="l02376"></a><span class="lineno"> 2376</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="l02377"></a><span class="lineno"> 2377</span>&#160;#<span class="keywordflow">if</span> defined(ADD_VEC_C)</div><div class="line"><a name="l02378"></a><span class="lineno"> 2378</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a40a6eb9f2a7712f08d6bb8ff6c9e6ca7">VECTOR_DECLARATION</a>(src2),</div><div class="line"><a name="l02379"></a><span class="lineno"> 2379</span>&#160;#endif <span class="comment">/* defined(ADD_VEC_C) */</span></div><div class="line"><a name="l02380"></a><span class="lineno"> 2380</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="l02381"></a><span class="lineno"> 2381</span>&#160; uint src0_stride_z,</div><div class="line"><a name="l02382"></a><span class="lineno"> 2382</span>&#160; uint src1_stride_z,</div><div class="line"><a name="l02383"></a><span class="lineno"> 2383</span>&#160; uint dst_stride_z</div><div class="line"><a name="l02384"></a><span class="lineno"> 2384</span>&#160;#<span class="keywordflow">if</span> defined(REINTERPRET_OUTPUT_AS_3D)</div><div class="line"><a name="l02385"></a><span class="lineno"> 2385</span>&#160; ,</div><div class="line"><a name="l02386"></a><span class="lineno"> 2386</span>&#160; uint cross_plane_pad</div><div class="line"><a name="l02387"></a><span class="lineno"> 2387</span>&#160;#endif <span class="comment">// REINTERPRET_OUTPUT_AS_3D</span></div><div class="line"><a name="l02388"></a><span class="lineno"> 2388</span>&#160; )</div><div class="line"><a name="l02389"></a><span class="lineno"> 2389</span>&#160;{</div><div class="line"><a name="l02390"></a><span class="lineno"> 2390</span>&#160; <span class="keywordtype">int</span> x = get_global_id(0) / MULT_TRANSPOSE1XW_WIDTH;</div><div class="line"><a name="l02391"></a><span class="lineno"> 2391</span>&#160; <span class="keywordtype">int</span> y = get_global_id(1) / MULT_INTERLEAVE4X4_HEIGHT;</div><div class="line"><a name="l02392"></a><span class="lineno"> 2392</span>&#160; <span class="keywordtype">int</span> z = get_global_id(2);</div><div class="line"><a name="l02393"></a><span class="lineno"> 2393</span>&#160;</div><div class="line"><a name="l02394"></a><span class="lineno"> 2394</span>&#160; <span class="comment">// Offset</span></div><div class="line"><a name="l02395"></a><span class="lineno"> 2395</span>&#160; <span class="keyword">const</span> <span class="keywordtype">int</span> offset_row_a = (get_global_id(1) % MULT_INTERLEAVE4X4_HEIGHT) * 4;</div><div class="line"><a name="l02396"></a><span class="lineno"> 2396</span>&#160; <span class="keyword">const</span> <span class="keywordtype">int</span> offset_row_b = (get_global_id(0) % MULT_TRANSPOSE1XW_WIDTH) * 8;</div><div class="line"><a name="l02397"></a><span class="lineno"> 2397</span>&#160;</div><div class="line"><a name="l02398"></a><span class="lineno"> 2398</span>&#160; <span class="comment">// src_addr_a = address of matrix A</span></div><div class="line"><a name="l02399"></a><span class="lineno"> 2399</span>&#160; <span class="comment">// src_addr_b = address of matrix B</span></div><div class="line"><a name="l02400"></a><span class="lineno"> 2400</span>&#160; <span class="keywordtype">int</span> src0_addr_in_bytes = z * src0_stride_z + y * src0_stride_y + src0_offset_first_element_in_bytes;</div><div class="line"><a name="l02401"></a><span class="lineno"> 2401</span>&#160; <span class="keywordtype">int</span> src1_addr_in_bytes = x * src1_stride_y + src1_offset_first_element_in_bytes;</div><div class="line"><a name="l02402"></a><span class="lineno"> 2402</span>&#160;</div><div class="line"><a name="l02403"></a><span class="lineno"> 2403</span>&#160;<span class="preprocessor">#if defined(MATRIX_B_DEPTH)</span></div><div class="line"><a name="l02404"></a><span class="lineno"> 2404</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="l02405"></a><span class="lineno"> 2405</span>&#160; src1_addr_in_bytes += (z % MATRIX_B_DEPTH) * src1_stride_z;</div><div class="line"><a name="l02406"></a><span class="lineno"> 2406</span>&#160;<span class="preprocessor">#else // defined(MATRIX_B_DEPTH)</span></div><div class="line"><a name="l02407"></a><span class="lineno"> 2407</span>&#160; src1_addr_in_bytes += z * src1_stride_z;</div><div class="line"><a name="l02408"></a><span class="lineno"> 2408</span>&#160;<span class="preprocessor">#endif // defined(MATRIX_B_DEPTH)</span></div><div class="line"><a name="l02409"></a><span class="lineno"> 2409</span>&#160;</div><div class="line"><a name="l02410"></a><span class="lineno"> 2410</span>&#160; __global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *src_addr_a = (__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)(src0_ptr + src0_addr_in_bytes);</div><div class="line"><a name="l02411"></a><span class="lineno"> 2411</span>&#160; __global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *src_addr_b = (__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)(src1_ptr + src1_addr_in_bytes);</div><div class="line"><a name="l02412"></a><span class="lineno"> 2412</span>&#160;</div><div class="line"><a name="l02413"></a><span class="lineno"> 2413</span>&#160; <span class="comment">// Compute end row address for matrix B</span></div><div class="line"><a name="l02414"></a><span class="lineno"> 2414</span>&#160; __global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *src_end_addr_b = src_addr_b + COLS_B;</div><div class="line"><a name="l02415"></a><span class="lineno"> 2415</span>&#160;</div><div class="line"><a name="l02416"></a><span class="lineno"> 2416</span>&#160; src_addr_a += offset_row_a;</div><div class="line"><a name="l02417"></a><span class="lineno"> 2417</span>&#160; src_addr_b += offset_row_b;</div><div class="line"><a name="l02418"></a><span class="lineno"> 2418</span>&#160;</div><div class="line"><a name="l02419"></a><span class="lineno"> 2419</span>&#160; <span class="comment">// Reset accumulators</span></div><div class="line"><a name="l02420"></a><span class="lineno"> 2420</span>&#160; half8 c00 = 0.0f;</div><div class="line"><a name="l02421"></a><span class="lineno"> 2421</span>&#160; half8 c10 = 0.0f;</div><div class="line"><a name="l02422"></a><span class="lineno"> 2422</span>&#160; half8 c20 = 0.0f;</div><div class="line"><a name="l02423"></a><span class="lineno"> 2423</span>&#160; half8 c30 = 0.0f;</div><div class="line"><a name="l02424"></a><span class="lineno"> 2424</span>&#160;</div><div class="line"><a name="l02425"></a><span class="lineno"> 2425</span>&#160; <span class="keywordflow">for</span>(; src_addr_b &lt;= (src_end_addr_b - (int)(16 * MULT_TRANSPOSE1XW_WIDTH)); src_addr_a += 8 * MULT_INTERLEAVE4X4_HEIGHT, src_addr_b += 16 * MULT_TRANSPOSE1XW_WIDTH)</div><div class="line"><a name="l02426"></a><span class="lineno"> 2426</span>&#160; {</div><div class="line"><a name="l02427"></a><span class="lineno"> 2427</span>&#160; <span class="comment">// Load values from matrix A (interleaved) and matrix B (transposed)</span></div><div class="line"><a name="l02428"></a><span class="lineno"> 2428</span>&#160; half4 a0 = vload4(0, src_addr_a);</div><div class="line"><a name="l02429"></a><span class="lineno"> 2429</span>&#160; half8 b0 = vload8(0, src_addr_b);</div><div class="line"><a name="l02430"></a><span class="lineno"> 2430</span>&#160;</div><div class="line"><a name="l02431"></a><span class="lineno"> 2431</span>&#160; c00 += (half8)a0.s0 * b0;</div><div class="line"><a name="l02432"></a><span class="lineno"> 2432</span>&#160; c10 += (half8)a0.s1 * b0;</div><div class="line"><a name="l02433"></a><span class="lineno"> 2433</span>&#160; c20 += (half8)a0.s2 * b0;</div><div class="line"><a name="l02434"></a><span class="lineno"> 2434</span>&#160; c30 += (half8)a0.s3 * b0;</div><div class="line"><a name="l02435"></a><span class="lineno"> 2435</span>&#160;</div><div class="line"><a name="l02436"></a><span class="lineno"> 2436</span>&#160; <span class="comment">// Load values from matrix A (interleaved) and matrix B (transposed)</span></div><div class="line"><a name="l02437"></a><span class="lineno"> 2437</span>&#160; a0 = vload4(0, src_addr_a + 4 * MULT_INTERLEAVE4X4_HEIGHT);</div><div class="line"><a name="l02438"></a><span class="lineno"> 2438</span>&#160; b0 = vload8(0, src_addr_b + 8 * MULT_TRANSPOSE1XW_WIDTH);</div><div class="line"><a name="l02439"></a><span class="lineno"> 2439</span>&#160;</div><div class="line"><a name="l02440"></a><span class="lineno"> 2440</span>&#160; c00 += (half8)a0.s0 * b0;</div><div class="line"><a name="l02441"></a><span class="lineno"> 2441</span>&#160; c10 += (half8)a0.s1 * b0;</div><div class="line"><a name="l02442"></a><span class="lineno"> 2442</span>&#160; c20 += (half8)a0.s2 * b0;</div><div class="line"><a name="l02443"></a><span class="lineno"> 2443</span>&#160; c30 += (half8)a0.s3 * b0;</div><div class="line"><a name="l02444"></a><span class="lineno"> 2444</span>&#160; }</div><div class="line"><a name="l02445"></a><span class="lineno"> 2445</span>&#160;</div><div class="line"><a name="l02446"></a><span class="lineno"> 2446</span>&#160; <span class="keywordflow">for</span>(; src_addr_b &lt; src_end_addr_b; src_addr_a += 4 * MULT_INTERLEAVE4X4_HEIGHT, src_addr_b += 8 * MULT_TRANSPOSE1XW_WIDTH)</div><div class="line"><a name="l02447"></a><span class="lineno"> 2447</span>&#160; {</div><div class="line"><a name="l02448"></a><span class="lineno"> 2448</span>&#160; <span class="comment">// Load values from matrix A (interleaved) and matrix B (transposed)</span></div><div class="line"><a name="l02449"></a><span class="lineno"> 2449</span>&#160; half4 a0 = vload4(0, src_addr_a);</div><div class="line"><a name="l02450"></a><span class="lineno"> 2450</span>&#160; half8 b0 = vload8(0, src_addr_b);</div><div class="line"><a name="l02451"></a><span class="lineno"> 2451</span>&#160;</div><div class="line"><a name="l02452"></a><span class="lineno"> 2452</span>&#160; c00 += (half8)a0.s0 * b0;</div><div class="line"><a name="l02453"></a><span class="lineno"> 2453</span>&#160; c10 += (half8)a0.s1 * b0;</div><div class="line"><a name="l02454"></a><span class="lineno"> 2454</span>&#160; c20 += (half8)a0.s2 * b0;</div><div class="line"><a name="l02455"></a><span class="lineno"> 2455</span>&#160; c30 += (half8)a0.s3 * b0;</div><div class="line"><a name="l02456"></a><span class="lineno"> 2456</span>&#160; }</div><div class="line"><a name="l02457"></a><span class="lineno"> 2457</span>&#160;</div><div class="line"><a name="l02458"></a><span class="lineno"> 2458</span>&#160; <span class="comment">// Compute destination address</span></div><div class="line"><a name="l02459"></a><span class="lineno"> 2459</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="l02460"></a><span class="lineno"> 2460</span>&#160;</div><div class="line"><a name="l02461"></a><span class="lineno"> 2461</span>&#160;<span class="preprocessor">#if defined(ALPHA)</span></div><div class="line"><a name="l02462"></a><span class="lineno"> 2462</span>&#160; <span class="comment">// Multiply by the weight of matrix product</span></div><div class="line"><a name="l02463"></a><span class="lineno"> 2463</span>&#160; c00 = c00 * (half8)ALPHA;</div><div class="line"><a name="l02464"></a><span class="lineno"> 2464</span>&#160; c10 = c10 * (half8)ALPHA;</div><div class="line"><a name="l02465"></a><span class="lineno"> 2465</span>&#160; c20 = c20 * (half8)ALPHA;</div><div class="line"><a name="l02466"></a><span class="lineno"> 2466</span>&#160; c30 = c30 * (half8)ALPHA;</div><div class="line"><a name="l02467"></a><span class="lineno"> 2467</span>&#160;<span class="preprocessor">#endif // defined(ALPHA)</span></div><div class="line"><a name="l02468"></a><span class="lineno"> 2468</span>&#160;</div><div class="line"><a name="l02469"></a><span class="lineno"> 2469</span>&#160;<span class="preprocessor">#if defined(ADD_VEC_C)</span></div><div class="line"><a name="l02470"></a><span class="lineno"> 2470</span>&#160; <span class="comment">// *INDENT-OFF*</span></div><div class="line"><a name="l02471"></a><span class="lineno"> 2471</span>&#160; <span class="comment">// clang-format off</span></div><div class="line"><a name="l02472"></a><span class="lineno"> 2472</span>&#160; __global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *src2_addr = (__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)(src2_ptr + src2_offset_first_element_in_bytes + get_global_id(0) * src2_step_x);</div><div class="line"><a name="l02473"></a><span class="lineno"> 2473</span>&#160; half8 c0 = vload8(0, src2_addr);</div><div class="line"><a name="l02474"></a><span class="lineno"> 2474</span>&#160; <span class="comment">// clang-format on</span></div><div class="line"><a name="l02475"></a><span class="lineno"> 2475</span>&#160; <span class="comment">// *INDENT-ON*</span></div><div class="line"><a name="l02476"></a><span class="lineno"> 2476</span>&#160;</div><div class="line"><a name="l02477"></a><span class="lineno"> 2477</span>&#160; c00 += c0;</div><div class="line"><a name="l02478"></a><span class="lineno"> 2478</span>&#160; c10 += c0;</div><div class="line"><a name="l02479"></a><span class="lineno"> 2479</span>&#160; c20 += c0;</div><div class="line"><a name="l02480"></a><span class="lineno"> 2480</span>&#160; c30 += c0;</div><div class="line"><a name="l02481"></a><span class="lineno"> 2481</span>&#160;<span class="preprocessor">#endif </span><span class="comment">/* defined(ADD_VEC_C) */</span><span class="preprocessor"></span></div><div class="line"><a name="l02482"></a><span class="lineno"> 2482</span>&#160;</div><div class="line"><a name="l02483"></a><span class="lineno"> 2483</span>&#160; <span class="comment">// Compute dst address</span></div><div class="line"><a name="l02484"></a><span class="lineno"> 2484</span>&#160; __global uchar *dst_addr = <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>(&amp;<a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#adbf67dcee294e673cf796f1ed8aeb6a4">dst</a>, 0, 0);</div><div class="line"><a name="l02485"></a><span class="lineno"> 2485</span>&#160;</div><div class="line"><a name="l02486"></a><span class="lineno"> 2486</span>&#160;<span class="preprocessor">#if defined(REINTERPRET_OUTPUT_AS_3D)</span></div><div class="line"><a name="l02487"></a><span class="lineno"> 2487</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="l02488"></a><span class="lineno"> 2488</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="l02489"></a><span class="lineno"> 2489</span>&#160; <span class="comment">//</span></div><div class="line"><a name="l02490"></a><span class="lineno"> 2490</span>&#160; <span class="comment">// | |</span></div><div class="line"><a name="l02491"></a><span class="lineno"> 2491</span>&#160; <span class="comment">// | plane0 |</span></div><div class="line"><a name="l02492"></a><span class="lineno"> 2492</span>&#160; <span class="comment">// | |</span></div><div class="line"><a name="l02493"></a><span class="lineno"> 2493</span>&#160; <span class="comment">// |__________________|</span></div><div class="line"><a name="l02494"></a><span class="lineno"> 2494</span>&#160; <span class="comment">// |******************|</span></div><div class="line"><a name="l02495"></a><span class="lineno"> 2495</span>&#160; <span class="comment">// | cross_plane_pad |</span></div><div class="line"><a name="l02496"></a><span class="lineno"> 2496</span>&#160; <span class="comment">// |******************|</span></div><div class="line"><a name="l02497"></a><span class="lineno"> 2497</span>&#160; <span class="comment">// | |</span></div><div class="line"><a name="l02498"></a><span class="lineno"> 2498</span>&#160; <span class="comment">// | plane1 |</span></div><div class="line"><a name="l02499"></a><span class="lineno"> 2499</span>&#160; <span class="comment">// | |</span></div><div class="line"><a name="l02500"></a><span class="lineno"> 2500</span>&#160; <span class="comment">// |__________________|</span></div><div class="line"><a name="l02501"></a><span class="lineno"> 2501</span>&#160;</div><div class="line"><a name="l02502"></a><span class="lineno"> 2502</span>&#160; <span class="comment">// The plane (zout) is calculated dividing M (get_global_id(1) * 4) by HEIGHT_GEMM3D</span></div><div class="line"><a name="l02503"></a><span class="lineno"> 2503</span>&#160; uint4 zout = ((uint4)(0, 1, 2, 3) + (uint4)(get_global_id(1) * 4)) / (uint4)HEIGHT_GEMM3D;</div><div class="line"><a name="l02504"></a><span class="lineno"> 2504</span>&#160; zout = min(DEPTH_GEMM3D - 1, zout);</div><div class="line"><a name="l02505"></a><span class="lineno"> 2505</span>&#160;</div><div class="line"><a name="l02506"></a><span class="lineno"> 2506</span>&#160; <span class="comment">// Add offset due to the cross plane paddings</span></div><div class="line"><a name="l02507"></a><span class="lineno"> 2507</span>&#160; zout *= (cross_plane_pad * dst_stride_y);</div><div class="line"><a name="l02508"></a><span class="lineno"> 2508</span>&#160;</div><div class="line"><a name="l02509"></a><span class="lineno"> 2509</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="l02510"></a><span class="lineno"> 2510</span>&#160; <span class="comment">// multiply dst_stride_z by DEPTH_GEMM3D</span></div><div class="line"><a name="l02511"></a><span class="lineno"> 2511</span>&#160; dst_addr += z * dst_stride_z * DEPTH_GEMM3D;</div><div class="line"><a name="l02512"></a><span class="lineno"> 2512</span>&#160;</div><div class="line"><a name="l02513"></a><span class="lineno"> 2513</span>&#160; <span class="comment">// Store 4x8 block</span></div><div class="line"><a name="l02514"></a><span class="lineno"> 2514</span>&#160; vstore8(c00, 0, (__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)(dst_addr + 0 * dst_stride_y + zout.s0));</div><div class="line"><a name="l02515"></a><span class="lineno"> 2515</span>&#160; vstore8(c10, 0, (__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)(dst_addr + 1 * dst_stride_y + zout.s1));</div><div class="line"><a name="l02516"></a><span class="lineno"> 2516</span>&#160; vstore8(c20, 0, (__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)(dst_addr + 2 * dst_stride_y + zout.s2));</div><div class="line"><a name="l02517"></a><span class="lineno"> 2517</span>&#160; vstore8(c30, 0, (__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)(dst_addr + 3 * dst_stride_y + zout.s3));</div><div class="line"><a name="l02518"></a><span class="lineno"> 2518</span>&#160;</div><div class="line"><a name="l02519"></a><span class="lineno"> 2519</span>&#160;<span class="preprocessor">#else // defined(REINTERPRET_OUTPUT_AS_3D)</span></div><div class="line"><a name="l02520"></a><span class="lineno"> 2520</span>&#160; <span class="comment">// Add offset for batched GEMM</span></div><div class="line"><a name="l02521"></a><span class="lineno"> 2521</span>&#160; dst_addr += z * dst_stride_z;</div><div class="line"><a name="l02522"></a><span class="lineno"> 2522</span>&#160;</div><div class="line"><a name="l02523"></a><span class="lineno"> 2523</span>&#160; <span class="comment">// Store 4x8 block</span></div><div class="line"><a name="l02524"></a><span class="lineno"> 2524</span>&#160; vstore8(c00, 0, (__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)(dst_addr + 0 * dst_stride_y));</div><div class="line"><a name="l02525"></a><span class="lineno"> 2525</span>&#160; vstore8(c10, 0, (__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)(dst_addr + 1 * dst_stride_y));</div><div class="line"><a name="l02526"></a><span class="lineno"> 2526</span>&#160; vstore8(c20, 0, (__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)(dst_addr + 2 * dst_stride_y));</div><div class="line"><a name="l02527"></a><span class="lineno"> 2527</span>&#160; vstore8(c30, 0, (__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)(dst_addr + 3 * dst_stride_y));</div><div class="line"><a name="l02528"></a><span class="lineno"> 2528</span>&#160;<span class="preprocessor">#endif // defined(REINTERPRET_OUTPUT_AS_3D)</span></div><div class="line"><a name="l02529"></a><span class="lineno"> 2529</span>&#160;}</div><div class="line"><a name="l02530"></a><span class="lineno"> 2530</span>&#160;</div><div class="line"><a name="l02577"></a><span class="lineno"> 2577</span>&#160;__kernel <span class="keywordtype">void</span> gemm_mm_interleaved_transposed_f16_acc32(<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="l02578"></a><span class="lineno"> 2578</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="l02579"></a><span class="lineno"> 2579</span>&#160;#<span class="keywordflow">if</span> defined(ADD_VEC_C)</div><div class="line"><a name="l02580"></a><span class="lineno"> 2580</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a40a6eb9f2a7712f08d6bb8ff6c9e6ca7">VECTOR_DECLARATION</a>(src2),</div><div class="line"><a name="l02581"></a><span class="lineno"> 2581</span>&#160;#endif <span class="comment">/* defined(ADD_VEC_C) */</span></div><div class="line"><a name="l02582"></a><span class="lineno"> 2582</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="l02583"></a><span class="lineno"> 2583</span>&#160; uint src0_stride_z,</div><div class="line"><a name="l02584"></a><span class="lineno"> 2584</span>&#160; uint src1_stride_z,</div><div class="line"><a name="l02585"></a><span class="lineno"> 2585</span>&#160; uint dst_stride_z</div><div class="line"><a name="l02586"></a><span class="lineno"> 2586</span>&#160;#<span class="keywordflow">if</span> defined(REINTERPRET_OUTPUT_AS_3D)</div><div class="line"><a name="l02587"></a><span class="lineno"> 2587</span>&#160; ,</div><div class="line"><a name="l02588"></a><span class="lineno"> 2588</span>&#160; uint cross_plane_pad</div><div class="line"><a name="l02589"></a><span class="lineno"> 2589</span>&#160;#endif <span class="comment">// REINTERPRET_OUTPUT_AS_3D</span></div><div class="line"><a name="l02590"></a><span class="lineno"> 2590</span>&#160; )</div><div class="line"><a name="l02591"></a><span class="lineno"> 2591</span>&#160;{</div><div class="line"><a name="l02592"></a><span class="lineno"> 2592</span>&#160; <span class="keywordtype">int</span> x = get_global_id(0) / MULT_TRANSPOSE1XW_WIDTH;</div><div class="line"><a name="l02593"></a><span class="lineno"> 2593</span>&#160; <span class="keywordtype">int</span> y = get_global_id(1) / MULT_INTERLEAVE4X4_HEIGHT;</div><div class="line"><a name="l02594"></a><span class="lineno"> 2594</span>&#160; <span class="keywordtype">int</span> z = get_global_id(2);</div><div class="line"><a name="l02595"></a><span class="lineno"> 2595</span>&#160;</div><div class="line"><a name="l02596"></a><span class="lineno"> 2596</span>&#160; <span class="comment">// Offset</span></div><div class="line"><a name="l02597"></a><span class="lineno"> 2597</span>&#160; <span class="keyword">const</span> <span class="keywordtype">int</span> offset_row_a = (get_global_id(1) % MULT_INTERLEAVE4X4_HEIGHT) * 4;</div><div class="line"><a name="l02598"></a><span class="lineno"> 2598</span>&#160; <span class="keyword">const</span> <span class="keywordtype">int</span> offset_row_b = (get_global_id(0) % MULT_TRANSPOSE1XW_WIDTH) * 8;</div><div class="line"><a name="l02599"></a><span class="lineno"> 2599</span>&#160;</div><div class="line"><a name="l02600"></a><span class="lineno"> 2600</span>&#160; <span class="comment">// src_addr_a = address of matrix A</span></div><div class="line"><a name="l02601"></a><span class="lineno"> 2601</span>&#160; <span class="comment">// src_addr_b = address of matrix B</span></div><div class="line"><a name="l02602"></a><span class="lineno"> 2602</span>&#160; <span class="keywordtype">int</span> src0_addr_in_bytes = z * src0_stride_z + y * src0_stride_y + src0_offset_first_element_in_bytes;</div><div class="line"><a name="l02603"></a><span class="lineno"> 2603</span>&#160; <span class="keywordtype">int</span> src1_addr_in_bytes = x * src1_stride_y + src1_offset_first_element_in_bytes;</div><div class="line"><a name="l02604"></a><span class="lineno"> 2604</span>&#160;</div><div class="line"><a name="l02605"></a><span class="lineno"> 2605</span>&#160;<span class="preprocessor">#if defined(MATRIX_B_DEPTH)</span></div><div class="line"><a name="l02606"></a><span class="lineno"> 2606</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="l02607"></a><span class="lineno"> 2607</span>&#160; src1_addr_in_bytes += (z % MATRIX_B_DEPTH) * src1_stride_z;</div><div class="line"><a name="l02608"></a><span class="lineno"> 2608</span>&#160;<span class="preprocessor">#else // defined(MATRIX_B_DEPTH)</span></div><div class="line"><a name="l02609"></a><span class="lineno"> 2609</span>&#160; src1_addr_in_bytes += z * src1_stride_z;</div><div class="line"><a name="l02610"></a><span class="lineno"> 2610</span>&#160;<span class="preprocessor">#endif // defined(MATRIX_B_DEPTH)</span></div><div class="line"><a name="l02611"></a><span class="lineno"> 2611</span>&#160;</div><div class="line"><a name="l02612"></a><span class="lineno"> 2612</span>&#160; __global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *src_addr_a = (__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)(src0_ptr + src0_addr_in_bytes);</div><div class="line"><a name="l02613"></a><span class="lineno"> 2613</span>&#160; __global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *src_addr_b = (__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)(src1_ptr + src1_addr_in_bytes);</div><div class="line"><a name="l02614"></a><span class="lineno"> 2614</span>&#160;</div><div class="line"><a name="l02615"></a><span class="lineno"> 2615</span>&#160; <span class="comment">// Compute end row address for matrix B</span></div><div class="line"><a name="l02616"></a><span class="lineno"> 2616</span>&#160; __global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *src_end_addr_b = src_addr_b + COLS_B;</div><div class="line"><a name="l02617"></a><span class="lineno"> 2617</span>&#160;</div><div class="line"><a name="l02618"></a><span class="lineno"> 2618</span>&#160; src_addr_a += offset_row_a;</div><div class="line"><a name="l02619"></a><span class="lineno"> 2619</span>&#160; src_addr_b += offset_row_b;</div><div class="line"><a name="l02620"></a><span class="lineno"> 2620</span>&#160;</div><div class="line"><a name="l02621"></a><span class="lineno"> 2621</span>&#160; <span class="comment">// Reset accumulators</span></div><div class="line"><a name="l02622"></a><span class="lineno"> 2622</span>&#160; float8 c00 = 0.0f;</div><div class="line"><a name="l02623"></a><span class="lineno"> 2623</span>&#160; float8 c10 = 0.0f;</div><div class="line"><a name="l02624"></a><span class="lineno"> 2624</span>&#160; float8 c20 = 0.0f;</div><div class="line"><a name="l02625"></a><span class="lineno"> 2625</span>&#160; float8 c30 = 0.0f;</div><div class="line"><a name="l02626"></a><span class="lineno"> 2626</span>&#160;</div><div class="line"><a name="l02627"></a><span class="lineno"> 2627</span>&#160; <span class="keywordflow">for</span>(; src_addr_b &lt;= (src_end_addr_b - (int)(16 * MULT_TRANSPOSE1XW_WIDTH)); src_addr_a += 8 * MULT_INTERLEAVE4X4_HEIGHT, src_addr_b += 16 * MULT_TRANSPOSE1XW_WIDTH)</div><div class="line"><a name="l02628"></a><span class="lineno"> 2628</span>&#160; {</div><div class="line"><a name="l02629"></a><span class="lineno"> 2629</span>&#160; <span class="comment">// Load values from matrix A (interleaved) and matrix B (transposed)</span></div><div class="line"><a name="l02630"></a><span class="lineno"> 2630</span>&#160; float4 a0 = convert_float4(vload4(0, src_addr_a));</div><div class="line"><a name="l02631"></a><span class="lineno"> 2631</span>&#160; float8 b0 = convert_float8(vload8(0, src_addr_b));</div><div class="line"><a name="l02632"></a><span class="lineno"> 2632</span>&#160;</div><div class="line"><a name="l02633"></a><span class="lineno"> 2633</span>&#160; c00 += (float8)a0.s0 * b0;</div><div class="line"><a name="l02634"></a><span class="lineno"> 2634</span>&#160; c10 += (float8)a0.s1 * b0;</div><div class="line"><a name="l02635"></a><span class="lineno"> 2635</span>&#160; c20 += (float8)a0.s2 * b0;</div><div class="line"><a name="l02636"></a><span class="lineno"> 2636</span>&#160; c30 += (float8)a0.s3 * b0;</div><div class="line"><a name="l02637"></a><span class="lineno"> 2637</span>&#160;</div><div class="line"><a name="l02638"></a><span class="lineno"> 2638</span>&#160; <span class="comment">// Load values from matrix A (interleaved) and matrix B (transposed)</span></div><div class="line"><a name="l02639"></a><span class="lineno"> 2639</span>&#160; a0 = convert_float4(vload4(0, src_addr_a + 4 * MULT_INTERLEAVE4X4_HEIGHT));</div><div class="line"><a name="l02640"></a><span class="lineno"> 2640</span>&#160; b0 = convert_float8(vload8(0, src_addr_b + 8 * MULT_TRANSPOSE1XW_WIDTH));</div><div class="line"><a name="l02641"></a><span class="lineno"> 2641</span>&#160;</div><div class="line"><a name="l02642"></a><span class="lineno"> 2642</span>&#160; c00 += (float8)a0.s0 * b0;</div><div class="line"><a name="l02643"></a><span class="lineno"> 2643</span>&#160; c10 += (float8)a0.s1 * b0;</div><div class="line"><a name="l02644"></a><span class="lineno"> 2644</span>&#160; c20 += (float8)a0.s2 * b0;</div><div class="line"><a name="l02645"></a><span class="lineno"> 2645</span>&#160; c30 += (float8)a0.s3 * b0;</div><div class="line"><a name="l02646"></a><span class="lineno"> 2646</span>&#160; }</div><div class="line"><a name="l02647"></a><span class="lineno"> 2647</span>&#160;</div><div class="line"><a name="l02648"></a><span class="lineno"> 2648</span>&#160; <span class="keywordflow">for</span>(; src_addr_b &lt; src_end_addr_b; src_addr_a += 4 * MULT_INTERLEAVE4X4_HEIGHT, src_addr_b += 8 * MULT_TRANSPOSE1XW_WIDTH)</div><div class="line"><a name="l02649"></a><span class="lineno"> 2649</span>&#160; {</div><div class="line"><a name="l02650"></a><span class="lineno"> 2650</span>&#160; <span class="comment">// Load values from matrix A (interleaved) and matrix B (transposed)</span></div><div class="line"><a name="l02651"></a><span class="lineno"> 2651</span>&#160; float4 a0 = convert_float4(vload4(0, src_addr_a));</div><div class="line"><a name="l02652"></a><span class="lineno"> 2652</span>&#160; float8 b0 = convert_float8(vload8(0, src_addr_b));</div><div class="line"><a name="l02653"></a><span class="lineno"> 2653</span>&#160;</div><div class="line"><a name="l02654"></a><span class="lineno"> 2654</span>&#160; c00 += (float8)a0.s0 * b0;</div><div class="line"><a name="l02655"></a><span class="lineno"> 2655</span>&#160; c10 += (float8)a0.s1 * b0;</div><div class="line"><a name="l02656"></a><span class="lineno"> 2656</span>&#160; c20 += (float8)a0.s2 * b0;</div><div class="line"><a name="l02657"></a><span class="lineno"> 2657</span>&#160; c30 += (float8)a0.s3 * b0;</div><div class="line"><a name="l02658"></a><span class="lineno"> 2658</span>&#160; }</div><div class="line"><a name="l02659"></a><span class="lineno"> 2659</span>&#160;</div><div class="line"><a name="l02660"></a><span class="lineno"> 2660</span>&#160; <span class="comment">// Compute destination address</span></div><div class="line"><a name="l02661"></a><span class="lineno"> 2661</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="l02662"></a><span class="lineno"> 2662</span>&#160;</div><div class="line"><a name="l02663"></a><span class="lineno"> 2663</span>&#160;<span class="preprocessor">#if defined(ALPHA)</span></div><div class="line"><a name="l02664"></a><span class="lineno"> 2664</span>&#160; <span class="comment">// Multiply by the weight of matrix product</span></div><div class="line"><a name="l02665"></a><span class="lineno"> 2665</span>&#160; c00 = c00 * (float8)ALPHA;</div><div class="line"><a name="l02666"></a><span class="lineno"> 2666</span>&#160; c10 = c10 * (float8)ALPHA;</div><div class="line"><a name="l02667"></a><span class="lineno"> 2667</span>&#160; c20 = c20 * (float8)ALPHA;</div><div class="line"><a name="l02668"></a><span class="lineno"> 2668</span>&#160; c30 = c30 * (float8)ALPHA;</div><div class="line"><a name="l02669"></a><span class="lineno"> 2669</span>&#160;<span class="preprocessor">#endif // defined(ALPHA)</span></div><div class="line"><a name="l02670"></a><span class="lineno"> 2670</span>&#160;</div><div class="line"><a name="l02671"></a><span class="lineno"> 2671</span>&#160;<span class="preprocessor">#if defined(ADD_VEC_C)</span></div><div class="line"><a name="l02672"></a><span class="lineno"> 2672</span>&#160; <span class="comment">// *INDENT-OFF*</span></div><div class="line"><a name="l02673"></a><span class="lineno"> 2673</span>&#160; <span class="comment">// clang-format off</span></div><div class="line"><a name="l02674"></a><span class="lineno"> 2674</span>&#160; __global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *src2_addr = (__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)(src2_ptr + src2_offset_first_element_in_bytes + get_global_id(0) * src2_step_x);</div><div class="line"><a name="l02675"></a><span class="lineno"> 2675</span>&#160; float8 c0 = convert_float8(vload8(0, src2_addr));</div><div class="line"><a name="l02676"></a><span class="lineno"> 2676</span>&#160; <span class="comment">// clang-format on</span></div><div class="line"><a name="l02677"></a><span class="lineno"> 2677</span>&#160; <span class="comment">// *INDENT-ON*</span></div><div class="line"><a name="l02678"></a><span class="lineno"> 2678</span>&#160;</div><div class="line"><a name="l02679"></a><span class="lineno"> 2679</span>&#160; c00 += c0;</div><div class="line"><a name="l02680"></a><span class="lineno"> 2680</span>&#160; c10 += c0;</div><div class="line"><a name="l02681"></a><span class="lineno"> 2681</span>&#160; c20 += c0;</div><div class="line"><a name="l02682"></a><span class="lineno"> 2682</span>&#160; c30 += c0;</div><div class="line"><a name="l02683"></a><span class="lineno"> 2683</span>&#160;<span class="preprocessor">#endif </span><span class="comment">/* defined(ADD_VEC_C) */</span><span class="preprocessor"></span></div><div class="line"><a name="l02684"></a><span class="lineno"> 2684</span>&#160;</div><div class="line"><a name="l02685"></a><span class="lineno"> 2685</span>&#160; <span class="comment">// Compute dst address</span></div><div class="line"><a name="l02686"></a><span class="lineno"> 2686</span>&#160; __global uchar *dst_addr = <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>(&amp;<a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#adbf67dcee294e673cf796f1ed8aeb6a4">dst</a>, 0, 0);</div><div class="line"><a name="l02687"></a><span class="lineno"> 2687</span>&#160;</div><div class="line"><a name="l02688"></a><span class="lineno"> 2688</span>&#160;<span class="preprocessor">#if defined(REINTERPRET_OUTPUT_AS_3D)</span></div><div class="line"><a name="l02689"></a><span class="lineno"> 2689</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="l02690"></a><span class="lineno"> 2690</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="l02691"></a><span class="lineno"> 2691</span>&#160; <span class="comment">//</span></div><div class="line"><a name="l02692"></a><span class="lineno"> 2692</span>&#160; <span class="comment">// | |</span></div><div class="line"><a name="l02693"></a><span class="lineno"> 2693</span>&#160; <span class="comment">// | plane0 |</span></div><div class="line"><a name="l02694"></a><span class="lineno"> 2694</span>&#160; <span class="comment">// | |</span></div><div class="line"><a name="l02695"></a><span class="lineno"> 2695</span>&#160; <span class="comment">// |__________________|</span></div><div class="line"><a name="l02696"></a><span class="lineno"> 2696</span>&#160; <span class="comment">// |******************|</span></div><div class="line"><a name="l02697"></a><span class="lineno"> 2697</span>&#160; <span class="comment">// | cross_plane_pad |</span></div><div class="line"><a name="l02698"></a><span class="lineno"> 2698</span>&#160; <span class="comment">// |******************|</span></div><div class="line"><a name="l02699"></a><span class="lineno"> 2699</span>&#160; <span class="comment">// | |</span></div><div class="line"><a name="l02700"></a><span class="lineno"> 2700</span>&#160; <span class="comment">// | plane1 |</span></div><div class="line"><a name="l02701"></a><span class="lineno"> 2701</span>&#160; <span class="comment">// | |</span></div><div class="line"><a name="l02702"></a><span class="lineno"> 2702</span>&#160; <span class="comment">// |__________________|</span></div><div class="line"><a name="l02703"></a><span class="lineno"> 2703</span>&#160;</div><div class="line"><a name="l02704"></a><span class="lineno"> 2704</span>&#160; <span class="comment">// The plane (zout) is calculated dividing M (get_global_id(1) * 4) by HEIGHT_GEMM3D</span></div><div class="line"><a name="l02705"></a><span class="lineno"> 2705</span>&#160; uint4 zout = ((uint4)(0, 1, 2, 3) + (uint4)(get_global_id(1) * 4)) / (uint4)HEIGHT_GEMM3D;</div><div class="line"><a name="l02706"></a><span class="lineno"> 2706</span>&#160; zout = min(DEPTH_GEMM3D - 1, zout);</div><div class="line"><a name="l02707"></a><span class="lineno"> 2707</span>&#160;</div><div class="line"><a name="l02708"></a><span class="lineno"> 2708</span>&#160; <span class="comment">// Add offset due to the cross plane paddings</span></div><div class="line"><a name="l02709"></a><span class="lineno"> 2709</span>&#160; zout *= (cross_plane_pad * dst_stride_y);</div><div class="line"><a name="l02710"></a><span class="lineno"> 2710</span>&#160;</div><div class="line"><a name="l02711"></a><span class="lineno"> 2711</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="l02712"></a><span class="lineno"> 2712</span>&#160; <span class="comment">// multiply dst_stride_z by DEPTH_GEMM3D</span></div><div class="line"><a name="l02713"></a><span class="lineno"> 2713</span>&#160; dst_addr += z * dst_stride_z * DEPTH_GEMM3D;</div><div class="line"><a name="l02714"></a><span class="lineno"> 2714</span>&#160;</div><div class="line"><a name="l02715"></a><span class="lineno"> 2715</span>&#160; <span class="comment">// Store 4x8 block</span></div><div class="line"><a name="l02716"></a><span class="lineno"> 2716</span>&#160; vstore8(convert_half8(c00), 0, (__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)(dst_addr + 0 * dst_stride_y + zout.s0));</div><div class="line"><a name="l02717"></a><span class="lineno"> 2717</span>&#160; vstore8(convert_half8(c10), 0, (__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)(dst_addr + 1 * dst_stride_y + zout.s1));</div><div class="line"><a name="l02718"></a><span class="lineno"> 2718</span>&#160; vstore8(convert_half8(c20), 0, (__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)(dst_addr + 2 * dst_stride_y + zout.s2));</div><div class="line"><a name="l02719"></a><span class="lineno"> 2719</span>&#160; vstore8(convert_half8(c30), 0, (__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)(dst_addr + 3 * dst_stride_y + zout.s3));</div><div class="line"><a name="l02720"></a><span class="lineno"> 2720</span>&#160;</div><div class="line"><a name="l02721"></a><span class="lineno"> 2721</span>&#160;<span class="preprocessor">#else // defined(REINTERPRET_OUTPUT_AS_3D)</span></div><div class="line"><a name="l02722"></a><span class="lineno"> 2722</span>&#160; <span class="comment">// Add offset for batched GEMM</span></div><div class="line"><a name="l02723"></a><span class="lineno"> 2723</span>&#160; dst_addr += z * dst_stride_z;</div><div class="line"><a name="l02724"></a><span class="lineno"> 2724</span>&#160;</div><div class="line"><a name="l02725"></a><span class="lineno"> 2725</span>&#160; <span class="comment">// Store 4x8 block</span></div><div class="line"><a name="l02726"></a><span class="lineno"> 2726</span>&#160; vstore8(convert_half8(c00), 0, (__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)(dst_addr + 0 * dst_stride_y));</div><div class="line"><a name="l02727"></a><span class="lineno"> 2727</span>&#160; vstore8(convert_half8(c10), 0, (__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)(dst_addr + 1 * dst_stride_y));</div><div class="line"><a name="l02728"></a><span class="lineno"> 2728</span>&#160; vstore8(convert_half8(c20), 0, (__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)(dst_addr + 2 * dst_stride_y));</div><div class="line"><a name="l02729"></a><span class="lineno"> 2729</span>&#160; vstore8(convert_half8(c30), 0, (__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)(dst_addr + 3 * dst_stride_y));</div><div class="line"><a name="l02730"></a><span class="lineno"> 2730</span>&#160;<span class="preprocessor">#endif // defined(REINTERPRET_OUTPUT_AS_3D)</span></div><div class="line"><a name="l02731"></a><span class="lineno"> 2731</span>&#160;}</div><div class="line"><a name="l02732"></a><span class="lineno"> 2732</span>&#160;</div><div class="line"><a name="l02776"></a><span class="lineno"> 2776</span>&#160;__kernel <span class="keywordtype">void</span> gemm_mm_interleaved_transposed_f16_bifrost(<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="l02777"></a><span class="lineno"> 2777</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="l02778"></a><span class="lineno"> 2778</span>&#160;#<span class="keywordflow">if</span> defined(ADD_VEC_C)</div><div class="line"><a name="l02779"></a><span class="lineno"> 2779</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a40a6eb9f2a7712f08d6bb8ff6c9e6ca7">VECTOR_DECLARATION</a>(src2),</div><div class="line"><a name="l02780"></a><span class="lineno"> 2780</span>&#160;#endif <span class="comment">/* defined(ADD_VEC_C) */</span></div><div class="line"><a name="l02781"></a><span class="lineno"> 2781</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="l02782"></a><span class="lineno"> 2782</span>&#160; uint src0_stride_z,</div><div class="line"><a name="l02783"></a><span class="lineno"> 2783</span>&#160; uint src1_stride_z,</div><div class="line"><a name="l02784"></a><span class="lineno"> 2784</span>&#160; uint dst_stride_z</div><div class="line"><a name="l02785"></a><span class="lineno"> 2785</span>&#160;#<span class="keywordflow">if</span> defined(REINTERPRET_OUTPUT_AS_3D)</div><div class="line"><a name="l02786"></a><span class="lineno"> 2786</span>&#160; ,</div><div class="line"><a name="l02787"></a><span class="lineno"> 2787</span>&#160; uint cross_plane_pad</div><div class="line"><a name="l02788"></a><span class="lineno"> 2788</span>&#160;#endif <span class="comment">// REINTERPRET_OUTPUT_AS_3D</span></div><div class="line"><a name="l02789"></a><span class="lineno"> 2789</span>&#160; )</div><div class="line"><a name="l02790"></a><span class="lineno"> 2790</span>&#160;{</div><div class="line"><a name="l02791"></a><span class="lineno"> 2791</span>&#160; <span class="keywordtype">int</span> x = get_global_id(0) / MULT_TRANSPOSE1XW_WIDTH;</div><div class="line"><a name="l02792"></a><span class="lineno"> 2792</span>&#160; <span class="keywordtype">int</span> y = get_global_id(1) / MULT_INTERLEAVE4X4_HEIGHT;</div><div class="line"><a name="l02793"></a><span class="lineno"> 2793</span>&#160; <span class="keywordtype">int</span> z = get_global_id(2);</div><div class="line"><a name="l02794"></a><span class="lineno"> 2794</span>&#160;</div><div class="line"><a name="l02795"></a><span class="lineno"> 2795</span>&#160; <span class="comment">// Offset</span></div><div class="line"><a name="l02796"></a><span class="lineno"> 2796</span>&#160; <span class="keyword">const</span> <span class="keywordtype">int</span> offset_row_a = (get_global_id(1) % MULT_INTERLEAVE4X4_HEIGHT) * 4;</div><div class="line"><a name="l02797"></a><span class="lineno"> 2797</span>&#160; <span class="keyword">const</span> <span class="keywordtype">int</span> offset_row_b = (get_global_id(0) % MULT_TRANSPOSE1XW_WIDTH) * 8;</div><div class="line"><a name="l02798"></a><span class="lineno"> 2798</span>&#160;</div><div class="line"><a name="l02799"></a><span class="lineno"> 2799</span>&#160; <span class="comment">// src_addr_a = address of matrix A</span></div><div class="line"><a name="l02800"></a><span class="lineno"> 2800</span>&#160; <span class="comment">// src_addr_b = address of matrix B</span></div><div class="line"><a name="l02801"></a><span class="lineno"> 2801</span>&#160; <span class="keywordtype">int</span> src0_addr_in_bytes = z * src0_stride_z + y * src0_stride_y + src0_offset_first_element_in_bytes;</div><div class="line"><a name="l02802"></a><span class="lineno"> 2802</span>&#160; <span class="keywordtype">int</span> src1_addr_in_bytes = x * src1_stride_y + src1_offset_first_element_in_bytes;</div><div class="line"><a name="l02803"></a><span class="lineno"> 2803</span>&#160;</div><div class="line"><a name="l02804"></a><span class="lineno"> 2804</span>&#160;<span class="preprocessor">#if defined(MATRIX_B_DEPTH)</span></div><div class="line"><a name="l02805"></a><span class="lineno"> 2805</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="l02806"></a><span class="lineno"> 2806</span>&#160; src1_addr_in_bytes += (z % MATRIX_B_DEPTH) * src1_stride_z;</div><div class="line"><a name="l02807"></a><span class="lineno"> 2807</span>&#160;<span class="preprocessor">#else // defined(MATRIX_B_DEPTH)</span></div><div class="line"><a name="l02808"></a><span class="lineno"> 2808</span>&#160; src1_addr_in_bytes += z * src1_stride_z;</div><div class="line"><a name="l02809"></a><span class="lineno"> 2809</span>&#160;<span class="preprocessor">#endif // defined(MATRIX_B_DEPTH)</span></div><div class="line"><a name="l02810"></a><span class="lineno"> 2810</span>&#160;</div><div class="line"><a name="l02811"></a><span class="lineno"> 2811</span>&#160; __global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *src_addr_a = (__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)(src0_ptr + src0_addr_in_bytes);</div><div class="line"><a name="l02812"></a><span class="lineno"> 2812</span>&#160; __global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *src_addr_b = (__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)(src1_ptr + src1_addr_in_bytes);</div><div class="line"><a name="l02813"></a><span class="lineno"> 2813</span>&#160;</div><div class="line"><a name="l02814"></a><span class="lineno"> 2814</span>&#160; <span class="comment">// Compute end row address for matrix B</span></div><div class="line"><a name="l02815"></a><span class="lineno"> 2815</span>&#160; __global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *src_end_addr_b = src_addr_b + COLS_B;</div><div class="line"><a name="l02816"></a><span class="lineno"> 2816</span>&#160;</div><div class="line"><a name="l02817"></a><span class="lineno"> 2817</span>&#160; src_addr_a += offset_row_a;</div><div class="line"><a name="l02818"></a><span class="lineno"> 2818</span>&#160; src_addr_b += offset_row_b;</div><div class="line"><a name="l02819"></a><span class="lineno"> 2819</span>&#160;</div><div class="line"><a name="l02820"></a><span class="lineno"> 2820</span>&#160; <span class="comment">// Reset accumulators</span></div><div class="line"><a name="l02821"></a><span class="lineno"> 2821</span>&#160; half8 c00 = 0.0f;</div><div class="line"><a name="l02822"></a><span class="lineno"> 2822</span>&#160; half8 c10 = 0.0f;</div><div class="line"><a name="l02823"></a><span class="lineno"> 2823</span>&#160; half8 c20 = 0.0f;</div><div class="line"><a name="l02824"></a><span class="lineno"> 2824</span>&#160; half8 c30 = 0.0f;</div><div class="line"><a name="l02825"></a><span class="lineno"> 2825</span>&#160;</div><div class="line"><a name="l02826"></a><span class="lineno"> 2826</span>&#160;<span class="preprocessor">#define COLS_MTX_B (COLS_B / (8 * MULT_TRANSPOSE1XW_WIDTH))</span></div><div class="line"><a name="l02827"></a><span class="lineno"> 2827</span>&#160;</div><div class="line"><a name="l02828"></a><span class="lineno"> 2828</span>&#160; <span class="keywordtype">int</span> i = 0;</div><div class="line"><a name="l02829"></a><span class="lineno"> 2829</span>&#160; <span class="keywordflow">for</span>(; i &lt;= (int)(COLS_MTX_B - 4); i += 4)</div><div class="line"><a name="l02830"></a><span class="lineno"> 2830</span>&#160; {</div><div class="line"><a name="l02831"></a><span class="lineno"> 2831</span>&#160;<span class="preprocessor">#if MULT_INTERLEAVE4X4_HEIGHT == 1</span></div><div class="line"><a name="l02832"></a><span class="lineno"> 2832</span>&#160; <span class="comment">// Load values from matrix A (interleaved) and matrix B (transposed)</span></div><div class="line"><a name="l02833"></a><span class="lineno"> 2833</span>&#160; half8 a0 = vload8(0, src_addr_a);</div><div class="line"><a name="l02834"></a><span class="lineno"> 2834</span>&#160; half8 b0 = vload8(0, src_addr_b);</div><div class="line"><a name="l02835"></a><span class="lineno"> 2835</span>&#160;</div><div class="line"><a name="l02836"></a><span class="lineno"> 2836</span>&#160; src_addr_a += 8 * MULT_INTERLEAVE4X4_HEIGHT;</div><div class="line"><a name="l02837"></a><span class="lineno"> 2837</span>&#160; src_addr_b += 8 * MULT_TRANSPOSE1XW_WIDTH;</div><div class="line"><a name="l02838"></a><span class="lineno"> 2838</span>&#160;</div><div class="line"><a name="l02839"></a><span class="lineno"> 2839</span>&#160; c00 = fma((half8)a0.s0, b0, c00);</div><div class="line"><a name="l02840"></a><span class="lineno"> 2840</span>&#160; c10 = fma((half8)a0.s1, b0, c10);</div><div class="line"><a name="l02841"></a><span class="lineno"> 2841</span>&#160; c20 = fma((half8)a0.s2, b0, c20);</div><div class="line"><a name="l02842"></a><span class="lineno"> 2842</span>&#160; c30 = fma((half8)a0.s3, b0, c30);</div><div class="line"><a name="l02843"></a><span class="lineno"> 2843</span>&#160;</div><div class="line"><a name="l02844"></a><span class="lineno"> 2844</span>&#160; <span class="comment">// Load values from matrix B (transposed)</span></div><div class="line"><a name="l02845"></a><span class="lineno"> 2845</span>&#160; b0 = vload8(0, src_addr_b);</div><div class="line"><a name="l02846"></a><span class="lineno"> 2846</span>&#160;</div><div class="line"><a name="l02847"></a><span class="lineno"> 2847</span>&#160; src_addr_b += 8 * MULT_TRANSPOSE1XW_WIDTH;</div><div class="line"><a name="l02848"></a><span class="lineno"> 2848</span>&#160;</div><div class="line"><a name="l02849"></a><span class="lineno"> 2849</span>&#160; c00 = fma((half8)a0.s4, b0, c00);</div><div class="line"><a name="l02850"></a><span class="lineno"> 2850</span>&#160; c10 = fma((half8)a0.s5, b0, c10);</div><div class="line"><a name="l02851"></a><span class="lineno"> 2851</span>&#160; c20 = fma((half8)a0.s6, b0, c20);</div><div class="line"><a name="l02852"></a><span class="lineno"> 2852</span>&#160; c30 = fma((half8)a0.s7, b0, c30);</div><div class="line"><a name="l02853"></a><span class="lineno"> 2853</span>&#160;</div><div class="line"><a name="l02854"></a><span class="lineno"> 2854</span>&#160; <span class="comment">// Load values from matrix A (interleaved) and matrix B (transposed)</span></div><div class="line"><a name="l02855"></a><span class="lineno"> 2855</span>&#160; a0 = vload8(0, src_addr_a);</div><div class="line"><a name="l02856"></a><span class="lineno"> 2856</span>&#160; b0 = vload8(0, src_addr_b);</div><div class="line"><a name="l02857"></a><span class="lineno"> 2857</span>&#160;</div><div class="line"><a name="l02858"></a><span class="lineno"> 2858</span>&#160; src_addr_a += 8 * MULT_INTERLEAVE4X4_HEIGHT;</div><div class="line"><a name="l02859"></a><span class="lineno"> 2859</span>&#160; src_addr_b += 8 * MULT_TRANSPOSE1XW_WIDTH;</div><div class="line"><a name="l02860"></a><span class="lineno"> 2860</span>&#160;</div><div class="line"><a name="l02861"></a><span class="lineno"> 2861</span>&#160; c00 = fma((half8)a0.s0, b0, c00);</div><div class="line"><a name="l02862"></a><span class="lineno"> 2862</span>&#160; c10 = fma((half8)a0.s1, b0, c10);</div><div class="line"><a name="l02863"></a><span class="lineno"> 2863</span>&#160; c20 = fma((half8)a0.s2, b0, c20);</div><div class="line"><a name="l02864"></a><span class="lineno"> 2864</span>&#160; c30 = fma((half8)a0.s3, b0, c30);</div><div class="line"><a name="l02865"></a><span class="lineno"> 2865</span>&#160;</div><div class="line"><a name="l02866"></a><span class="lineno"> 2866</span>&#160; <span class="comment">// Load values from matrix B (transposed)</span></div><div class="line"><a name="l02867"></a><span class="lineno"> 2867</span>&#160; b0 = vload8(0, src_addr_b);</div><div class="line"><a name="l02868"></a><span class="lineno"> 2868</span>&#160;</div><div class="line"><a name="l02869"></a><span class="lineno"> 2869</span>&#160; src_addr_b += 8 * MULT_TRANSPOSE1XW_WIDTH;</div><div class="line"><a name="l02870"></a><span class="lineno"> 2870</span>&#160;</div><div class="line"><a name="l02871"></a><span class="lineno"> 2871</span>&#160; c00 = fma((half8)a0.s4, b0, c00);</div><div class="line"><a name="l02872"></a><span class="lineno"> 2872</span>&#160; c10 = fma((half8)a0.s5, b0, c10);</div><div class="line"><a name="l02873"></a><span class="lineno"> 2873</span>&#160; c20 = fma((half8)a0.s6, b0, c20);</div><div class="line"><a name="l02874"></a><span class="lineno"> 2874</span>&#160; c30 = fma((half8)a0.s7, b0, c30);</div><div class="line"><a name="l02875"></a><span class="lineno"> 2875</span>&#160;<span class="preprocessor">#else // MULT_INTERLEAVE4X4_HEIGHT == 1</span></div><div class="line"><a name="l02876"></a><span class="lineno"> 2876</span>&#160; <span class="comment">// Load values from matrix A (interleaved) and matrix B (transposed)</span></div><div class="line"><a name="l02877"></a><span class="lineno"> 2877</span>&#160; half4 a0 = vload4(0, src_addr_a);</div><div class="line"><a name="l02878"></a><span class="lineno"> 2878</span>&#160; half8 b0 = vload8(0, src_addr_b);</div><div class="line"><a name="l02879"></a><span class="lineno"> 2879</span>&#160;</div><div class="line"><a name="l02880"></a><span class="lineno"> 2880</span>&#160; src_addr_a += 4 * MULT_INTERLEAVE4X4_HEIGHT;</div><div class="line"><a name="l02881"></a><span class="lineno"> 2881</span>&#160; src_addr_b += 8 * MULT_TRANSPOSE1XW_WIDTH;</div><div class="line"><a name="l02882"></a><span class="lineno"> 2882</span>&#160;</div><div class="line"><a name="l02883"></a><span class="lineno"> 2883</span>&#160; c00 = fma((half8)a0.s0, b0, c00);</div><div class="line"><a name="l02884"></a><span class="lineno"> 2884</span>&#160; c10 = fma((half8)a0.s1, b0, c10);</div><div class="line"><a name="l02885"></a><span class="lineno"> 2885</span>&#160; c20 = fma((half8)a0.s2, b0, c20);</div><div class="line"><a name="l02886"></a><span class="lineno"> 2886</span>&#160; c30 = fma((half8)a0.s3, b0, c30);</div><div class="line"><a name="l02887"></a><span class="lineno"> 2887</span>&#160;</div><div class="line"><a name="l02888"></a><span class="lineno"> 2888</span>&#160; <span class="comment">// Load values from matrix A (interleaved) and matrix B (transposed)</span></div><div class="line"><a name="l02889"></a><span class="lineno"> 2889</span>&#160; a0 = vload4(0, src_addr_a);</div><div class="line"><a name="l02890"></a><span class="lineno"> 2890</span>&#160; b0 = vload8(0, src_addr_b);</div><div class="line"><a name="l02891"></a><span class="lineno"> 2891</span>&#160;</div><div class="line"><a name="l02892"></a><span class="lineno"> 2892</span>&#160; src_addr_a += 4 * MULT_INTERLEAVE4X4_HEIGHT;</div><div class="line"><a name="l02893"></a><span class="lineno"> 2893</span>&#160; src_addr_b += 8 * MULT_TRANSPOSE1XW_WIDTH;</div><div class="line"><a name="l02894"></a><span class="lineno"> 2894</span>&#160;</div><div class="line"><a name="l02895"></a><span class="lineno"> 2895</span>&#160; c00 = fma((half8)a0.s0, b0, c00);</div><div class="line"><a name="l02896"></a><span class="lineno"> 2896</span>&#160; c10 = fma((half8)a0.s1, b0, c10);</div><div class="line"><a name="l02897"></a><span class="lineno"> 2897</span>&#160; c20 = fma((half8)a0.s2, b0, c20);</div><div class="line"><a name="l02898"></a><span class="lineno"> 2898</span>&#160; c30 = fma((half8)a0.s3, b0, c30);</div><div class="line"><a name="l02899"></a><span class="lineno"> 2899</span>&#160;</div><div class="line"><a name="l02900"></a><span class="lineno"> 2900</span>&#160; <span class="comment">// Load values from matrix A (interleaved) and matrix B (transposed)</span></div><div class="line"><a name="l02901"></a><span class="lineno"> 2901</span>&#160; a0 = vload4(0, src_addr_a);</div><div class="line"><a name="l02902"></a><span class="lineno"> 2902</span>&#160; b0 = vload8(0, src_addr_b);</div><div class="line"><a name="l02903"></a><span class="lineno"> 2903</span>&#160;</div><div class="line"><a name="l02904"></a><span class="lineno"> 2904</span>&#160; src_addr_a += 4 * MULT_INTERLEAVE4X4_HEIGHT;</div><div class="line"><a name="l02905"></a><span class="lineno"> 2905</span>&#160; src_addr_b += 8 * MULT_TRANSPOSE1XW_WIDTH;</div><div class="line"><a name="l02906"></a><span class="lineno"> 2906</span>&#160;</div><div class="line"><a name="l02907"></a><span class="lineno"> 2907</span>&#160; c00 = fma((half8)a0.s0, b0, c00);</div><div class="line"><a name="l02908"></a><span class="lineno"> 2908</span>&#160; c10 = fma((half8)a0.s1, b0, c10);</div><div class="line"><a name="l02909"></a><span class="lineno"> 2909</span>&#160; c20 = fma((half8)a0.s2, b0, c20);</div><div class="line"><a name="l02910"></a><span class="lineno"> 2910</span>&#160; c30 = fma((half8)a0.s3, b0, c30);</div><div class="line"><a name="l02911"></a><span class="lineno"> 2911</span>&#160;</div><div class="line"><a name="l02912"></a><span class="lineno"> 2912</span>&#160; <span class="comment">// Load values from matrix A (interleaved) and matrix B (transposed)</span></div><div class="line"><a name="l02913"></a><span class="lineno"> 2913</span>&#160; a0 = vload4(0, src_addr_a);</div><div class="line"><a name="l02914"></a><span class="lineno"> 2914</span>&#160; b0 = vload8(0, src_addr_b);</div><div class="line"><a name="l02915"></a><span class="lineno"> 2915</span>&#160;</div><div class="line"><a name="l02916"></a><span class="lineno"> 2916</span>&#160; src_addr_a += 4 * MULT_INTERLEAVE4X4_HEIGHT;</div><div class="line"><a name="l02917"></a><span class="lineno"> 2917</span>&#160; src_addr_b += 8 * MULT_TRANSPOSE1XW_WIDTH;</div><div class="line"><a name="l02918"></a><span class="lineno"> 2918</span>&#160;</div><div class="line"><a name="l02919"></a><span class="lineno"> 2919</span>&#160; c00 = fma((half8)a0.s0, b0, c00);</div><div class="line"><a name="l02920"></a><span class="lineno"> 2920</span>&#160; c10 = fma((half8)a0.s1, b0, c10);</div><div class="line"><a name="l02921"></a><span class="lineno"> 2921</span>&#160; c20 = fma((half8)a0.s2, b0, c20);</div><div class="line"><a name="l02922"></a><span class="lineno"> 2922</span>&#160; c30 = fma((half8)a0.s3, b0, c30);</div><div class="line"><a name="l02923"></a><span class="lineno"> 2923</span>&#160;<span class="preprocessor">#endif // MULT_INTERLEAVE4X4_HEIGHT == 1</span></div><div class="line"><a name="l02924"></a><span class="lineno"> 2924</span>&#160; }</div><div class="line"><a name="l02925"></a><span class="lineno"> 2925</span>&#160;</div><div class="line"><a name="l02926"></a><span class="lineno"> 2926</span>&#160; <span class="keywordflow">for</span>(; i &lt; (int)(COLS_MTX_B); ++i)</div><div class="line"><a name="l02927"></a><span class="lineno"> 2927</span>&#160; {</div><div class="line"><a name="l02928"></a><span class="lineno"> 2928</span>&#160; <span class="comment">// Load values from matrix A (interleaved) and matrix B (transposed)</span></div><div class="line"><a name="l02929"></a><span class="lineno"> 2929</span>&#160; half4 a0 = vload4(0, src_addr_a);</div><div class="line"><a name="l02930"></a><span class="lineno"> 2930</span>&#160; half8 b0 = vload8(0, src_addr_b);</div><div class="line"><a name="l02931"></a><span class="lineno"> 2931</span>&#160;</div><div class="line"><a name="l02932"></a><span class="lineno"> 2932</span>&#160; src_addr_a += 4 * MULT_INTERLEAVE4X4_HEIGHT;</div><div class="line"><a name="l02933"></a><span class="lineno"> 2933</span>&#160; src_addr_b += 8 * MULT_TRANSPOSE1XW_WIDTH;</div><div class="line"><a name="l02934"></a><span class="lineno"> 2934</span>&#160;</div><div class="line"><a name="l02935"></a><span class="lineno"> 2935</span>&#160; c00 = fma((half8)a0.s0, b0, c00);</div><div class="line"><a name="l02936"></a><span class="lineno"> 2936</span>&#160; c10 = fma((half8)a0.s1, b0, c10);</div><div class="line"><a name="l02937"></a><span class="lineno"> 2937</span>&#160; c20 = fma((half8)a0.s2, b0, c20);</div><div class="line"><a name="l02938"></a><span class="lineno"> 2938</span>&#160; c30 = fma((half8)a0.s3, b0, c30);</div><div class="line"><a name="l02939"></a><span class="lineno"> 2939</span>&#160; }</div><div class="line"><a name="l02940"></a><span class="lineno"> 2940</span>&#160;</div><div class="line"><a name="l02941"></a><span class="lineno"> 2941</span>&#160; <span class="comment">// Compute destination address</span></div><div class="line"><a name="l02942"></a><span class="lineno"> 2942</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="l02943"></a><span class="lineno"> 2943</span>&#160;</div><div class="line"><a name="l02944"></a><span class="lineno"> 2944</span>&#160;<span class="preprocessor">#if defined(ALPHA)</span></div><div class="line"><a name="l02945"></a><span class="lineno"> 2945</span>&#160; <span class="comment">// Multiply by the weight of matrix product</span></div><div class="line"><a name="l02946"></a><span class="lineno"> 2946</span>&#160; c00 = c00 * (half8)ALPHA;</div><div class="line"><a name="l02947"></a><span class="lineno"> 2947</span>&#160; c10 = c10 * (half8)ALPHA;</div><div class="line"><a name="l02948"></a><span class="lineno"> 2948</span>&#160; c20 = c20 * (half8)ALPHA;</div><div class="line"><a name="l02949"></a><span class="lineno"> 2949</span>&#160; c30 = c30 * (half8)ALPHA;</div><div class="line"><a name="l02950"></a><span class="lineno"> 2950</span>&#160;<span class="preprocessor">#endif // defined(ALPHA)</span></div><div class="line"><a name="l02951"></a><span class="lineno"> 2951</span>&#160;</div><div class="line"><a name="l02952"></a><span class="lineno"> 2952</span>&#160;<span class="preprocessor">#if defined(ADD_VEC_C)</span></div><div class="line"><a name="l02953"></a><span class="lineno"> 2953</span>&#160; <span class="comment">// *INDENT-OFF*</span></div><div class="line"><a name="l02954"></a><span class="lineno"> 2954</span>&#160; <span class="comment">// clang-format off</span></div><div class="line"><a name="l02955"></a><span class="lineno"> 2955</span>&#160; __global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *src2_addr = (__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)(src2_ptr + src2_offset_first_element_in_bytes + get_global_id(0) * src2_step_x);</div><div class="line"><a name="l02956"></a><span class="lineno"> 2956</span>&#160; half8 c0 = vload8(0, src2_addr);</div><div class="line"><a name="l02957"></a><span class="lineno"> 2957</span>&#160; <span class="comment">// clang-format on</span></div><div class="line"><a name="l02958"></a><span class="lineno"> 2958</span>&#160; <span class="comment">// *INDENT-ON*</span></div><div class="line"><a name="l02959"></a><span class="lineno"> 2959</span>&#160;</div><div class="line"><a name="l02960"></a><span class="lineno"> 2960</span>&#160; c00 += c0;</div><div class="line"><a name="l02961"></a><span class="lineno"> 2961</span>&#160; c10 += c0;</div><div class="line"><a name="l02962"></a><span class="lineno"> 2962</span>&#160; c20 += c0;</div><div class="line"><a name="l02963"></a><span class="lineno"> 2963</span>&#160; c30 += c0;</div><div class="line"><a name="l02964"></a><span class="lineno"> 2964</span>&#160;<span class="preprocessor">#endif </span><span class="comment">/* defined(ADD_VEC_C) */</span><span class="preprocessor"></span></div><div class="line"><a name="l02965"></a><span class="lineno"> 2965</span>&#160;</div><div class="line"><a name="l02966"></a><span class="lineno"> 2966</span>&#160; <span class="comment">// Compute dst address</span></div><div class="line"><a name="l02967"></a><span class="lineno"> 2967</span>&#160; __global uchar *dst_addr = <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>(&amp;<a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#adbf67dcee294e673cf796f1ed8aeb6a4">dst</a>, 0, 0);</div><div class="line"><a name="l02968"></a><span class="lineno"> 2968</span>&#160;</div><div class="line"><a name="l02969"></a><span class="lineno"> 2969</span>&#160;<span class="preprocessor">#if defined(REINTERPRET_OUTPUT_AS_3D)</span></div><div class="line"><a name="l02970"></a><span class="lineno"> 2970</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="l02971"></a><span class="lineno"> 2971</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="l02972"></a><span class="lineno"> 2972</span>&#160; <span class="comment">//</span></div><div class="line"><a name="l02973"></a><span class="lineno"> 2973</span>&#160; <span class="comment">// | |</span></div><div class="line"><a name="l02974"></a><span class="lineno"> 2974</span>&#160; <span class="comment">// | plane0 |</span></div><div class="line"><a name="l02975"></a><span class="lineno"> 2975</span>&#160; <span class="comment">// | |</span></div><div class="line"><a name="l02976"></a><span class="lineno"> 2976</span>&#160; <span class="comment">// |__________________|</span></div><div class="line"><a name="l02977"></a><span class="lineno"> 2977</span>&#160; <span class="comment">// |******************|</span></div><div class="line"><a name="l02978"></a><span class="lineno"> 2978</span>&#160; <span class="comment">// | cross_plane_pad |</span></div><div class="line"><a name="l02979"></a><span class="lineno"> 2979</span>&#160; <span class="comment">// |******************|</span></div><div class="line"><a name="l02980"></a><span class="lineno"> 2980</span>&#160; <span class="comment">// | |</span></div><div class="line"><a name="l02981"></a><span class="lineno"> 2981</span>&#160; <span class="comment">// | plane1 |</span></div><div class="line"><a name="l02982"></a><span class="lineno"> 2982</span>&#160; <span class="comment">// | |</span></div><div class="line"><a name="l02983"></a><span class="lineno"> 2983</span>&#160; <span class="comment">// |__________________|</span></div><div class="line"><a name="l02984"></a><span class="lineno"> 2984</span>&#160;</div><div class="line"><a name="l02985"></a><span class="lineno"> 2985</span>&#160; <span class="comment">// The plane (zout) is calculated dividing M (get_global_id(1) * 4) by HEIGHT_GEMM3D</span></div><div class="line"><a name="l02986"></a><span class="lineno"> 2986</span>&#160; uint4 zout = ((uint4)(0, 1, 2, 3) + (uint4)(get_global_id(1) * 4)) / (uint4)HEIGHT_GEMM3D;</div><div class="line"><a name="l02987"></a><span class="lineno"> 2987</span>&#160; zout = min(DEPTH_GEMM3D - 1, zout);</div><div class="line"><a name="l02988"></a><span class="lineno"> 2988</span>&#160;</div><div class="line"><a name="l02989"></a><span class="lineno"> 2989</span>&#160; <span class="comment">// Add offset due to the cross plane paddings</span></div><div class="line"><a name="l02990"></a><span class="lineno"> 2990</span>&#160; zout *= (cross_plane_pad * dst_stride_y);</div><div class="line"><a name="l02991"></a><span class="lineno"> 2991</span>&#160;</div><div class="line"><a name="l02992"></a><span class="lineno"> 2992</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="l02993"></a><span class="lineno"> 2993</span>&#160; <span class="comment">// multiply dst_stride_z by DEPTH_GEMM3D</span></div><div class="line"><a name="l02994"></a><span class="lineno"> 2994</span>&#160; dst_addr += z * dst_stride_z * DEPTH_GEMM3D;</div><div class="line"><a name="l02995"></a><span class="lineno"> 2995</span>&#160;</div><div class="line"><a name="l02996"></a><span class="lineno"> 2996</span>&#160; <span class="comment">// Store 4x8 block</span></div><div class="line"><a name="l02997"></a><span class="lineno"> 2997</span>&#160; vstore8(c00, 0, (__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)(dst_addr + 0 * dst_stride_y + zout.s0));</div><div class="line"><a name="l02998"></a><span class="lineno"> 2998</span>&#160; vstore8(c10, 0, (__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)(dst_addr + 1 * dst_stride_y + zout.s1));</div><div class="line"><a name="l02999"></a><span class="lineno"> 2999</span>&#160; vstore8(c20, 0, (__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)(dst_addr + 2 * dst_stride_y + zout.s2));</div><div class="line"><a name="l03000"></a><span class="lineno"> 3000</span>&#160; vstore8(c30, 0, (__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)(dst_addr + 3 * dst_stride_y + zout.s3));</div><div class="line"><a name="l03001"></a><span class="lineno"> 3001</span>&#160;</div><div class="line"><a name="l03002"></a><span class="lineno"> 3002</span>&#160;<span class="preprocessor">#else // defined(REINTERPRET_OUTPUT_AS_3D)</span></div><div class="line"><a name="l03003"></a><span class="lineno"> 3003</span>&#160; <span class="comment">// Add offset for batched GEMM</span></div><div class="line"><a name="l03004"></a><span class="lineno"> 3004</span>&#160; dst_addr += z * dst_stride_z;</div><div class="line"><a name="l03005"></a><span class="lineno"> 3005</span>&#160;</div><div class="line"><a name="l03006"></a><span class="lineno"> 3006</span>&#160; <span class="comment">// Store 4x8 block</span></div><div class="line"><a name="l03007"></a><span class="lineno"> 3007</span>&#160; vstore8(c00, 0, (__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)(dst_addr + 0 * dst_stride_y));</div><div class="line"><a name="l03008"></a><span class="lineno"> 3008</span>&#160; vstore8(c10, 0, (__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)(dst_addr + 1 * dst_stride_y));</div><div class="line"><a name="l03009"></a><span class="lineno"> 3009</span>&#160; vstore8(c20, 0, (__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)(dst_addr + 2 * dst_stride_y));</div><div class="line"><a name="l03010"></a><span class="lineno"> 3010</span>&#160; vstore8(c30, 0, (__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)(dst_addr + 3 * dst_stride_y));</div><div class="line"><a name="l03011"></a><span class="lineno"> 3011</span>&#160;<span class="preprocessor">#endif // defined(REINTERPRET_OUTPUT_AS_3D)</span></div><div class="line"><a name="l03012"></a><span class="lineno"> 3012</span>&#160;}</div><div class="line"><a name="l03013"></a><span class="lineno"> 3013</span>&#160;</div><div class="line"><a name="l03014"></a><span class="lineno"> 3014</span>&#160;<span class="comment">// Undefine local defines</span></div><div class="line"><a name="l03015"></a><span class="lineno"> 3015</span>&#160;<span class="preprocessor">#undef COLS_MTX_B</span></div><div class="line"><a name="l03016"></a><span class="lineno"> 3016</span>&#160;</div><div class="line"><a name="l03017"></a><span class="lineno"> 3017</span>&#160;<span class="preprocessor">#endif // defined(ARM_COMPUTE_OPENCL_FP16_ENABLED)</span></div><div class="line"><a name="l03018"></a><span class="lineno"> 3018</span>&#160;</div><div class="line"><a name="l03019"></a><span class="lineno"> 3019</span>&#160;<span class="preprocessor">#endif // defined(COLS_B) &amp;&amp; defined(MULT_TRANSPOSE1XW_WIDTH) &amp;&amp; defined(MULT_INTERLEAVE4X4_HEIGHT)</span></div><div class="line"><a name="l03020"></a><span class="lineno"> 3020</span>&#160;</div><div class="line"><a name="l03021"></a><span class="lineno"> 3021</span>&#160;<span class="preprocessor">#if defined(COLS_A) &amp;&amp; defined(NUM_ELEMS_PROCESSED_PER_THREAD_X) &amp;&amp; (NUM_ELEMS_PROCESSED_PER_THREAD_Y)</span></div><div class="line"><a name="l03022"></a><span class="lineno"> 3022</span>&#160;<span class="preprocessor">#if defined(DATA_TYPE)</span></div><div class="line"><a name="l03023"></a><span class="lineno"> 3023</span>&#160;<span class="preprocessor">#define VECTOR_TYPE VEC_DATA_TYPE(DATA_TYPE, NUM_ELEMS_PROCESSED_PER_THREAD_X)</span></div><div class="line"><a name="l03024"></a><span class="lineno"> 3024</span>&#160;</div><div class="line"><a name="l03072"></a><span class="lineno"> 3072</span>&#160;__kernel <span class="keywordtype">void</span> gemm_mm_floating_point(<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="l03073"></a><span class="lineno"> 3073</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="l03074"></a><span class="lineno"> 3074</span>&#160;#<span class="keywordflow">if</span> defined(ADD_VEC_C)</div><div class="line"><a name="l03075"></a><span class="lineno"> 3075</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a40a6eb9f2a7712f08d6bb8ff6c9e6ca7">VECTOR_DECLARATION</a>(src2),</div><div class="line"><a name="l03076"></a><span class="lineno"> 3076</span>&#160;#endif <span class="comment">/* defined(ADD_VEC_C) */</span></div><div class="line"><a name="l03077"></a><span class="lineno"> 3077</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="l03078"></a><span class="lineno"> 3078</span>&#160; uint src0_stride_z,</div><div class="line"><a name="l03079"></a><span class="lineno"> 3079</span>&#160; uint src1_stride_z,</div><div class="line"><a name="l03080"></a><span class="lineno"> 3080</span>&#160; uint dst_stride_z</div><div class="line"><a name="l03081"></a><span class="lineno"> 3081</span>&#160;#<span class="keywordflow">if</span> defined(REINTERPRET_INPUT_AS_3D)</div><div class="line"><a name="l03082"></a><span class="lineno"> 3082</span>&#160; ,</div><div class="line"><a name="l03083"></a><span class="lineno"> 3083</span>&#160; uint src_cross_plane_pad</div><div class="line"><a name="l03084"></a><span class="lineno"> 3084</span>&#160;#endif <span class="comment">// REINTERPRET_INPUT_AS_3D</span></div><div class="line"><a name="l03085"></a><span class="lineno"> 3085</span>&#160;#<span class="keywordflow">if</span> defined(REINTERPRET_OUTPUT_AS_3D)</div><div class="line"><a name="l03086"></a><span class="lineno"> 3086</span>&#160; ,</div><div class="line"><a name="l03087"></a><span class="lineno"> 3087</span>&#160; uint dst_cross_plane_pad</div><div class="line"><a name="l03088"></a><span class="lineno"> 3088</span>&#160;#endif <span class="comment">// REINTERPRET_OUTPUT_AS_3D</span></div><div class="line"><a name="l03089"></a><span class="lineno"> 3089</span>&#160; )</div><div class="line"><a name="l03090"></a><span class="lineno"> 3090</span>&#160;{</div><div class="line"><a name="l03091"></a><span class="lineno"> 3091</span>&#160; <span class="keywordtype">int</span> idx = get_global_id(0) * NUM_ELEMS_PROCESSED_PER_THREAD_X;</div><div class="line"><a name="l03092"></a><span class="lineno"> 3092</span>&#160;</div><div class="line"><a name="l03093"></a><span class="lineno"> 3093</span>&#160; <span class="comment">// Compute starting address for matrix A and Matrix B</span></div><div class="line"><a name="l03094"></a><span class="lineno"> 3094</span>&#160; int2 src_addr = ((int2)(src0_offset_first_element_in_bytes, src1_offset_first_element_in_bytes));</div><div class="line"><a name="l03095"></a><span class="lineno"> 3095</span>&#160;</div><div class="line"><a name="l03096"></a><span class="lineno"> 3096</span>&#160; <span class="comment">// Update address for the matrix A</span></div><div class="line"><a name="l03097"></a><span class="lineno"> 3097</span>&#160; src_addr.s0 += get_global_id(1) * src0_stride_y * NUM_ELEMS_PROCESSED_PER_THREAD_Y;</div><div class="line"><a name="l03098"></a><span class="lineno"> 3098</span>&#160;</div><div class="line"><a name="l03099"></a><span class="lineno"> 3099</span>&#160; <span class="comment">// Update address for the matrix B</span></div><div class="line"><a name="l03100"></a><span class="lineno"> 3100</span>&#160; src_addr.s1 += idx * <span class="keyword">sizeof</span>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>);</div><div class="line"><a name="l03101"></a><span class="lineno"> 3101</span>&#160;</div><div class="line"><a name="l03102"></a><span class="lineno"> 3102</span>&#160;<span class="preprocessor">#if defined(REINTERPRET_INPUT_AS_3D)</span></div><div class="line"><a name="l03103"></a><span class="lineno"> 3103</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="l03104"></a><span class="lineno"> 3104</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="l03105"></a><span class="lineno"> 3105</span>&#160; <span class="comment">//</span></div><div class="line"><a name="l03106"></a><span class="lineno"> 3106</span>&#160; <span class="comment">// | |</span></div><div class="line"><a name="l03107"></a><span class="lineno"> 3107</span>&#160; <span class="comment">// | plane0 |</span></div><div class="line"><a name="l03108"></a><span class="lineno"> 3108</span>&#160; <span class="comment">// | |</span></div><div class="line"><a name="l03109"></a><span class="lineno"> 3109</span>&#160; <span class="comment">// |__________________|</span></div><div class="line"><a name="l03110"></a><span class="lineno"> 3110</span>&#160; <span class="comment">// |******************|</span></div><div class="line"><a name="l03111"></a><span class="lineno"> 3111</span>&#160; <span class="comment">// | cross_plane_pad |</span></div><div class="line"><a name="l03112"></a><span class="lineno"> 3112</span>&#160; <span class="comment">// |******************|</span></div><div class="line"><a name="l03113"></a><span class="lineno"> 3113</span>&#160; <span class="comment">// | |</span></div><div class="line"><a name="l03114"></a><span class="lineno"> 3114</span>&#160; <span class="comment">// | plane1 |</span></div><div class="line"><a name="l03115"></a><span class="lineno"> 3115</span>&#160; <span class="comment">// | |</span></div><div class="line"><a name="l03116"></a><span class="lineno"> 3116</span>&#160; <span class="comment">// |__________________|</span></div><div class="line"><a name="l03117"></a><span class="lineno"> 3117</span>&#160;</div><div class="line"><a name="l03118"></a><span class="lineno"> 3118</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="l03119"></a><span class="lineno"> 3119</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="l03120"></a><span class="lineno"> 3120</span>&#160; zin = min(DEPTH_GEMM3D - 1, zin);</div><div class="line"><a name="l03121"></a><span class="lineno"> 3121</span>&#160;</div><div class="line"><a name="l03122"></a><span class="lineno"> 3122</span>&#160; <span class="comment">// Add offset due to the cross plane paddings</span></div><div class="line"><a name="l03123"></a><span class="lineno"> 3123</span>&#160; zin *= (src_cross_plane_pad * src0_stride_y);</div><div class="line"><a name="l03124"></a><span class="lineno"> 3124</span>&#160;</div><div class="line"><a name="l03125"></a><span class="lineno"> 3125</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="l03126"></a><span class="lineno"> 3126</span>&#160; <span class="comment">// multiply src0_stride_z by DEPTH_GEMM3D</span></div><div class="line"><a name="l03127"></a><span class="lineno"> 3127</span>&#160; src_addr.s0 += get_global_id(2) * src0_stride_z * DEPTH_GEMM3D;</div><div class="line"><a name="l03128"></a><span class="lineno"> 3128</span>&#160;</div><div class="line"><a name="l03129"></a><span class="lineno"> 3129</span>&#160;<span class="preprocessor">#else // defined(REINTERPRET_INPUT_AS_3D)</span></div><div class="line"><a name="l03130"></a><span class="lineno"> 3130</span>&#160;</div><div class="line"><a name="l03131"></a><span class="lineno"> 3131</span>&#160; <span class="comment">// Add offset for batched GEMM</span></div><div class="line"><a name="l03132"></a><span class="lineno"> 3132</span>&#160; src_addr.s0 += get_global_id(2) * src0_stride_z;</div><div class="line"><a name="l03133"></a><span class="lineno"> 3133</span>&#160;</div><div class="line"><a name="l03134"></a><span class="lineno"> 3134</span>&#160;<span class="preprocessor">#endif // defined(REINTERPRET_INPUT_AS_3D)</span></div><div class="line"><a name="l03135"></a><span class="lineno"> 3135</span>&#160;</div><div class="line"><a name="l03136"></a><span class="lineno"> 3136</span>&#160;<span class="preprocessor">#if defined(MATRIX_B_DEPTH)</span></div><div class="line"><a name="l03137"></a><span class="lineno"> 3137</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="l03138"></a><span class="lineno"> 3138</span>&#160; src_addr.s1 += (get_global_id(2) % MATRIX_B_DEPTH) * src1_stride_z;</div><div class="line"><a name="l03139"></a><span class="lineno"> 3139</span>&#160;<span class="preprocessor">#else // defined(MATRIX_B_DEPTH)</span></div><div class="line"><a name="l03140"></a><span class="lineno"> 3140</span>&#160; src_addr.s1 += get_global_id(2) * src1_stride_z;</div><div class="line"><a name="l03141"></a><span class="lineno"> 3141</span>&#160;<span class="preprocessor">#endif // defined(MATRIX_B_DEPTH)</span></div><div class="line"><a name="l03142"></a><span class="lineno"> 3142</span>&#160;</div><div class="line"><a name="l03143"></a><span class="lineno"> 3143</span>&#160; <span class="keywordtype">int</span> end_row_vec_a = src_addr.s0 + (COLS_A * <span class="keyword">sizeof</span>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>));</div><div class="line"><a name="l03144"></a><span class="lineno"> 3144</span>&#160;</div><div class="line"><a name="l03145"></a><span class="lineno"> 3145</span>&#160; VECTOR_TYPE acc0 = 0.0f;</div><div class="line"><a name="l03146"></a><span class="lineno"> 3146</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 1</span></div><div class="line"><a name="l03147"></a><span class="lineno"> 3147</span>&#160; VECTOR_TYPE acc1 = 0.0f;</div><div class="line"><a name="l03148"></a><span class="lineno"> 3148</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 1</span></div><div class="line"><a name="l03149"></a><span class="lineno"> 3149</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 2</span></div><div class="line"><a name="l03150"></a><span class="lineno"> 3150</span>&#160; VECTOR_TYPE acc2 = 0.0f;</div><div class="line"><a name="l03151"></a><span class="lineno"> 3151</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 2</span></div><div class="line"><a name="l03152"></a><span class="lineno"> 3152</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 3</span></div><div class="line"><a name="l03153"></a><span class="lineno"> 3153</span>&#160; VECTOR_TYPE acc3 = 0.0f;</div><div class="line"><a name="l03154"></a><span class="lineno"> 3154</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 3</span></div><div class="line"><a name="l03155"></a><span class="lineno"> 3155</span>&#160;</div><div class="line"><a name="l03156"></a><span class="lineno"> 3156</span>&#160; <span class="keywordflow">for</span>(; src_addr.s0 &lt;= (end_row_vec_a - 2 * (int)<span class="keyword">sizeof</span>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>)); src_addr += (int2)(2 * <span class="keyword">sizeof</span>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>), 2 * src1_stride_y))</div><div class="line"><a name="l03157"></a><span class="lineno"> 3157</span>&#160; {</div><div class="line"><a name="l03158"></a><span class="lineno"> 3158</span>&#160;<span class="preprocessor">#if defined(REINTERPRET_INPUT_AS_3D)</span></div><div class="line"><a name="l03159"></a><span class="lineno"> 3159</span>&#160; <span class="comment">// Load values from matrix A</span></div><div class="line"><a name="l03160"></a><span class="lineno"> 3160</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, 2)</div><div class="line"><a name="l03161"></a><span class="lineno"> 3161</span>&#160; a0 = vload2(0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src0_ptr + src_addr.s0 + 0 * src0_stride_y + zin.s0));</div><div class="line"><a name="l03162"></a><span class="lineno"> 3162</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 1</span></div><div class="line"><a name="l03163"></a><span class="lineno"> 3163</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, 2)</div><div class="line"><a name="l03164"></a><span class="lineno"> 3164</span>&#160; a1 = vload2(0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src0_ptr + src_addr.s0 + 1 * src0_stride_y + zin.s1));</div><div class="line"><a name="l03165"></a><span class="lineno"> 3165</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 1</span></div><div class="line"><a name="l03166"></a><span class="lineno"> 3166</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 2</span></div><div class="line"><a name="l03167"></a><span class="lineno"> 3167</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, 2)</div><div class="line"><a name="l03168"></a><span class="lineno"> 3168</span>&#160; a2 = vload2(0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src0_ptr + src_addr.s0 + 2 * src0_stride_y + zin.s2));</div><div class="line"><a name="l03169"></a><span class="lineno"> 3169</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 2</span></div><div class="line"><a name="l03170"></a><span class="lineno"> 3170</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 3</span></div><div class="line"><a name="l03171"></a><span class="lineno"> 3171</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, 2)</div><div class="line"><a name="l03172"></a><span class="lineno"> 3172</span>&#160; a3 = vload2(0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src0_ptr + src_addr.s0 + 3 * src0_stride_y + zin.s3));</div><div class="line"><a name="l03173"></a><span class="lineno"> 3173</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 3</span></div><div class="line"><a name="l03174"></a><span class="lineno"> 3174</span>&#160;<span class="preprocessor">#else // defined(REINTERPRET_INPUT_AS_3D)</span></div><div class="line"><a name="l03175"></a><span class="lineno"> 3175</span>&#160; <span class="comment">// Load values from matrix A</span></div><div class="line"><a name="l03176"></a><span class="lineno"> 3176</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, 2)</div><div class="line"><a name="l03177"></a><span class="lineno"> 3177</span>&#160; a0 = vload2(0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src0_ptr + src_addr.s0 + 0 * src0_stride_y));</div><div class="line"><a name="l03178"></a><span class="lineno"> 3178</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 1</span></div><div class="line"><a name="l03179"></a><span class="lineno"> 3179</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, 2)</div><div class="line"><a name="l03180"></a><span class="lineno"> 3180</span>&#160; a1 = vload2(0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src0_ptr + src_addr.s0 + 1 * src0_stride_y));</div><div class="line"><a name="l03181"></a><span class="lineno"> 3181</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 1</span></div><div class="line"><a name="l03182"></a><span class="lineno"> 3182</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 2</span></div><div class="line"><a name="l03183"></a><span class="lineno"> 3183</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, 2)</div><div class="line"><a name="l03184"></a><span class="lineno"> 3184</span>&#160; a2 = vload2(0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src0_ptr + src_addr.s0 + 2 * src0_stride_y));</div><div class="line"><a name="l03185"></a><span class="lineno"> 3185</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 2</span></div><div class="line"><a name="l03186"></a><span class="lineno"> 3186</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 3</span></div><div class="line"><a name="l03187"></a><span class="lineno"> 3187</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, 2)</div><div class="line"><a name="l03188"></a><span class="lineno"> 3188</span>&#160; a3 = vload2(0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src0_ptr + src_addr.s0 + 3 * src0_stride_y));</div><div class="line"><a name="l03189"></a><span class="lineno"> 3189</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 3</span></div><div class="line"><a name="l03190"></a><span class="lineno"> 3190</span>&#160;<span class="preprocessor">#endif // defined(REINTERPRET_INPUT_AS_3D)</span></div><div class="line"><a name="l03191"></a><span class="lineno"> 3191</span>&#160;</div><div class="line"><a name="l03192"></a><span class="lineno"> 3192</span>&#160; <span class="comment">// Load values from matrix B</span></div><div class="line"><a name="l03193"></a><span class="lineno"> 3193</span>&#160; VECTOR_TYPE 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, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src1_ptr + src_addr.s1));</div><div class="line"><a name="l03194"></a><span class="lineno"> 3194</span>&#160; VECTOR_TYPE 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, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src1_ptr + src_addr.s1 + src1_stride_y));</div><div class="line"><a name="l03195"></a><span class="lineno"> 3195</span>&#160;</div><div class="line"><a name="l03196"></a><span class="lineno"> 3196</span>&#160; <span class="comment">// Accumulate</span></div><div class="line"><a name="l03197"></a><span class="lineno"> 3197</span>&#160; acc0 += b0 * (VECTOR_TYPE)a0.s0;</div><div class="line"><a name="l03198"></a><span class="lineno"> 3198</span>&#160; acc0 += b1 * (VECTOR_TYPE)a0.s1;</div><div class="line"><a name="l03199"></a><span class="lineno"> 3199</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 1</span></div><div class="line"><a name="l03200"></a><span class="lineno"> 3200</span>&#160; acc1 += b0 * (VECTOR_TYPE)a1.s0;</div><div class="line"><a name="l03201"></a><span class="lineno"> 3201</span>&#160; acc1 += b1 * (VECTOR_TYPE)a1.s1;</div><div class="line"><a name="l03202"></a><span class="lineno"> 3202</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 1</span></div><div class="line"><a name="l03203"></a><span class="lineno"> 3203</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 2</span></div><div class="line"><a name="l03204"></a><span class="lineno"> 3204</span>&#160; acc2 += b0 * (VECTOR_TYPE)a2.s0;</div><div class="line"><a name="l03205"></a><span class="lineno"> 3205</span>&#160; acc2 += b1 * (VECTOR_TYPE)a2.s1;</div><div class="line"><a name="l03206"></a><span class="lineno"> 3206</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 2</span></div><div class="line"><a name="l03207"></a><span class="lineno"> 3207</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 3</span></div><div class="line"><a name="l03208"></a><span class="lineno"> 3208</span>&#160; acc3 += b0 * (VECTOR_TYPE)a3.s0;</div><div class="line"><a name="l03209"></a><span class="lineno"> 3209</span>&#160; acc3 += b1 * (VECTOR_TYPE)a3.s1;</div><div class="line"><a name="l03210"></a><span class="lineno"> 3210</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 3</span></div><div class="line"><a name="l03211"></a><span class="lineno"> 3211</span>&#160; }</div><div class="line"><a name="l03212"></a><span class="lineno"> 3212</span>&#160;</div><div class="line"><a name="l03213"></a><span class="lineno"> 3213</span>&#160; <span class="keywordflow">for</span>(; src_addr.s0 &lt; end_row_vec_a; src_addr += (int2)(<span class="keyword">sizeof</span>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>), src1_stride_y))</div><div class="line"><a name="l03214"></a><span class="lineno"> 3214</span>&#160; {</div><div class="line"><a name="l03215"></a><span class="lineno"> 3215</span>&#160;<span class="preprocessor">#if defined(REINTERPRET_INPUT_AS_3D)</span></div><div class="line"><a name="l03216"></a><span class="lineno"> 3216</span>&#160; <span class="comment">// Load values from matrix A</span></div><div class="line"><a name="l03217"></a><span class="lineno"> 3217</span>&#160; <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> a0 = *((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src0_ptr + src_addr.s0 + 0 * src0_stride_y + zin.s0));</div><div class="line"><a name="l03218"></a><span class="lineno"> 3218</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 1</span></div><div class="line"><a name="l03219"></a><span class="lineno"> 3219</span>&#160; <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> a1 = *((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src0_ptr + src_addr.s0 + 1 * src0_stride_y + zin.s1));</div><div class="line"><a name="l03220"></a><span class="lineno"> 3220</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 1</span></div><div class="line"><a name="l03221"></a><span class="lineno"> 3221</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 2</span></div><div class="line"><a name="l03222"></a><span class="lineno"> 3222</span>&#160; <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> a2 = *((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src0_ptr + src_addr.s0 + 2 * src0_stride_y + zin.s2));</div><div class="line"><a name="l03223"></a><span class="lineno"> 3223</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 2</span></div><div class="line"><a name="l03224"></a><span class="lineno"> 3224</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 3</span></div><div class="line"><a name="l03225"></a><span class="lineno"> 3225</span>&#160; <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> a3 = *((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src0_ptr + src_addr.s0 + 3 * src0_stride_y + zin.s3));</div><div class="line"><a name="l03226"></a><span class="lineno"> 3226</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 3</span></div><div class="line"><a name="l03227"></a><span class="lineno"> 3227</span>&#160;<span class="preprocessor">#else // defined(REINTERPRET_INPUT_AS_3D)</span></div><div class="line"><a name="l03228"></a><span class="lineno"> 3228</span>&#160; <span class="comment">// Load values from matrix A</span></div><div class="line"><a name="l03229"></a><span class="lineno"> 3229</span>&#160; <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> a0 = *((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src0_ptr + src_addr.s0 + 0 * src0_stride_y));</div><div class="line"><a name="l03230"></a><span class="lineno"> 3230</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 1</span></div><div class="line"><a name="l03231"></a><span class="lineno"> 3231</span>&#160; <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> a1 = *((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src0_ptr + src_addr.s0 + 1 * src0_stride_y));</div><div class="line"><a name="l03232"></a><span class="lineno"> 3232</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 1</span></div><div class="line"><a name="l03233"></a><span class="lineno"> 3233</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 2</span></div><div class="line"><a name="l03234"></a><span class="lineno"> 3234</span>&#160; <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> a2 = *((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src0_ptr + src_addr.s0 + 2 * src0_stride_y));</div><div class="line"><a name="l03235"></a><span class="lineno"> 3235</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 2</span></div><div class="line"><a name="l03236"></a><span class="lineno"> 3236</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 3</span></div><div class="line"><a name="l03237"></a><span class="lineno"> 3237</span>&#160; <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> a3 = *((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src0_ptr + src_addr.s0 + 3 * src0_stride_y));</div><div class="line"><a name="l03238"></a><span class="lineno"> 3238</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 3</span></div><div class="line"><a name="l03239"></a><span class="lineno"> 3239</span>&#160;<span class="preprocessor">#endif // defined(REINTERPRET_INPUT_AS_3D)</span></div><div class="line"><a name="l03240"></a><span class="lineno"> 3240</span>&#160;</div><div class="line"><a name="l03241"></a><span class="lineno"> 3241</span>&#160; <span class="comment">// Load values from matrix B</span></div><div class="line"><a name="l03242"></a><span class="lineno"> 3242</span>&#160; VECTOR_TYPE 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, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src1_ptr + src_addr.s1));</div><div class="line"><a name="l03243"></a><span class="lineno"> 3243</span>&#160;</div><div class="line"><a name="l03244"></a><span class="lineno"> 3244</span>&#160; <span class="comment">// Accumulate</span></div><div class="line"><a name="l03245"></a><span class="lineno"> 3245</span>&#160; acc0 += b0 * (VECTOR_TYPE)a0;</div><div class="line"><a name="l03246"></a><span class="lineno"> 3246</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 1</span></div><div class="line"><a name="l03247"></a><span class="lineno"> 3247</span>&#160; acc1 += b0 * (VECTOR_TYPE)a1;</div><div class="line"><a name="l03248"></a><span class="lineno"> 3248</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 1</span></div><div class="line"><a name="l03249"></a><span class="lineno"> 3249</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 2</span></div><div class="line"><a name="l03250"></a><span class="lineno"> 3250</span>&#160; acc2 += b0 * (VECTOR_TYPE)a2;</div><div class="line"><a name="l03251"></a><span class="lineno"> 3251</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 2</span></div><div class="line"><a name="l03252"></a><span class="lineno"> 3252</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 3</span></div><div class="line"><a name="l03253"></a><span class="lineno"> 3253</span>&#160; acc3 += b0 * (VECTOR_TYPE)a3;</div><div class="line"><a name="l03254"></a><span class="lineno"> 3254</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 3</span></div><div class="line"><a name="l03255"></a><span class="lineno"> 3255</span>&#160; }</div><div class="line"><a name="l03256"></a><span class="lineno"> 3256</span>&#160;</div><div class="line"><a name="l03257"></a><span class="lineno"> 3257</span>&#160; <span class="comment">// Compute destination address</span></div><div class="line"><a name="l03258"></a><span class="lineno"> 3258</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="l03259"></a><span class="lineno"> 3259</span>&#160;</div><div class="line"><a name="l03260"></a><span class="lineno"> 3260</span>&#160; <span class="comment">// Compute dst address</span></div><div class="line"><a name="l03261"></a><span class="lineno"> 3261</span>&#160; __global uchar *dst_addr = <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>(&amp;<a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#adbf67dcee294e673cf796f1ed8aeb6a4">dst</a>, 0, 0);</div><div class="line"><a name="l03262"></a><span class="lineno"> 3262</span>&#160;</div><div class="line"><a name="l03263"></a><span class="lineno"> 3263</span>&#160; <span class="comment">// Multiply by the weight of matrix-matrix product and store the result</span></div><div class="line"><a name="l03264"></a><span class="lineno"> 3264</span>&#160;<span class="preprocessor">#if defined(ALPHA)</span></div><div class="line"><a name="l03265"></a><span class="lineno"> 3265</span>&#160; acc0 = acc0 * (VECTOR_TYPE)ALPHA;</div><div class="line"><a name="l03266"></a><span class="lineno"> 3266</span>&#160;<span class="preprocessor">#endif // defined(ALPHA)</span></div><div class="line"><a name="l03267"></a><span class="lineno"> 3267</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 1 &amp;&amp; defined(ALPHA)</span></div><div class="line"><a name="l03268"></a><span class="lineno"> 3268</span>&#160; acc1 = acc1 * (VECTOR_TYPE)ALPHA;</div><div class="line"><a name="l03269"></a><span class="lineno"> 3269</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 1 &amp;&amp; defined(ALPHA)</span></div><div class="line"><a name="l03270"></a><span class="lineno"> 3270</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 2 &amp;&amp; defined(ALPHA)</span></div><div class="line"><a name="l03271"></a><span class="lineno"> 3271</span>&#160; acc2 = acc2 * (VECTOR_TYPE)ALPHA;</div><div class="line"><a name="l03272"></a><span class="lineno"> 3272</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 2 &amp;&amp; defined(ALPHA)</span></div><div class="line"><a name="l03273"></a><span class="lineno"> 3273</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 3 &amp;&amp; defined(ALPHA)</span></div><div class="line"><a name="l03274"></a><span class="lineno"> 3274</span>&#160; acc3 = acc3 * (VECTOR_TYPE)ALPHA;</div><div class="line"><a name="l03275"></a><span class="lineno"> 3275</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 3 &amp;&amp; defined(ALPHA)</span></div><div class="line"><a name="l03276"></a><span class="lineno"> 3276</span>&#160;</div><div class="line"><a name="l03277"></a><span class="lineno"> 3277</span>&#160;<span class="preprocessor">#if defined(ADD_VEC_C)</span></div><div class="line"><a name="l03278"></a><span class="lineno"> 3278</span>&#160; <span class="comment">// *INDENT-OFF*</span></div><div class="line"><a name="l03279"></a><span class="lineno"> 3279</span>&#160; <span class="comment">// clang-format off</span></div><div class="line"><a name="l03280"></a><span class="lineno"> 3280</span>&#160; __global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *src2_addr = (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src2_ptr + src2_offset_first_element_in_bytes + get_global_id(0) * src2_step_x);</div><div class="line"><a name="l03281"></a><span class="lineno"> 3281</span>&#160; VECTOR_TYPE c0 = <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, src2_addr);</div><div class="line"><a name="l03282"></a><span class="lineno"> 3282</span>&#160; <span class="comment">// clang-format on</span></div><div class="line"><a name="l03283"></a><span class="lineno"> 3283</span>&#160; <span class="comment">// *INDENT-ON*</span></div><div class="line"><a name="l03284"></a><span class="lineno"> 3284</span>&#160;</div><div class="line"><a name="l03285"></a><span class="lineno"> 3285</span>&#160; acc0 += c0;</div><div class="line"><a name="l03286"></a><span class="lineno"> 3286</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 1</span></div><div class="line"><a name="l03287"></a><span class="lineno"> 3287</span>&#160; acc1 += c0;</div><div class="line"><a name="l03288"></a><span class="lineno"> 3288</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 1</span></div><div class="line"><a name="l03289"></a><span class="lineno"> 3289</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 2</span></div><div class="line"><a name="l03290"></a><span class="lineno"> 3290</span>&#160; acc2 += c0;</div><div class="line"><a name="l03291"></a><span class="lineno"> 3291</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 2</span></div><div class="line"><a name="l03292"></a><span class="lineno"> 3292</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 3</span></div><div class="line"><a name="l03293"></a><span class="lineno"> 3293</span>&#160; acc3 += c0;</div><div class="line"><a name="l03294"></a><span class="lineno"> 3294</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 3</span></div><div class="line"><a name="l03295"></a><span class="lineno"> 3295</span>&#160;<span class="preprocessor">#endif </span><span class="comment">/* defined(ADD_VEC_C) */</span><span class="preprocessor"></span></div><div class="line"><a name="l03296"></a><span class="lineno"> 3296</span>&#160;</div><div class="line"><a name="l03297"></a><span class="lineno"> 3297</span>&#160; <span class="keywordtype">int</span> z = get_global_id(2);</div><div class="line"><a name="l03298"></a><span class="lineno"> 3298</span>&#160;</div><div class="line"><a name="l03299"></a><span class="lineno"> 3299</span>&#160;<span class="preprocessor">#if defined(REINTERPRET_OUTPUT_AS_3D)</span></div><div class="line"><a name="l03300"></a><span class="lineno"> 3300</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="l03301"></a><span class="lineno"> 3301</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="l03302"></a><span class="lineno"> 3302</span>&#160; <span class="comment">//</span></div><div class="line"><a name="l03303"></a><span class="lineno"> 3303</span>&#160; <span class="comment">// | |</span></div><div class="line"><a name="l03304"></a><span class="lineno"> 3304</span>&#160; <span class="comment">// | plane0 |</span></div><div class="line"><a name="l03305"></a><span class="lineno"> 3305</span>&#160; <span class="comment">// | |</span></div><div class="line"><a name="l03306"></a><span class="lineno"> 3306</span>&#160; <span class="comment">// |__________________|</span></div><div class="line"><a name="l03307"></a><span class="lineno"> 3307</span>&#160; <span class="comment">// |******************|</span></div><div class="line"><a name="l03308"></a><span class="lineno"> 3308</span>&#160; <span class="comment">// | cross_plane_pad |</span></div><div class="line"><a name="l03309"></a><span class="lineno"> 3309</span>&#160; <span class="comment">// |******************|</span></div><div class="line"><a name="l03310"></a><span class="lineno"> 3310</span>&#160; <span class="comment">// | |</span></div><div class="line"><a name="l03311"></a><span class="lineno"> 3311</span>&#160; <span class="comment">// | plane1 |</span></div><div class="line"><a name="l03312"></a><span class="lineno"> 3312</span>&#160; <span class="comment">// | |</span></div><div class="line"><a name="l03313"></a><span class="lineno"> 3313</span>&#160; <span class="comment">// |__________________|</span></div><div class="line"><a name="l03314"></a><span class="lineno"> 3314</span>&#160;</div><div class="line"><a name="l03315"></a><span class="lineno"> 3315</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="l03316"></a><span class="lineno"> 3316</span>&#160; uint4 zout = ((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="l03317"></a><span class="lineno"> 3317</span>&#160; zout = min(DEPTH_GEMM3D - 1, zout);</div><div class="line"><a name="l03318"></a><span class="lineno"> 3318</span>&#160;</div><div class="line"><a name="l03319"></a><span class="lineno"> 3319</span>&#160; <span class="comment">// Add offset due to the cross plane paddings</span></div><div class="line"><a name="l03320"></a><span class="lineno"> 3320</span>&#160; zout *= (dst_cross_plane_pad * dst_stride_y);</div><div class="line"><a name="l03321"></a><span class="lineno"> 3321</span>&#160;</div><div class="line"><a name="l03322"></a><span class="lineno"> 3322</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="l03323"></a><span class="lineno"> 3323</span>&#160; <span class="comment">// multiply dst_stride_z by DEPTH_GEMM3D</span></div><div class="line"><a name="l03324"></a><span class="lineno"> 3324</span>&#160; dst_addr += z * dst_stride_z * DEPTH_GEMM3D;</div><div class="line"><a name="l03325"></a><span class="lineno"> 3325</span>&#160;</div><div class="line"><a name="l03326"></a><span class="lineno"> 3326</span>&#160; <span class="comment">// Store output block</span></div><div class="line"><a name="l03327"></a><span class="lineno"> 3327</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="l03328"></a><span class="lineno"> 3328</span>&#160; (acc0, 0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 0 * dst_stride_y + zout.s0));</div><div class="line"><a name="l03329"></a><span class="lineno"> 3329</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 1</span></div><div class="line"><a name="l03330"></a><span class="lineno"> 3330</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="l03331"></a><span class="lineno"> 3331</span>&#160; (acc1, 0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 1 * dst_stride_y + zout.s1));</div><div class="line"><a name="l03332"></a><span class="lineno"> 3332</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 1</span></div><div class="line"><a name="l03333"></a><span class="lineno"> 3333</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 2</span></div><div class="line"><a name="l03334"></a><span class="lineno"> 3334</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="l03335"></a><span class="lineno"> 3335</span>&#160; (acc2, 0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 2 * dst_stride_y + zout.s2));</div><div class="line"><a name="l03336"></a><span class="lineno"> 3336</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 2</span></div><div class="line"><a name="l03337"></a><span class="lineno"> 3337</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 3</span></div><div class="line"><a name="l03338"></a><span class="lineno"> 3338</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="l03339"></a><span class="lineno"> 3339</span>&#160; (acc3, 0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 3 * dst_stride_y + zout.s3));</div><div class="line"><a name="l03340"></a><span class="lineno"> 3340</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 3</span></div><div class="line"><a name="l03341"></a><span class="lineno"> 3341</span>&#160;</div><div class="line"><a name="l03342"></a><span class="lineno"> 3342</span>&#160;<span class="preprocessor">#else // defined(REINTERPRET_OUTPUT_AS_3D)</span></div><div class="line"><a name="l03343"></a><span class="lineno"> 3343</span>&#160; <span class="comment">// Add offset for batched GEMM</span></div><div class="line"><a name="l03344"></a><span class="lineno"> 3344</span>&#160; dst_addr += z * dst_stride_z;</div><div class="line"><a name="l03345"></a><span class="lineno"> 3345</span>&#160;</div><div class="line"><a name="l03346"></a><span class="lineno"> 3346</span>&#160; <span class="comment">// Store output block</span></div><div class="line"><a name="l03347"></a><span class="lineno"> 3347</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="l03348"></a><span class="lineno"> 3348</span>&#160; (acc0, 0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 0 * dst_stride_y));</div><div class="line"><a name="l03349"></a><span class="lineno"> 3349</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 1</span></div><div class="line"><a name="l03350"></a><span class="lineno"> 3350</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="l03351"></a><span class="lineno"> 3351</span>&#160; (acc1, 0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 1 * dst_stride_y));</div><div class="line"><a name="l03352"></a><span class="lineno"> 3352</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 1</span></div><div class="line"><a name="l03353"></a><span class="lineno"> 3353</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 2</span></div><div class="line"><a name="l03354"></a><span class="lineno"> 3354</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="l03355"></a><span class="lineno"> 3355</span>&#160; (acc2, 0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 2 * dst_stride_y));</div><div class="line"><a name="l03356"></a><span class="lineno"> 3356</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 2</span></div><div class="line"><a name="l03357"></a><span class="lineno"> 3357</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 3</span></div><div class="line"><a name="l03358"></a><span class="lineno"> 3358</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="l03359"></a><span class="lineno"> 3359</span>&#160; (acc3, 0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(dst_addr + 3 * dst_stride_y));</div><div class="line"><a name="l03360"></a><span class="lineno"> 3360</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 3</span></div><div class="line"><a name="l03361"></a><span class="lineno"> 3361</span>&#160;<span class="preprocessor">#endif // defined(REINTERPRET_OUTPUT_AS_3D)</span></div><div class="line"><a name="l03362"></a><span class="lineno"> 3362</span>&#160;}</div><div class="line"><a name="l03363"></a><span class="lineno"> 3363</span>&#160;<span class="preprocessor">#endif // defined(DATA_TYPE)</span></div><div class="line"><a name="l03364"></a><span class="lineno"> 3364</span>&#160;</div><div class="line"><a name="l03414"></a><span class="lineno"> 3414</span>&#160;__kernel <span class="keywordtype">void</span> gemm_mm_floating_point_f32_bifrost(<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="l03415"></a><span class="lineno"> 3415</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="l03416"></a><span class="lineno"> 3416</span>&#160;#<span class="keywordflow">if</span> defined(ADD_VEC_C)</div><div class="line"><a name="l03417"></a><span class="lineno"> 3417</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a40a6eb9f2a7712f08d6bb8ff6c9e6ca7">VECTOR_DECLARATION</a>(src2),</div><div class="line"><a name="l03418"></a><span class="lineno"> 3418</span>&#160;#endif <span class="comment">/* defined(ADD_VEC_C) */</span></div><div class="line"><a name="l03419"></a><span class="lineno"> 3419</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="l03420"></a><span class="lineno"> 3420</span>&#160; uint src0_stride_z,</div><div class="line"><a name="l03421"></a><span class="lineno"> 3421</span>&#160; uint src1_stride_z,</div><div class="line"><a name="l03422"></a><span class="lineno"> 3422</span>&#160; uint dst_stride_z</div><div class="line"><a name="l03423"></a><span class="lineno"> 3423</span>&#160;#<span class="keywordflow">if</span> defined(REINTERPRET_INPUT_AS_3D)</div><div class="line"><a name="l03424"></a><span class="lineno"> 3424</span>&#160; ,</div><div class="line"><a name="l03425"></a><span class="lineno"> 3425</span>&#160; uint src_cross_plane_pad</div><div class="line"><a name="l03426"></a><span class="lineno"> 3426</span>&#160;#endif <span class="comment">// REINTERPRET_INPUT_AS_3D</span></div><div class="line"><a name="l03427"></a><span class="lineno"> 3427</span>&#160;#<span class="keywordflow">if</span> defined(REINTERPRET_OUTPUT_AS_3D)</div><div class="line"><a name="l03428"></a><span class="lineno"> 3428</span>&#160; ,</div><div class="line"><a name="l03429"></a><span class="lineno"> 3429</span>&#160; uint dst_cross_plane_pad</div><div class="line"><a name="l03430"></a><span class="lineno"> 3430</span>&#160;#endif <span class="comment">// REINTERPRET_OUTPUT_AS_3D</span></div><div class="line"><a name="l03431"></a><span class="lineno"> 3431</span>&#160; )</div><div class="line"><a name="l03432"></a><span class="lineno"> 3432</span>&#160;{</div><div class="line"><a name="l03433"></a><span class="lineno"> 3433</span>&#160; <span class="keywordtype">int</span> idx = get_global_id(0) * NUM_ELEMS_PROCESSED_PER_THREAD_X;</div><div class="line"><a name="l03434"></a><span class="lineno"> 3434</span>&#160;</div><div class="line"><a name="l03435"></a><span class="lineno"> 3435</span>&#160; <span class="comment">// Compute starting address for matrix A and matrix B</span></div><div class="line"><a name="l03436"></a><span class="lineno"> 3436</span>&#160; int2 src_addr = ((int2)(src0_offset_first_element_in_bytes, src1_offset_first_element_in_bytes));</div><div class="line"><a name="l03437"></a><span class="lineno"> 3437</span>&#160;</div><div class="line"><a name="l03438"></a><span class="lineno"> 3438</span>&#160; <span class="comment">// Update address for matrix A</span></div><div class="line"><a name="l03439"></a><span class="lineno"> 3439</span>&#160; src_addr.s0 += get_global_id(1) * src0_stride_y * NUM_ELEMS_PROCESSED_PER_THREAD_Y;</div><div class="line"><a name="l03440"></a><span class="lineno"> 3440</span>&#160;</div><div class="line"><a name="l03441"></a><span class="lineno"> 3441</span>&#160; <span class="comment">// Update address for matrix B</span></div><div class="line"><a name="l03442"></a><span class="lineno"> 3442</span>&#160; src_addr.s1 += idx * <span class="keyword">sizeof</span>(float);</div><div class="line"><a name="l03443"></a><span class="lineno"> 3443</span>&#160;</div><div class="line"><a name="l03444"></a><span class="lineno"> 3444</span>&#160;<span class="preprocessor">#if defined(REINTERPRET_INPUT_AS_3D)</span></div><div class="line"><a name="l03445"></a><span class="lineno"> 3445</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="l03446"></a><span class="lineno"> 3446</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="l03447"></a><span class="lineno"> 3447</span>&#160; <span class="comment">//</span></div><div class="line"><a name="l03448"></a><span class="lineno"> 3448</span>&#160; <span class="comment">// | |</span></div><div class="line"><a name="l03449"></a><span class="lineno"> 3449</span>&#160; <span class="comment">// | plane0 |</span></div><div class="line"><a name="l03450"></a><span class="lineno"> 3450</span>&#160; <span class="comment">// | |</span></div><div class="line"><a name="l03451"></a><span class="lineno"> 3451</span>&#160; <span class="comment">// |__________________|</span></div><div class="line"><a name="l03452"></a><span class="lineno"> 3452</span>&#160; <span class="comment">// |******************|</span></div><div class="line"><a name="l03453"></a><span class="lineno"> 3453</span>&#160; <span class="comment">// | cross_plane_pad |</span></div><div class="line"><a name="l03454"></a><span class="lineno"> 3454</span>&#160; <span class="comment">// |******************|</span></div><div class="line"><a name="l03455"></a><span class="lineno"> 3455</span>&#160; <span class="comment">// | |</span></div><div class="line"><a name="l03456"></a><span class="lineno"> 3456</span>&#160; <span class="comment">// | plane1 |</span></div><div class="line"><a name="l03457"></a><span class="lineno"> 3457</span>&#160; <span class="comment">// | |</span></div><div class="line"><a name="l03458"></a><span class="lineno"> 3458</span>&#160; <span class="comment">// |__________________|</span></div><div class="line"><a name="l03459"></a><span class="lineno"> 3459</span>&#160;</div><div class="line"><a name="l03460"></a><span class="lineno"> 3460</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="l03461"></a><span class="lineno"> 3461</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="l03462"></a><span class="lineno"> 3462</span>&#160; zin = min(DEPTH_GEMM3D - 1, zin);</div><div class="line"><a name="l03463"></a><span class="lineno"> 3463</span>&#160;</div><div class="line"><a name="l03464"></a><span class="lineno"> 3464</span>&#160; <span class="comment">// Add offset due to the cross plane paddings</span></div><div class="line"><a name="l03465"></a><span class="lineno"> 3465</span>&#160; zin *= (src_cross_plane_pad * src0_stride_y);</div><div class="line"><a name="l03466"></a><span class="lineno"> 3466</span>&#160;</div><div class="line"><a name="l03467"></a><span class="lineno"> 3467</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="l03468"></a><span class="lineno"> 3468</span>&#160; <span class="comment">// multiply src0_stride_z by DEPTH_GEMM3D</span></div><div class="line"><a name="l03469"></a><span class="lineno"> 3469</span>&#160; src_addr.s0 += get_global_id(2) * src0_stride_z * DEPTH_GEMM3D;</div><div class="line"><a name="l03470"></a><span class="lineno"> 3470</span>&#160;</div><div class="line"><a name="l03471"></a><span class="lineno"> 3471</span>&#160;<span class="preprocessor">#else // defined(REINTERPRET_INPUT_AS_3D)</span></div><div class="line"><a name="l03472"></a><span class="lineno"> 3472</span>&#160;</div><div class="line"><a name="l03473"></a><span class="lineno"> 3473</span>&#160; <span class="comment">// Add offset for batched GEMM</span></div><div class="line"><a name="l03474"></a><span class="lineno"> 3474</span>&#160; src_addr.s0 += get_global_id(2) * src0_stride_z;</div><div class="line"><a name="l03475"></a><span class="lineno"> 3475</span>&#160;</div><div class="line"><a name="l03476"></a><span class="lineno"> 3476</span>&#160;<span class="preprocessor">#endif // defined(REINTERPRET_INPUT_AS_3D)</span></div><div class="line"><a name="l03477"></a><span class="lineno"> 3477</span>&#160;</div><div class="line"><a name="l03478"></a><span class="lineno"> 3478</span>&#160;<span class="preprocessor">#if defined(MATRIX_B_DEPTH)</span></div><div class="line"><a name="l03479"></a><span class="lineno"> 3479</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="l03480"></a><span class="lineno"> 3480</span>&#160; src_addr.s1 += (get_global_id(2) % MATRIX_B_DEPTH) * src1_stride_z;</div><div class="line"><a name="l03481"></a><span class="lineno"> 3481</span>&#160;<span class="preprocessor">#else // defined(MATRIX_B_DEPTH)</span></div><div class="line"><a name="l03482"></a><span class="lineno"> 3482</span>&#160; src_addr.s1 += get_global_id(2) * src1_stride_z;</div><div class="line"><a name="l03483"></a><span class="lineno"> 3483</span>&#160;<span class="preprocessor">#endif // defined(MATRIX_B_DEPTH)</span></div><div class="line"><a name="l03484"></a><span class="lineno"> 3484</span>&#160;</div><div class="line"><a name="l03485"></a><span class="lineno"> 3485</span>&#160; <span class="comment">// Initialize accumulators</span></div><div class="line"><a name="l03486"></a><span class="lineno"> 3486</span>&#160; <span class="keywordtype">float</span> acc00 = 0.0f;</div><div class="line"><a name="l03487"></a><span class="lineno"> 3487</span>&#160; <span class="keywordtype">float</span> acc01 = 0.0f;</div><div class="line"><a name="l03488"></a><span class="lineno"> 3488</span>&#160; <span class="keywordtype">float</span> acc02 = 0.0f;</div><div class="line"><a name="l03489"></a><span class="lineno"> 3489</span>&#160; <span class="keywordtype">float</span> acc03 = 0.0f;</div><div class="line"><a name="l03490"></a><span class="lineno"> 3490</span>&#160;</div><div class="line"><a name="l03491"></a><span class="lineno"> 3491</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 1</span></div><div class="line"><a name="l03492"></a><span class="lineno"> 3492</span>&#160; <span class="keywordtype">float</span> acc10 = 0.0f;</div><div class="line"><a name="l03493"></a><span class="lineno"> 3493</span>&#160; <span class="keywordtype">float</span> acc11 = 0.0f;</div><div class="line"><a name="l03494"></a><span class="lineno"> 3494</span>&#160; <span class="keywordtype">float</span> acc12 = 0.0f;</div><div class="line"><a name="l03495"></a><span class="lineno"> 3495</span>&#160; <span class="keywordtype">float</span> acc13 = 0.0f;</div><div class="line"><a name="l03496"></a><span class="lineno"> 3496</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 1</span></div><div class="line"><a name="l03497"></a><span class="lineno"> 3497</span>&#160;</div><div class="line"><a name="l03498"></a><span class="lineno"> 3498</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 2</span></div><div class="line"><a name="l03499"></a><span class="lineno"> 3499</span>&#160; <span class="keywordtype">float</span> acc20 = 0.0f;</div><div class="line"><a name="l03500"></a><span class="lineno"> 3500</span>&#160; <span class="keywordtype">float</span> acc21 = 0.0f;</div><div class="line"><a name="l03501"></a><span class="lineno"> 3501</span>&#160; <span class="keywordtype">float</span> acc22 = 0.0f;</div><div class="line"><a name="l03502"></a><span class="lineno"> 3502</span>&#160; <span class="keywordtype">float</span> acc23 = 0.0f;</div><div class="line"><a name="l03503"></a><span class="lineno"> 3503</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 2</span></div><div class="line"><a name="l03504"></a><span class="lineno"> 3504</span>&#160;</div><div class="line"><a name="l03505"></a><span class="lineno"> 3505</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 3</span></div><div class="line"><a name="l03506"></a><span class="lineno"> 3506</span>&#160; <span class="keywordtype">float</span> acc30 = 0.0f;</div><div class="line"><a name="l03507"></a><span class="lineno"> 3507</span>&#160; <span class="keywordtype">float</span> acc31 = 0.0f;</div><div class="line"><a name="l03508"></a><span class="lineno"> 3508</span>&#160; <span class="keywordtype">float</span> acc32 = 0.0f;</div><div class="line"><a name="l03509"></a><span class="lineno"> 3509</span>&#160; <span class="keywordtype">float</span> acc33 = 0.0f;</div><div class="line"><a name="l03510"></a><span class="lineno"> 3510</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 3</span></div><div class="line"><a name="l03511"></a><span class="lineno"> 3511</span>&#160;</div><div class="line"><a name="l03512"></a><span class="lineno"> 3512</span>&#160; <span class="comment">// A and B src indices get incremented at the same time.</span></div><div class="line"><a name="l03513"></a><span class="lineno"> 3513</span>&#160; <span class="keywordtype">int</span> i = 0;</div><div class="line"><a name="l03514"></a><span class="lineno"> 3514</span>&#160; <span class="keywordflow">for</span>(; i &lt;= ((int)COLS_A - 4); i += 4)</div><div class="line"><a name="l03515"></a><span class="lineno"> 3515</span>&#160; {</div><div class="line"><a name="l03516"></a><span class="lineno"> 3516</span>&#160;<span class="preprocessor">#if defined(REINTERPRET_INPUT_AS_3D)</span></div><div class="line"><a name="l03517"></a><span class="lineno"> 3517</span>&#160; <span class="comment">// Load values from matrix A and matrix B</span></div><div class="line"><a name="l03518"></a><span class="lineno"> 3518</span>&#160; float4 a0 = vload4(0, (__global <span class="keywordtype">float</span> *)(src0_ptr + src_addr.s0 + 0 * src0_stride_y + zin.s0));</div><div class="line"><a name="l03519"></a><span class="lineno"> 3519</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 1</span></div><div class="line"><a name="l03520"></a><span class="lineno"> 3520</span>&#160; float4 a1 = vload4(0, (__global <span class="keywordtype">float</span> *)(src0_ptr + src_addr.s0 + 1 * src0_stride_y + zin.s1));</div><div class="line"><a name="l03521"></a><span class="lineno"> 3521</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 1</span></div><div class="line"><a name="l03522"></a><span class="lineno"> 3522</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 2</span></div><div class="line"><a name="l03523"></a><span class="lineno"> 3523</span>&#160; float4 a2 = vload4(0, (__global <span class="keywordtype">float</span> *)(src0_ptr + src_addr.s0 + 2 * src0_stride_y + zin.s2));</div><div class="line"><a name="l03524"></a><span class="lineno"> 3524</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 2</span></div><div class="line"><a name="l03525"></a><span class="lineno"> 3525</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 3</span></div><div class="line"><a name="l03526"></a><span class="lineno"> 3526</span>&#160; float4 a3 = vload4(0, (__global <span class="keywordtype">float</span> *)(src0_ptr + src_addr.s0 + 3 * src0_stride_y + zin.s3));</div><div class="line"><a name="l03527"></a><span class="lineno"> 3527</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 3</span></div><div class="line"><a name="l03528"></a><span class="lineno"> 3528</span>&#160;<span class="preprocessor">#else // defined(REINTERPRET_INPUT_AS_3D)</span></div><div class="line"><a name="l03529"></a><span class="lineno"> 3529</span>&#160; <span class="comment">// Load values from matrix A and matrix B</span></div><div class="line"><a name="l03530"></a><span class="lineno"> 3530</span>&#160; float4 a0 = vload4(0, (__global <span class="keywordtype">float</span> *)(src0_ptr + src_addr.s0 + 0 * src0_stride_y));</div><div class="line"><a name="l03531"></a><span class="lineno"> 3531</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 1</span></div><div class="line"><a name="l03532"></a><span class="lineno"> 3532</span>&#160; float4 a1 = vload4(0, (__global <span class="keywordtype">float</span> *)(src0_ptr + src_addr.s0 + 1 * src0_stride_y));</div><div class="line"><a name="l03533"></a><span class="lineno"> 3533</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 1</span></div><div class="line"><a name="l03534"></a><span class="lineno"> 3534</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 2</span></div><div class="line"><a name="l03535"></a><span class="lineno"> 3535</span>&#160; float4 a2 = vload4(0, (__global <span class="keywordtype">float</span> *)(src0_ptr + src_addr.s0 + 2 * src0_stride_y));</div><div class="line"><a name="l03536"></a><span class="lineno"> 3536</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 2</span></div><div class="line"><a name="l03537"></a><span class="lineno"> 3537</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 3</span></div><div class="line"><a name="l03538"></a><span class="lineno"> 3538</span>&#160; float4 a3 = vload4(0, (__global <span class="keywordtype">float</span> *)(src0_ptr + src_addr.s0 + 3 * src0_stride_y));</div><div class="line"><a name="l03539"></a><span class="lineno"> 3539</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 3</span></div><div class="line"><a name="l03540"></a><span class="lineno"> 3540</span>&#160;<span class="preprocessor">#endif // defined(REINTERPRET_INPUT_AS_3D)</span></div><div class="line"><a name="l03541"></a><span class="lineno"> 3541</span>&#160;</div><div class="line"><a name="l03542"></a><span class="lineno"> 3542</span>&#160; float4 b0 = vload4(0, (__global <span class="keywordtype">float</span> *)(src1_ptr + src_addr.s1));</div><div class="line"><a name="l03543"></a><span class="lineno"> 3543</span>&#160; src_addr.s1 += src1_stride_y;</div><div class="line"><a name="l03544"></a><span class="lineno"> 3544</span>&#160;</div><div class="line"><a name="l03545"></a><span class="lineno"> 3545</span>&#160; <span class="comment">// Multiply and accumulate</span></div><div class="line"><a name="l03546"></a><span class="lineno"> 3546</span>&#160; acc00 = fma(a0.s0, b0.s0, acc00);</div><div class="line"><a name="l03547"></a><span class="lineno"> 3547</span>&#160; acc01 = fma(a0.s0, b0.s1, acc01);</div><div class="line"><a name="l03548"></a><span class="lineno"> 3548</span>&#160; acc02 = fma(a0.s0, b0.s2, acc02);</div><div class="line"><a name="l03549"></a><span class="lineno"> 3549</span>&#160; acc03 = fma(a0.s0, b0.s3, acc03);</div><div class="line"><a name="l03550"></a><span class="lineno"> 3550</span>&#160;</div><div class="line"><a name="l03551"></a><span class="lineno"> 3551</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 1</span></div><div class="line"><a name="l03552"></a><span class="lineno"> 3552</span>&#160;</div><div class="line"><a name="l03553"></a><span class="lineno"> 3553</span>&#160; acc10 = fma(a1.s0, b0.s0, acc10);</div><div class="line"><a name="l03554"></a><span class="lineno"> 3554</span>&#160; acc11 = fma(a1.s0, b0.s1, acc11);</div><div class="line"><a name="l03555"></a><span class="lineno"> 3555</span>&#160; acc12 = fma(a1.s0, b0.s2, acc12);</div><div class="line"><a name="l03556"></a><span class="lineno"> 3556</span>&#160; acc13 = fma(a1.s0, b0.s3, acc13);</div><div class="line"><a name="l03557"></a><span class="lineno"> 3557</span>&#160;</div><div class="line"><a name="l03558"></a><span class="lineno"> 3558</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 1</span></div><div class="line"><a name="l03559"></a><span class="lineno"> 3559</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 2</span></div><div class="line"><a name="l03560"></a><span class="lineno"> 3560</span>&#160;</div><div class="line"><a name="l03561"></a><span class="lineno"> 3561</span>&#160; acc20 = fma(a2.s0, b0.s0, acc20);</div><div class="line"><a name="l03562"></a><span class="lineno"> 3562</span>&#160; acc21 = fma(a2.s0, b0.s1, acc21);</div><div class="line"><a name="l03563"></a><span class="lineno"> 3563</span>&#160; acc22 = fma(a2.s0, b0.s2, acc22);</div><div class="line"><a name="l03564"></a><span class="lineno"> 3564</span>&#160; acc23 = fma(a2.s0, b0.s3, acc23);</div><div class="line"><a name="l03565"></a><span class="lineno"> 3565</span>&#160;</div><div class="line"><a name="l03566"></a><span class="lineno"> 3566</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 2</span></div><div class="line"><a name="l03567"></a><span class="lineno"> 3567</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 3</span></div><div class="line"><a name="l03568"></a><span class="lineno"> 3568</span>&#160;</div><div class="line"><a name="l03569"></a><span class="lineno"> 3569</span>&#160; acc30 = fma(a3.s0, b0.s0, acc30);</div><div class="line"><a name="l03570"></a><span class="lineno"> 3570</span>&#160; acc31 = fma(a3.s0, b0.s1, acc31);</div><div class="line"><a name="l03571"></a><span class="lineno"> 3571</span>&#160; acc32 = fma(a3.s0, b0.s2, acc32);</div><div class="line"><a name="l03572"></a><span class="lineno"> 3572</span>&#160; acc33 = fma(a3.s0, b0.s3, acc33);</div><div class="line"><a name="l03573"></a><span class="lineno"> 3573</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 3</span></div><div class="line"><a name="l03574"></a><span class="lineno"> 3574</span>&#160;</div><div class="line"><a name="l03575"></a><span class="lineno"> 3575</span>&#160; <span class="comment">// Load values from matrix A and matrix B</span></div><div class="line"><a name="l03576"></a><span class="lineno"> 3576</span>&#160; b0 = vload4(0, (__global <span class="keywordtype">float</span> *)(src1_ptr + src_addr.s1));</div><div class="line"><a name="l03577"></a><span class="lineno"> 3577</span>&#160; src_addr.s1 += src1_stride_y;</div><div class="line"><a name="l03578"></a><span class="lineno"> 3578</span>&#160;</div><div class="line"><a name="l03579"></a><span class="lineno"> 3579</span>&#160; <span class="comment">// Multiply and accumulate</span></div><div class="line"><a name="l03580"></a><span class="lineno"> 3580</span>&#160; acc00 = fma(a0.s1, b0.s0, acc00);</div><div class="line"><a name="l03581"></a><span class="lineno"> 3581</span>&#160; acc01 = fma(a0.s1, b0.s1, acc01);</div><div class="line"><a name="l03582"></a><span class="lineno"> 3582</span>&#160; acc02 = fma(a0.s1, b0.s2, acc02);</div><div class="line"><a name="l03583"></a><span class="lineno"> 3583</span>&#160; acc03 = fma(a0.s1, b0.s3, acc03);</div><div class="line"><a name="l03584"></a><span class="lineno"> 3584</span>&#160;</div><div class="line"><a name="l03585"></a><span class="lineno"> 3585</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 1</span></div><div class="line"><a name="l03586"></a><span class="lineno"> 3586</span>&#160;</div><div class="line"><a name="l03587"></a><span class="lineno"> 3587</span>&#160; acc10 = fma(a1.s1, b0.s0, acc10);</div><div class="line"><a name="l03588"></a><span class="lineno"> 3588</span>&#160; acc11 = fma(a1.s1, b0.s1, acc11);</div><div class="line"><a name="l03589"></a><span class="lineno"> 3589</span>&#160; acc12 = fma(a1.s1, b0.s2, acc12);</div><div class="line"><a name="l03590"></a><span class="lineno"> 3590</span>&#160; acc13 = fma(a1.s1, b0.s3, acc13);</div><div class="line"><a name="l03591"></a><span class="lineno"> 3591</span>&#160;</div><div class="line"><a name="l03592"></a><span class="lineno"> 3592</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 1</span></div><div class="line"><a name="l03593"></a><span class="lineno"> 3593</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 2</span></div><div class="line"><a name="l03594"></a><span class="lineno"> 3594</span>&#160;</div><div class="line"><a name="l03595"></a><span class="lineno"> 3595</span>&#160; acc20 = fma(a2.s1, b0.s0, acc20);</div><div class="line"><a name="l03596"></a><span class="lineno"> 3596</span>&#160; acc21 = fma(a2.s1, b0.s1, acc21);</div><div class="line"><a name="l03597"></a><span class="lineno"> 3597</span>&#160; acc22 = fma(a2.s1, b0.s2, acc22);</div><div class="line"><a name="l03598"></a><span class="lineno"> 3598</span>&#160; acc23 = fma(a2.s1, b0.s3, acc23);</div><div class="line"><a name="l03599"></a><span class="lineno"> 3599</span>&#160;</div><div class="line"><a name="l03600"></a><span class="lineno"> 3600</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 2</span></div><div class="line"><a name="l03601"></a><span class="lineno"> 3601</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 3</span></div><div class="line"><a name="l03602"></a><span class="lineno"> 3602</span>&#160;</div><div class="line"><a name="l03603"></a><span class="lineno"> 3603</span>&#160; acc30 = fma(a3.s1, b0.s0, acc30);</div><div class="line"><a name="l03604"></a><span class="lineno"> 3604</span>&#160; acc31 = fma(a3.s1, b0.s1, acc31);</div><div class="line"><a name="l03605"></a><span class="lineno"> 3605</span>&#160; acc32 = fma(a3.s1, b0.s2, acc32);</div><div class="line"><a name="l03606"></a><span class="lineno"> 3606</span>&#160; acc33 = fma(a3.s1, b0.s3, acc33);</div><div class="line"><a name="l03607"></a><span class="lineno"> 3607</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 3</span></div><div class="line"><a name="l03608"></a><span class="lineno"> 3608</span>&#160;</div><div class="line"><a name="l03609"></a><span class="lineno"> 3609</span>&#160; <span class="comment">// Load values from matrix A and matrix B</span></div><div class="line"><a name="l03610"></a><span class="lineno"> 3610</span>&#160; b0 = vload4(0, (__global <span class="keywordtype">float</span> *)(src1_ptr + src_addr.s1));</div><div class="line"><a name="l03611"></a><span class="lineno"> 3611</span>&#160; src_addr.s1 += src1_stride_y;</div><div class="line"><a name="l03612"></a><span class="lineno"> 3612</span>&#160;</div><div class="line"><a name="l03613"></a><span class="lineno"> 3613</span>&#160; <span class="comment">// Multiply and accumulate</span></div><div class="line"><a name="l03614"></a><span class="lineno"> 3614</span>&#160; acc00 = fma(a0.s2, b0.s0, acc00);</div><div class="line"><a name="l03615"></a><span class="lineno"> 3615</span>&#160; acc01 = fma(a0.s2, b0.s1, acc01);</div><div class="line"><a name="l03616"></a><span class="lineno"> 3616</span>&#160; acc02 = fma(a0.s2, b0.s2, acc02);</div><div class="line"><a name="l03617"></a><span class="lineno"> 3617</span>&#160; acc03 = fma(a0.s2, b0.s3, acc03);</div><div class="line"><a name="l03618"></a><span class="lineno"> 3618</span>&#160;</div><div class="line"><a name="l03619"></a><span class="lineno"> 3619</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 1</span></div><div class="line"><a name="l03620"></a><span class="lineno"> 3620</span>&#160;</div><div class="line"><a name="l03621"></a><span class="lineno"> 3621</span>&#160; acc10 = fma(a1.s2, b0.s0, acc10);</div><div class="line"><a name="l03622"></a><span class="lineno"> 3622</span>&#160; acc11 = fma(a1.s2, b0.s1, acc11);</div><div class="line"><a name="l03623"></a><span class="lineno"> 3623</span>&#160; acc12 = fma(a1.s2, b0.s2, acc12);</div><div class="line"><a name="l03624"></a><span class="lineno"> 3624</span>&#160; acc13 = fma(a1.s2, b0.s3, acc13);</div><div class="line"><a name="l03625"></a><span class="lineno"> 3625</span>&#160;</div><div class="line"><a name="l03626"></a><span class="lineno"> 3626</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 1</span></div><div class="line"><a name="l03627"></a><span class="lineno"> 3627</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 2</span></div><div class="line"><a name="l03628"></a><span class="lineno"> 3628</span>&#160;</div><div class="line"><a name="l03629"></a><span class="lineno"> 3629</span>&#160; acc20 = fma(a2.s2, b0.s0, acc20);</div><div class="line"><a name="l03630"></a><span class="lineno"> 3630</span>&#160; acc21 = fma(a2.s2, b0.s1, acc21);</div><div class="line"><a name="l03631"></a><span class="lineno"> 3631</span>&#160; acc22 = fma(a2.s2, b0.s2, acc22);</div><div class="line"><a name="l03632"></a><span class="lineno"> 3632</span>&#160; acc23 = fma(a2.s2, b0.s3, acc23);</div><div class="line"><a name="l03633"></a><span class="lineno"> 3633</span>&#160;</div><div class="line"><a name="l03634"></a><span class="lineno"> 3634</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 2</span></div><div class="line"><a name="l03635"></a><span class="lineno"> 3635</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 3</span></div><div class="line"><a name="l03636"></a><span class="lineno"> 3636</span>&#160;</div><div class="line"><a name="l03637"></a><span class="lineno"> 3637</span>&#160; acc30 = fma(a3.s2, b0.s0, acc30);</div><div class="line"><a name="l03638"></a><span class="lineno"> 3638</span>&#160; acc31 = fma(a3.s2, b0.s1, acc31);</div><div class="line"><a name="l03639"></a><span class="lineno"> 3639</span>&#160; acc32 = fma(a3.s2, b0.s2, acc32);</div><div class="line"><a name="l03640"></a><span class="lineno"> 3640</span>&#160; acc33 = fma(a3.s2, b0.s3, acc33);</div><div class="line"><a name="l03641"></a><span class="lineno"> 3641</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 3</span></div><div class="line"><a name="l03642"></a><span class="lineno"> 3642</span>&#160;</div><div class="line"><a name="l03643"></a><span class="lineno"> 3643</span>&#160; <span class="comment">// Load values from matrix A and matrix B</span></div><div class="line"><a name="l03644"></a><span class="lineno"> 3644</span>&#160; b0 = vload4(0, (__global <span class="keywordtype">float</span> *)(src1_ptr + src_addr.s1));</div><div class="line"><a name="l03645"></a><span class="lineno"> 3645</span>&#160; src_addr.s1 += src1_stride_y;</div><div class="line"><a name="l03646"></a><span class="lineno"> 3646</span>&#160;</div><div class="line"><a name="l03647"></a><span class="lineno"> 3647</span>&#160; <span class="comment">// Multiply and accumulate</span></div><div class="line"><a name="l03648"></a><span class="lineno"> 3648</span>&#160; acc00 = fma(a0.s3, b0.s0, acc00);</div><div class="line"><a name="l03649"></a><span class="lineno"> 3649</span>&#160; acc01 = fma(a0.s3, b0.s1, acc01);</div><div class="line"><a name="l03650"></a><span class="lineno"> 3650</span>&#160; acc02 = fma(a0.s3, b0.s2, acc02);</div><div class="line"><a name="l03651"></a><span class="lineno"> 3651</span>&#160; acc03 = fma(a0.s3, b0.s3, acc03);</div><div class="line"><a name="l03652"></a><span class="lineno"> 3652</span>&#160;</div><div class="line"><a name="l03653"></a><span class="lineno"> 3653</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 1</span></div><div class="line"><a name="l03654"></a><span class="lineno"> 3654</span>&#160;</div><div class="line"><a name="l03655"></a><span class="lineno"> 3655</span>&#160; acc10 = fma(a1.s3, b0.s0, acc10);</div><div class="line"><a name="l03656"></a><span class="lineno"> 3656</span>&#160; acc11 = fma(a1.s3, b0.s1, acc11);</div><div class="line"><a name="l03657"></a><span class="lineno"> 3657</span>&#160; acc12 = fma(a1.s3, b0.s2, acc12);</div><div class="line"><a name="l03658"></a><span class="lineno"> 3658</span>&#160; acc13 = fma(a1.s3, b0.s3, acc13);</div><div class="line"><a name="l03659"></a><span class="lineno"> 3659</span>&#160;</div><div class="line"><a name="l03660"></a><span class="lineno"> 3660</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 1</span></div><div class="line"><a name="l03661"></a><span class="lineno"> 3661</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 2</span></div><div class="line"><a name="l03662"></a><span class="lineno"> 3662</span>&#160;</div><div class="line"><a name="l03663"></a><span class="lineno"> 3663</span>&#160; acc20 = fma(a2.s3, b0.s0, acc20);</div><div class="line"><a name="l03664"></a><span class="lineno"> 3664</span>&#160; acc21 = fma(a2.s3, b0.s1, acc21);</div><div class="line"><a name="l03665"></a><span class="lineno"> 3665</span>&#160; acc22 = fma(a2.s3, b0.s2, acc22);</div><div class="line"><a name="l03666"></a><span class="lineno"> 3666</span>&#160; acc23 = fma(a2.s3, b0.s3, acc23);</div><div class="line"><a name="l03667"></a><span class="lineno"> 3667</span>&#160;</div><div class="line"><a name="l03668"></a><span class="lineno"> 3668</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 2</span></div><div class="line"><a name="l03669"></a><span class="lineno"> 3669</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 3</span></div><div class="line"><a name="l03670"></a><span class="lineno"> 3670</span>&#160;</div><div class="line"><a name="l03671"></a><span class="lineno"> 3671</span>&#160; acc30 = fma(a3.s3, b0.s0, acc30);</div><div class="line"><a name="l03672"></a><span class="lineno"> 3672</span>&#160; acc31 = fma(a3.s3, b0.s1, acc31);</div><div class="line"><a name="l03673"></a><span class="lineno"> 3673</span>&#160; acc32 = fma(a3.s3, b0.s2, acc32);</div><div class="line"><a name="l03674"></a><span class="lineno"> 3674</span>&#160; acc33 = fma(a3.s3, b0.s3, acc33);</div><div class="line"><a name="l03675"></a><span class="lineno"> 3675</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 3</span></div><div class="line"><a name="l03676"></a><span class="lineno"> 3676</span>&#160;</div><div class="line"><a name="l03677"></a><span class="lineno"> 3677</span>&#160; src_addr.s0 += 4 * <span class="keyword">sizeof</span>(float);</div><div class="line"><a name="l03678"></a><span class="lineno"> 3678</span>&#160; }</div><div class="line"><a name="l03679"></a><span class="lineno"> 3679</span>&#160;</div><div class="line"><a name="l03680"></a><span class="lineno"> 3680</span>&#160; <span class="keywordflow">for</span>(; i &lt; (int)COLS_A; ++i)</div><div class="line"><a name="l03681"></a><span class="lineno"> 3681</span>&#160; {</div><div class="line"><a name="l03682"></a><span class="lineno"> 3682</span>&#160;<span class="preprocessor">#if defined(REINTERPRET_INPUT_AS_3D)</span></div><div class="line"><a name="l03683"></a><span class="lineno"> 3683</span>&#160; <span class="comment">// Load values from matrix A</span></div><div class="line"><a name="l03684"></a><span class="lineno"> 3684</span>&#160; <span class="keywordtype">float</span> a0 = *((__global <span class="keywordtype">float</span> *)(src0_ptr + src_addr.s0 + 0 * src0_stride_y + zin.s0));</div><div class="line"><a name="l03685"></a><span class="lineno"> 3685</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 1</span></div><div class="line"><a name="l03686"></a><span class="lineno"> 3686</span>&#160; <span class="keywordtype">float</span> a1 = *((__global <span class="keywordtype">float</span> *)(src0_ptr + src_addr.s0 + 1 * src0_stride_y + zin.s1));</div><div class="line"><a name="l03687"></a><span class="lineno"> 3687</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 1</span></div><div class="line"><a name="l03688"></a><span class="lineno"> 3688</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 2</span></div><div class="line"><a name="l03689"></a><span class="lineno"> 3689</span>&#160; <span class="keywordtype">float</span> a2 = *((__global <span class="keywordtype">float</span> *)(src0_ptr + src_addr.s0 + 2 * src0_stride_y + zin.s2));</div><div class="line"><a name="l03690"></a><span class="lineno"> 3690</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 2</span></div><div class="line"><a name="l03691"></a><span class="lineno"> 3691</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 3</span></div><div class="line"><a name="l03692"></a><span class="lineno"> 3692</span>&#160; <span class="keywordtype">float</span> a3 = *((__global <span class="keywordtype">float</span> *)(src0_ptr + src_addr.s0 + 3 * src0_stride_y + zin.s3));</div><div class="line"><a name="l03693"></a><span class="lineno"> 3693</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 3</span></div><div class="line"><a name="l03694"></a><span class="lineno"> 3694</span>&#160;<span class="preprocessor">#else // defined(REINTERPRET_INPUT_AS_3D)</span></div><div class="line"><a name="l03695"></a><span class="lineno"> 3695</span>&#160; <span class="comment">// Load values from matrix A</span></div><div class="line"><a name="l03696"></a><span class="lineno"> 3696</span>&#160; <span class="keywordtype">float</span> a0 = *((__global <span class="keywordtype">float</span> *)(src0_ptr + src_addr.s0 + 0 * src0_stride_y));</div><div class="line"><a name="l03697"></a><span class="lineno"> 3697</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 1</span></div><div class="line"><a name="l03698"></a><span class="lineno"> 3698</span>&#160; <span class="keywordtype">float</span> a1 = *((__global <span class="keywordtype">float</span> *)(src0_ptr + src_addr.s0 + 1 * src0_stride_y));</div><div class="line"><a name="l03699"></a><span class="lineno"> 3699</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 1</span></div><div class="line"><a name="l03700"></a><span class="lineno"> 3700</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 2</span></div><div class="line"><a name="l03701"></a><span class="lineno"> 3701</span>&#160; <span class="keywordtype">float</span> a2 = *((__global <span class="keywordtype">float</span> *)(src0_ptr + src_addr.s0 + 2 * src0_stride_y));</div><div class="line"><a name="l03702"></a><span class="lineno"> 3702</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 2</span></div><div class="line"><a name="l03703"></a><span class="lineno"> 3703</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 3</span></div><div class="line"><a name="l03704"></a><span class="lineno"> 3704</span>&#160; <span class="keywordtype">float</span> a3 = *((__global <span class="keywordtype">float</span> *)(src0_ptr + src_addr.s0 + 3 * src0_stride_y));</div><div class="line"><a name="l03705"></a><span class="lineno"> 3705</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 3</span></div><div class="line"><a name="l03706"></a><span class="lineno"> 3706</span>&#160;<span class="preprocessor">#endif // defined(REINTERPRET_INPUT_AS_3D)</span></div><div class="line"><a name="l03707"></a><span class="lineno"> 3707</span>&#160;</div><div class="line"><a name="l03708"></a><span class="lineno"> 3708</span>&#160; <span class="comment">// Load values from matrix B</span></div><div class="line"><a name="l03709"></a><span class="lineno"> 3709</span>&#160; float4 b0 = vload4(0, (__global <span class="keywordtype">float</span> *)(src1_ptr + src_addr.s1));</div><div class="line"><a name="l03710"></a><span class="lineno"> 3710</span>&#160; src_addr.s1 += src1_stride_y;</div><div class="line"><a name="l03711"></a><span class="lineno"> 3711</span>&#160;</div><div class="line"><a name="l03712"></a><span class="lineno"> 3712</span>&#160; <span class="comment">// Multiply and accumulate</span></div><div class="line"><a name="l03713"></a><span class="lineno"> 3713</span>&#160; acc00 = fma(a0, b0.s0, acc00);</div><div class="line"><a name="l03714"></a><span class="lineno"> 3714</span>&#160; acc01 = fma(a0, b0.s1, acc01);</div><div class="line"><a name="l03715"></a><span class="lineno"> 3715</span>&#160; acc02 = fma(a0, b0.s2, acc02);</div><div class="line"><a name="l03716"></a><span class="lineno"> 3716</span>&#160; acc03 = fma(a0, b0.s3, acc03);</div><div class="line"><a name="l03717"></a><span class="lineno"> 3717</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 1</span></div><div class="line"><a name="l03718"></a><span class="lineno"> 3718</span>&#160; acc10 = fma(a1, b0.s0, acc10);</div><div class="line"><a name="l03719"></a><span class="lineno"> 3719</span>&#160; acc11 = fma(a1, b0.s1, acc11);</div><div class="line"><a name="l03720"></a><span class="lineno"> 3720</span>&#160; acc12 = fma(a1, b0.s2, acc12);</div><div class="line"><a name="l03721"></a><span class="lineno"> 3721</span>&#160; acc13 = fma(a1, b0.s3, acc13);</div><div class="line"><a name="l03722"></a><span class="lineno"> 3722</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 1</span></div><div class="line"><a name="l03723"></a><span class="lineno"> 3723</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 2</span></div><div class="line"><a name="l03724"></a><span class="lineno"> 3724</span>&#160; acc20 = fma(a2, b0.s0, acc20);</div><div class="line"><a name="l03725"></a><span class="lineno"> 3725</span>&#160; acc21 = fma(a2, b0.s1, acc21);</div><div class="line"><a name="l03726"></a><span class="lineno"> 3726</span>&#160; acc22 = fma(a2, b0.s2, acc22);</div><div class="line"><a name="l03727"></a><span class="lineno"> 3727</span>&#160; acc23 = fma(a2, b0.s3, acc23);</div><div class="line"><a name="l03728"></a><span class="lineno"> 3728</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 2</span></div><div class="line"><a name="l03729"></a><span class="lineno"> 3729</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 3</span></div><div class="line"><a name="l03730"></a><span class="lineno"> 3730</span>&#160; acc30 = fma(a3, b0.s0, acc30);</div><div class="line"><a name="l03731"></a><span class="lineno"> 3731</span>&#160; acc31 = fma(a3, b0.s1, acc31);</div><div class="line"><a name="l03732"></a><span class="lineno"> 3732</span>&#160; acc32 = fma(a3, b0.s2, acc32);</div><div class="line"><a name="l03733"></a><span class="lineno"> 3733</span>&#160; acc33 = fma(a3, b0.s3, acc33);</div><div class="line"><a name="l03734"></a><span class="lineno"> 3734</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 3</span></div><div class="line"><a name="l03735"></a><span class="lineno"> 3735</span>&#160;</div><div class="line"><a name="l03736"></a><span class="lineno"> 3736</span>&#160; src_addr.s0 += <span class="keyword">sizeof</span>(float);</div><div class="line"><a name="l03737"></a><span class="lineno"> 3737</span>&#160; }</div><div class="line"><a name="l03738"></a><span class="lineno"> 3738</span>&#160;</div><div class="line"><a name="l03739"></a><span class="lineno"> 3739</span>&#160; <span class="keywordtype">int</span> z = get_global_id(2);</div><div class="line"><a name="l03740"></a><span class="lineno"> 3740</span>&#160;</div><div class="line"><a name="l03741"></a><span class="lineno"> 3741</span>&#160; <span class="comment">// Compute destination address</span></div><div class="line"><a name="l03742"></a><span class="lineno"> 3742</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="l03743"></a><span class="lineno"> 3743</span>&#160;</div><div class="line"><a name="l03744"></a><span class="lineno"> 3744</span>&#160; <span class="comment">// Multiply by the weight of matrix-matrix product and store the result</span></div><div class="line"><a name="l03745"></a><span class="lineno"> 3745</span>&#160;<span class="preprocessor">#if defined(ALPHA)</span></div><div class="line"><a name="l03746"></a><span class="lineno"> 3746</span>&#160; acc00 = acc00 * ALPHA;</div><div class="line"><a name="l03747"></a><span class="lineno"> 3747</span>&#160; acc01 = acc01 * ALPHA;</div><div class="line"><a name="l03748"></a><span class="lineno"> 3748</span>&#160; acc02 = acc02 * ALPHA;</div><div class="line"><a name="l03749"></a><span class="lineno"> 3749</span>&#160; acc03 = acc03 * ALPHA;</div><div class="line"><a name="l03750"></a><span class="lineno"> 3750</span>&#160;<span class="preprocessor">#endif // defined(ALPHA)</span></div><div class="line"><a name="l03751"></a><span class="lineno"> 3751</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 1 &amp;&amp; defined(ALPHA)</span></div><div class="line"><a name="l03752"></a><span class="lineno"> 3752</span>&#160; acc10 = acc10 * ALPHA;</div><div class="line"><a name="l03753"></a><span class="lineno"> 3753</span>&#160; acc11 = acc11 * ALPHA;</div><div class="line"><a name="l03754"></a><span class="lineno"> 3754</span>&#160; acc12 = acc12 * ALPHA;</div><div class="line"><a name="l03755"></a><span class="lineno"> 3755</span>&#160; acc13 = acc13 * ALPHA;</div><div class="line"><a name="l03756"></a><span class="lineno"> 3756</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 1 &amp;&amp; defined(ALPHA)</span></div><div class="line"><a name="l03757"></a><span class="lineno"> 3757</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 2 &amp;&amp; defined(ALPHA)</span></div><div class="line"><a name="l03758"></a><span class="lineno"> 3758</span>&#160; acc20 = acc20 * ALPHA;</div><div class="line"><a name="l03759"></a><span class="lineno"> 3759</span>&#160; acc21 = acc21 * ALPHA;</div><div class="line"><a name="l03760"></a><span class="lineno"> 3760</span>&#160; acc22 = acc22 * ALPHA;</div><div class="line"><a name="l03761"></a><span class="lineno"> 3761</span>&#160; acc23 = acc23 * ALPHA;</div><div class="line"><a name="l03762"></a><span class="lineno"> 3762</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 2 &amp;&amp; defined(ALPHA)</span></div><div class="line"><a name="l03763"></a><span class="lineno"> 3763</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 3 &amp;&amp; defined(ALPHA)</span></div><div class="line"><a name="l03764"></a><span class="lineno"> 3764</span>&#160; acc30 = acc30 * ALPHA;</div><div class="line"><a name="l03765"></a><span class="lineno"> 3765</span>&#160; acc31 = acc31 * ALPHA;</div><div class="line"><a name="l03766"></a><span class="lineno"> 3766</span>&#160; acc32 = acc32 * ALPHA;</div><div class="line"><a name="l03767"></a><span class="lineno"> 3767</span>&#160; acc33 = acc33 * ALPHA;</div><div class="line"><a name="l03768"></a><span class="lineno"> 3768</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 3 &amp;&amp; defined(ALPHA)</span></div><div class="line"><a name="l03769"></a><span class="lineno"> 3769</span>&#160;</div><div class="line"><a name="l03770"></a><span class="lineno"> 3770</span>&#160; <span class="comment">// Compute dst address</span></div><div class="line"><a name="l03771"></a><span class="lineno"> 3771</span>&#160; __global uchar *dst_addr = <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>(&amp;<a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#adbf67dcee294e673cf796f1ed8aeb6a4">dst</a>, 0, 0);</div><div class="line"><a name="l03772"></a><span class="lineno"> 3772</span>&#160;</div><div class="line"><a name="l03773"></a><span class="lineno"> 3773</span>&#160;<span class="preprocessor">#if defined(ADD_VEC_C)</span></div><div class="line"><a name="l03774"></a><span class="lineno"> 3774</span>&#160; __global <span class="keywordtype">float</span> *src2_addr = (__global <span class="keywordtype">float</span> *)(src2_ptr + src2_offset_first_element_in_bytes + get_global_id(0) * src2_step_x);</div><div class="line"><a name="l03775"></a><span class="lineno"> 3775</span>&#160; float4 c0 = vload4(0, src2_addr);</div><div class="line"><a name="l03776"></a><span class="lineno"> 3776</span>&#160;</div><div class="line"><a name="l03777"></a><span class="lineno"> 3777</span>&#160; acc00 += c0.s0;</div><div class="line"><a name="l03778"></a><span class="lineno"> 3778</span>&#160; acc01 += c0.s1;</div><div class="line"><a name="l03779"></a><span class="lineno"> 3779</span>&#160; acc02 += c0.s2;</div><div class="line"><a name="l03780"></a><span class="lineno"> 3780</span>&#160; acc03 += c0.s3;</div><div class="line"><a name="l03781"></a><span class="lineno"> 3781</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 1</span></div><div class="line"><a name="l03782"></a><span class="lineno"> 3782</span>&#160; acc10 += c0.s0;</div><div class="line"><a name="l03783"></a><span class="lineno"> 3783</span>&#160; acc11 += c0.s1;</div><div class="line"><a name="l03784"></a><span class="lineno"> 3784</span>&#160; acc12 += c0.s2;</div><div class="line"><a name="l03785"></a><span class="lineno"> 3785</span>&#160; acc13 += c0.s3;</div><div class="line"><a name="l03786"></a><span class="lineno"> 3786</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 1</span></div><div class="line"><a name="l03787"></a><span class="lineno"> 3787</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 2</span></div><div class="line"><a name="l03788"></a><span class="lineno"> 3788</span>&#160; acc20 += c0.s0;</div><div class="line"><a name="l03789"></a><span class="lineno"> 3789</span>&#160; acc21 += c0.s1;</div><div class="line"><a name="l03790"></a><span class="lineno"> 3790</span>&#160; acc22 += c0.s2;</div><div class="line"><a name="l03791"></a><span class="lineno"> 3791</span>&#160; acc23 += c0.s3;</div><div class="line"><a name="l03792"></a><span class="lineno"> 3792</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 2</span></div><div class="line"><a name="l03793"></a><span class="lineno"> 3793</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 3</span></div><div class="line"><a name="l03794"></a><span class="lineno"> 3794</span>&#160; acc30 += c0.s0;</div><div class="line"><a name="l03795"></a><span class="lineno"> 3795</span>&#160; acc31 += c0.s1;</div><div class="line"><a name="l03796"></a><span class="lineno"> 3796</span>&#160; acc32 += c0.s2;</div><div class="line"><a name="l03797"></a><span class="lineno"> 3797</span>&#160; acc33 += c0.s3;</div><div class="line"><a name="l03798"></a><span class="lineno"> 3798</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 3</span></div><div class="line"><a name="l03799"></a><span class="lineno"> 3799</span>&#160;<span class="preprocessor">#endif </span><span class="comment">/* defined(ADD_VEC_C) */</span><span class="preprocessor"></span></div><div class="line"><a name="l03800"></a><span class="lineno"> 3800</span>&#160;</div><div class="line"><a name="l03801"></a><span class="lineno"> 3801</span>&#160;<span class="preprocessor">#if defined(REINTERPRET_OUTPUT_AS_3D)</span></div><div class="line"><a name="l03802"></a><span class="lineno"> 3802</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="l03803"></a><span class="lineno"> 3803</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="l03804"></a><span class="lineno"> 3804</span>&#160; <span class="comment">//</span></div><div class="line"><a name="l03805"></a><span class="lineno"> 3805</span>&#160; <span class="comment">// | |</span></div><div class="line"><a name="l03806"></a><span class="lineno"> 3806</span>&#160; <span class="comment">// | plane0 |</span></div><div class="line"><a name="l03807"></a><span class="lineno"> 3807</span>&#160; <span class="comment">// | |</span></div><div class="line"><a name="l03808"></a><span class="lineno"> 3808</span>&#160; <span class="comment">// |__________________|</span></div><div class="line"><a name="l03809"></a><span class="lineno"> 3809</span>&#160; <span class="comment">// |******************|</span></div><div class="line"><a name="l03810"></a><span class="lineno"> 3810</span>&#160; <span class="comment">// | cross_plane_pad |</span></div><div class="line"><a name="l03811"></a><span class="lineno"> 3811</span>&#160; <span class="comment">// |******************|</span></div><div class="line"><a name="l03812"></a><span class="lineno"> 3812</span>&#160; <span class="comment">// | |</span></div><div class="line"><a name="l03813"></a><span class="lineno"> 3813</span>&#160; <span class="comment">// | plane1 |</span></div><div class="line"><a name="l03814"></a><span class="lineno"> 3814</span>&#160; <span class="comment">// | |</span></div><div class="line"><a name="l03815"></a><span class="lineno"> 3815</span>&#160; <span class="comment">// |__________________|</span></div><div class="line"><a name="l03816"></a><span class="lineno"> 3816</span>&#160;</div><div class="line"><a name="l03817"></a><span class="lineno"> 3817</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="l03818"></a><span class="lineno"> 3818</span>&#160; uint4 zout = ((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="l03819"></a><span class="lineno"> 3819</span>&#160; zout = min(DEPTH_GEMM3D - 1, zout);</div><div class="line"><a name="l03820"></a><span class="lineno"> 3820</span>&#160;</div><div class="line"><a name="l03821"></a><span class="lineno"> 3821</span>&#160; <span class="comment">// Add offset due to the cross plane paddings</span></div><div class="line"><a name="l03822"></a><span class="lineno"> 3822</span>&#160; zout *= (dst_cross_plane_pad * dst_stride_y);</div><div class="line"><a name="l03823"></a><span class="lineno"> 3823</span>&#160;</div><div class="line"><a name="l03824"></a><span class="lineno"> 3824</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="l03825"></a><span class="lineno"> 3825</span>&#160; <span class="comment">// multiply dst_stride_z by DEPTH_GEMM3D</span></div><div class="line"><a name="l03826"></a><span class="lineno"> 3826</span>&#160; dst_addr += z * dst_stride_z * DEPTH_GEMM3D;</div><div class="line"><a name="l03827"></a><span class="lineno"> 3827</span>&#160;</div><div class="line"><a name="l03828"></a><span class="lineno"> 3828</span>&#160; <span class="comment">// Store the output block</span></div><div class="line"><a name="l03829"></a><span class="lineno"> 3829</span>&#160; vstore4((float4)(acc00, acc01, acc02, acc03), 0, (__global <span class="keywordtype">float</span> *)(dst_addr + 0 * dst_stride_y + zout.s0));</div><div class="line"><a name="l03830"></a><span class="lineno"> 3830</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 1</span></div><div class="line"><a name="l03831"></a><span class="lineno"> 3831</span>&#160; vstore4((float4)(acc10, acc11, acc12, acc13), 0, (__global <span class="keywordtype">float</span> *)(dst_addr + 1 * dst_stride_y + zout.s1));</div><div class="line"><a name="l03832"></a><span class="lineno"> 3832</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 1</span></div><div class="line"><a name="l03833"></a><span class="lineno"> 3833</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 2</span></div><div class="line"><a name="l03834"></a><span class="lineno"> 3834</span>&#160; vstore4((float4)(acc20, acc21, acc22, acc23), 0, (__global <span class="keywordtype">float</span> *)(dst_addr + 2 * dst_stride_y + zout.s2));</div><div class="line"><a name="l03835"></a><span class="lineno"> 3835</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 2</span></div><div class="line"><a name="l03836"></a><span class="lineno"> 3836</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 3</span></div><div class="line"><a name="l03837"></a><span class="lineno"> 3837</span>&#160; vstore4((float4)(acc30, acc31, acc32, acc33), 0, (__global <span class="keywordtype">float</span> *)(dst_addr + 3 * dst_stride_y + zout.s3));</div><div class="line"><a name="l03838"></a><span class="lineno"> 3838</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 3</span></div><div class="line"><a name="l03839"></a><span class="lineno"> 3839</span>&#160;</div><div class="line"><a name="l03840"></a><span class="lineno"> 3840</span>&#160;<span class="preprocessor">#else // defined(REINTERPRET_OUTPUT_AS_3D)</span></div><div class="line"><a name="l03841"></a><span class="lineno"> 3841</span>&#160; <span class="comment">// Add offset for batched GEMM</span></div><div class="line"><a name="l03842"></a><span class="lineno"> 3842</span>&#160; dst_addr += z * dst_stride_z;</div><div class="line"><a name="l03843"></a><span class="lineno"> 3843</span>&#160;</div><div class="line"><a name="l03844"></a><span class="lineno"> 3844</span>&#160; <span class="comment">// Store the output block</span></div><div class="line"><a name="l03845"></a><span class="lineno"> 3845</span>&#160; vstore4((float4)(acc00, acc01, acc02, acc03), 0, (__global <span class="keywordtype">float</span> *)(dst_addr + 0 * dst_stride_y));</div><div class="line"><a name="l03846"></a><span class="lineno"> 3846</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 1</span></div><div class="line"><a name="l03847"></a><span class="lineno"> 3847</span>&#160; vstore4((float4)(acc10, acc11, acc12, acc13), 0, (__global <span class="keywordtype">float</span> *)(dst_addr + 1 * dst_stride_y));</div><div class="line"><a name="l03848"></a><span class="lineno"> 3848</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 1</span></div><div class="line"><a name="l03849"></a><span class="lineno"> 3849</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 2</span></div><div class="line"><a name="l03850"></a><span class="lineno"> 3850</span>&#160; vstore4((float4)(acc20, acc21, acc22, acc23), 0, (__global <span class="keywordtype">float</span> *)(dst_addr + 2 * dst_stride_y));</div><div class="line"><a name="l03851"></a><span class="lineno"> 3851</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 2</span></div><div class="line"><a name="l03852"></a><span class="lineno"> 3852</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 3</span></div><div class="line"><a name="l03853"></a><span class="lineno"> 3853</span>&#160; vstore4((float4)(acc30, acc31, acc32, acc33), 0, (__global <span class="keywordtype">float</span> *)(dst_addr + 3 * dst_stride_y));</div><div class="line"><a name="l03854"></a><span class="lineno"> 3854</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 3</span></div><div class="line"><a name="l03855"></a><span class="lineno"> 3855</span>&#160;<span class="preprocessor">#endif // defined(REINTERPRET_OUTPUT_AS_3D)</span></div><div class="line"><a name="l03856"></a><span class="lineno"> 3856</span>&#160;}</div><div class="line"><a name="l03857"></a><span class="lineno"> 3857</span>&#160;</div><div class="line"><a name="l03908"></a><span class="lineno"> 3908</span>&#160;__kernel <span class="keywordtype">void</span> gemm_mm_floating_point_f32_bifrost_1000(<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="l03909"></a><span class="lineno"> 3909</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="l03910"></a><span class="lineno"> 3910</span>&#160;#<span class="keywordflow">if</span> defined(ADD_VEC_C)</div><div class="line"><a name="l03911"></a><span class="lineno"> 3911</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a40a6eb9f2a7712f08d6bb8ff6c9e6ca7">VECTOR_DECLARATION</a>(src2),</div><div class="line"><a name="l03912"></a><span class="lineno"> 3912</span>&#160;#endif <span class="comment">/* defined(ADD_VEC_C) */</span></div><div class="line"><a name="l03913"></a><span class="lineno"> 3913</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="l03914"></a><span class="lineno"> 3914</span>&#160; uint src0_stride_z,</div><div class="line"><a name="l03915"></a><span class="lineno"> 3915</span>&#160; uint src1_stride_z,</div><div class="line"><a name="l03916"></a><span class="lineno"> 3916</span>&#160; uint dst_stride_z</div><div class="line"><a name="l03917"></a><span class="lineno"> 3917</span>&#160;#<span class="keywordflow">if</span> defined(REINTERPRET_INPUT_AS_3D)</div><div class="line"><a name="l03918"></a><span class="lineno"> 3918</span>&#160; ,</div><div class="line"><a name="l03919"></a><span class="lineno"> 3919</span>&#160; uint src_cross_plane_pad</div><div class="line"><a name="l03920"></a><span class="lineno"> 3920</span>&#160;#endif <span class="comment">// REINTERPRET_INPUT_AS_3D</span></div><div class="line"><a name="l03921"></a><span class="lineno"> 3921</span>&#160;#<span class="keywordflow">if</span> defined(REINTERPRET_OUTPUT_AS_3D)</div><div class="line"><a name="l03922"></a><span class="lineno"> 3922</span>&#160; ,</div><div class="line"><a name="l03923"></a><span class="lineno"> 3923</span>&#160; uint dst_cross_plane_pad</div><div class="line"><a name="l03924"></a><span class="lineno"> 3924</span>&#160;#endif <span class="comment">// REINTERPRET_OUTPUT_AS_3D</span></div><div class="line"><a name="l03925"></a><span class="lineno"> 3925</span>&#160; )</div><div class="line"><a name="l03926"></a><span class="lineno"> 3926</span>&#160;{</div><div class="line"><a name="l03927"></a><span class="lineno"> 3927</span>&#160; <span class="comment">// Requires 2 NUM_ELEMS_PROCESSED_PER_THREAD_X, C vect2, A vect4, B (2 vload2) // to fix for NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 1</span></div><div class="line"><a name="l03928"></a><span class="lineno"> 3928</span>&#160; <span class="keywordtype">int</span> idx = get_global_id(0) * NUM_ELEMS_PROCESSED_PER_THREAD_X;</div><div class="line"><a name="l03929"></a><span class="lineno"> 3929</span>&#160;</div><div class="line"><a name="l03930"></a><span class="lineno"> 3930</span>&#160; <span class="comment">// Compute starting address for matrix A and Matrix B</span></div><div class="line"><a name="l03931"></a><span class="lineno"> 3931</span>&#160; int2 src_addr = ((int2)(src0_offset_first_element_in_bytes, src1_offset_first_element_in_bytes));</div><div class="line"><a name="l03932"></a><span class="lineno"> 3932</span>&#160;</div><div class="line"><a name="l03933"></a><span class="lineno"> 3933</span>&#160; <span class="comment">// Update address for the matrix A</span></div><div class="line"><a name="l03934"></a><span class="lineno"> 3934</span>&#160; src_addr.s0 += get_global_id(1) * src0_stride_y * NUM_ELEMS_PROCESSED_PER_THREAD_Y;</div><div class="line"><a name="l03935"></a><span class="lineno"> 3935</span>&#160;</div><div class="line"><a name="l03936"></a><span class="lineno"> 3936</span>&#160; <span class="comment">// Update address for the matrix B</span></div><div class="line"><a name="l03937"></a><span class="lineno"> 3937</span>&#160; src_addr.s1 += idx * <span class="keyword">sizeof</span>(float);</div><div class="line"><a name="l03938"></a><span class="lineno"> 3938</span>&#160;</div><div class="line"><a name="l03939"></a><span class="lineno"> 3939</span>&#160;<span class="preprocessor">#if defined(REINTERPRET_INPUT_AS_3D)</span></div><div class="line"><a name="l03940"></a><span class="lineno"> 3940</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="l03941"></a><span class="lineno"> 3941</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="l03942"></a><span class="lineno"> 3942</span>&#160; <span class="comment">//</span></div><div class="line"><a name="l03943"></a><span class="lineno"> 3943</span>&#160; <span class="comment">// | |</span></div><div class="line"><a name="l03944"></a><span class="lineno"> 3944</span>&#160; <span class="comment">// | plane0 |</span></div><div class="line"><a name="l03945"></a><span class="lineno"> 3945</span>&#160; <span class="comment">// | |</span></div><div class="line"><a name="l03946"></a><span class="lineno"> 3946</span>&#160; <span class="comment">// |__________________|</span></div><div class="line"><a name="l03947"></a><span class="lineno"> 3947</span>&#160; <span class="comment">// |******************|</span></div><div class="line"><a name="l03948"></a><span class="lineno"> 3948</span>&#160; <span class="comment">// | cross_plane_pad |</span></div><div class="line"><a name="l03949"></a><span class="lineno"> 3949</span>&#160; <span class="comment">// |******************|</span></div><div class="line"><a name="l03950"></a><span class="lineno"> 3950</span>&#160; <span class="comment">// | |</span></div><div class="line"><a name="l03951"></a><span class="lineno"> 3951</span>&#160; <span class="comment">// | plane1 |</span></div><div class="line"><a name="l03952"></a><span class="lineno"> 3952</span>&#160; <span class="comment">// | |</span></div><div class="line"><a name="l03953"></a><span class="lineno"> 3953</span>&#160; <span class="comment">// |__________________|</span></div><div class="line"><a name="l03954"></a><span class="lineno"> 3954</span>&#160;</div><div class="line"><a name="l03955"></a><span class="lineno"> 3955</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="l03956"></a><span class="lineno"> 3956</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="l03957"></a><span class="lineno"> 3957</span>&#160; zin = min(DEPTH_GEMM3D - 1, zin);</div><div class="line"><a name="l03958"></a><span class="lineno"> 3958</span>&#160;</div><div class="line"><a name="l03959"></a><span class="lineno"> 3959</span>&#160; <span class="comment">// Add offset due to the cross plane paddings</span></div><div class="line"><a name="l03960"></a><span class="lineno"> 3960</span>&#160; zin *= (src_cross_plane_pad * src0_stride_y);</div><div class="line"><a name="l03961"></a><span class="lineno"> 3961</span>&#160;</div><div class="line"><a name="l03962"></a><span class="lineno"> 3962</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="l03963"></a><span class="lineno"> 3963</span>&#160; <span class="comment">// multiply src0_stride_z by DEPTH_GEMM3D</span></div><div class="line"><a name="l03964"></a><span class="lineno"> 3964</span>&#160; src_addr.s0 += get_global_id(2) * src0_stride_z * DEPTH_GEMM3D;</div><div class="line"><a name="l03965"></a><span class="lineno"> 3965</span>&#160;</div><div class="line"><a name="l03966"></a><span class="lineno"> 3966</span>&#160;<span class="preprocessor">#else // defined(REINTERPRET_INPUT_AS_3D)</span></div><div class="line"><a name="l03967"></a><span class="lineno"> 3967</span>&#160;</div><div class="line"><a name="l03968"></a><span class="lineno"> 3968</span>&#160; <span class="comment">// Add offset for batched GEMM</span></div><div class="line"><a name="l03969"></a><span class="lineno"> 3969</span>&#160; src_addr.s0 += get_global_id(2) * src0_stride_z;</div><div class="line"><a name="l03970"></a><span class="lineno"> 3970</span>&#160;</div><div class="line"><a name="l03971"></a><span class="lineno"> 3971</span>&#160;<span class="preprocessor">#endif // defined(REINTERPRET_INPUT_AS_3D)</span></div><div class="line"><a name="l03972"></a><span class="lineno"> 3972</span>&#160;</div><div class="line"><a name="l03973"></a><span class="lineno"> 3973</span>&#160;<span class="preprocessor">#if defined(MATRIX_B_DEPTH)</span></div><div class="line"><a name="l03974"></a><span class="lineno"> 3974</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="l03975"></a><span class="lineno"> 3975</span>&#160; src_addr.s1 += (get_global_id(2) % MATRIX_B_DEPTH) * src1_stride_z;</div><div class="line"><a name="l03976"></a><span class="lineno"> 3976</span>&#160;<span class="preprocessor">#else // defined(MATRIX_B_DEPTH)</span></div><div class="line"><a name="l03977"></a><span class="lineno"> 3977</span>&#160; src_addr.s1 += get_global_id(2) * src1_stride_z;</div><div class="line"><a name="l03978"></a><span class="lineno"> 3978</span>&#160;<span class="preprocessor">#endif // defined(MATRIX_B_DEPTH)</span></div><div class="line"><a name="l03979"></a><span class="lineno"> 3979</span>&#160;</div><div class="line"><a name="l03980"></a><span class="lineno"> 3980</span>&#160; <span class="comment">// Initialize accumulators</span></div><div class="line"><a name="l03981"></a><span class="lineno"> 3981</span>&#160; <span class="keywordtype">float</span> acc00 = 0.0f;</div><div class="line"><a name="l03982"></a><span class="lineno"> 3982</span>&#160; <span class="keywordtype">float</span> acc01 = 0.0f;</div><div class="line"><a name="l03983"></a><span class="lineno"> 3983</span>&#160;</div><div class="line"><a name="l03984"></a><span class="lineno"> 3984</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 1</span></div><div class="line"><a name="l03985"></a><span class="lineno"> 3985</span>&#160; <span class="keywordtype">float</span> acc10 = 0.0f;</div><div class="line"><a name="l03986"></a><span class="lineno"> 3986</span>&#160; <span class="keywordtype">float</span> acc11 = 0.0f;</div><div class="line"><a name="l03987"></a><span class="lineno"> 3987</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 1</span></div><div class="line"><a name="l03988"></a><span class="lineno"> 3988</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 2</span></div><div class="line"><a name="l03989"></a><span class="lineno"> 3989</span>&#160; <span class="keywordtype">float</span> acc20 = 0.0f;</div><div class="line"><a name="l03990"></a><span class="lineno"> 3990</span>&#160; <span class="keywordtype">float</span> acc21 = 0.0f;</div><div class="line"><a name="l03991"></a><span class="lineno"> 3991</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 2</span></div><div class="line"><a name="l03992"></a><span class="lineno"> 3992</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 3</span></div><div class="line"><a name="l03993"></a><span class="lineno"> 3993</span>&#160; <span class="keywordtype">float</span> acc30 = 0.0f;</div><div class="line"><a name="l03994"></a><span class="lineno"> 3994</span>&#160; <span class="keywordtype">float</span> acc31 = 0.0f;</div><div class="line"><a name="l03995"></a><span class="lineno"> 3995</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 3</span></div><div class="line"><a name="l03996"></a><span class="lineno"> 3996</span>&#160;</div><div class="line"><a name="l03997"></a><span class="lineno"> 3997</span>&#160; <span class="comment">// A and B src indices get incremented at the same time.</span></div><div class="line"><a name="l03998"></a><span class="lineno"> 3998</span>&#160; <span class="keywordtype">int</span> i = 0;</div><div class="line"><a name="l03999"></a><span class="lineno"> 3999</span>&#160; <span class="keywordflow">for</span>(; i &lt;= ((int)COLS_A - 8); i += 8)</div><div class="line"><a name="l04000"></a><span class="lineno"> 4000</span>&#160; {</div><div class="line"><a name="l04001"></a><span class="lineno"> 4001</span>&#160;<span class="preprocessor">#if defined(REINTERPRET_INPUT_AS_3D)</span></div><div class="line"><a name="l04002"></a><span class="lineno"> 4002</span>&#160; <span class="comment">// Load values from matrix A</span></div><div class="line"><a name="l04003"></a><span class="lineno"> 4003</span>&#160; float8 a0 = vload8(0, (__global <span class="keywordtype">float</span> *)(src0_ptr + src_addr.s0 + zin.s0));</div><div class="line"><a name="l04004"></a><span class="lineno"> 4004</span>&#160;<span class="preprocessor">#else // defined(REINTERPRET_INPUT_AS_3D)</span></div><div class="line"><a name="l04005"></a><span class="lineno"> 4005</span>&#160; <span class="comment">// Load values from matrix A</span></div><div class="line"><a name="l04006"></a><span class="lineno"> 4006</span>&#160; float8 a0 = vload8(0, (__global <span class="keywordtype">float</span> *)(src0_ptr + src_addr.s0));</div><div class="line"><a name="l04007"></a><span class="lineno"> 4007</span>&#160;<span class="preprocessor">#endif // defined(REINTERPRET_INPUT_AS_3D)</span></div><div class="line"><a name="l04008"></a><span class="lineno"> 4008</span>&#160;</div><div class="line"><a name="l04009"></a><span class="lineno"> 4009</span>&#160; <span class="comment">// Load values from matrix B</span></div><div class="line"><a name="l04010"></a><span class="lineno"> 4010</span>&#160; float2 b0 = vload2(0, (__global <span class="keywordtype">float</span> *)(src1_ptr + src_addr.s1));</div><div class="line"><a name="l04011"></a><span class="lineno"> 4011</span>&#160; src_addr.s1 += src1_stride_y;</div><div class="line"><a name="l04012"></a><span class="lineno"> 4012</span>&#160; float2 b1 = vload2(0, (__global <span class="keywordtype">float</span> *)(src1_ptr + src_addr.s1));</div><div class="line"><a name="l04013"></a><span class="lineno"> 4013</span>&#160; src_addr.s1 += src1_stride_y;</div><div class="line"><a name="l04014"></a><span class="lineno"> 4014</span>&#160; float2 b2 = vload2(0, (__global <span class="keywordtype">float</span> *)(src1_ptr + src_addr.s1));</div><div class="line"><a name="l04015"></a><span class="lineno"> 4015</span>&#160; src_addr.s1 += src1_stride_y;</div><div class="line"><a name="l04016"></a><span class="lineno"> 4016</span>&#160; float2 b3 = vload2(0, (__global <span class="keywordtype">float</span> *)(src1_ptr + src_addr.s1));</div><div class="line"><a name="l04017"></a><span class="lineno"> 4017</span>&#160; src_addr.s1 += src1_stride_y;</div><div class="line"><a name="l04018"></a><span class="lineno"> 4018</span>&#160; float2 b4 = vload2(0, (__global <span class="keywordtype">float</span> *)(src1_ptr + src_addr.s1));</div><div class="line"><a name="l04019"></a><span class="lineno"> 4019</span>&#160; src_addr.s1 += src1_stride_y;</div><div class="line"><a name="l04020"></a><span class="lineno"> 4020</span>&#160; float2 b5 = vload2(0, (__global <span class="keywordtype">float</span> *)(src1_ptr + src_addr.s1));</div><div class="line"><a name="l04021"></a><span class="lineno"> 4021</span>&#160; src_addr.s1 += src1_stride_y;</div><div class="line"><a name="l04022"></a><span class="lineno"> 4022</span>&#160; float2 b6 = vload2(0, (__global <span class="keywordtype">float</span> *)(src1_ptr + src_addr.s1));</div><div class="line"><a name="l04023"></a><span class="lineno"> 4023</span>&#160; src_addr.s1 += src1_stride_y;</div><div class="line"><a name="l04024"></a><span class="lineno"> 4024</span>&#160; float2 b7 = vload2(0, (__global <span class="keywordtype">float</span> *)(src1_ptr + src_addr.s1));</div><div class="line"><a name="l04025"></a><span class="lineno"> 4025</span>&#160; src_addr.s1 += src1_stride_y;</div><div class="line"><a name="l04026"></a><span class="lineno"> 4026</span>&#160;</div><div class="line"><a name="l04027"></a><span class="lineno"> 4027</span>&#160; <span class="comment">// Multiply and accumulate</span></div><div class="line"><a name="l04028"></a><span class="lineno"> 4028</span>&#160; acc00 = fma(a0.s0, b0.s0, acc00);</div><div class="line"><a name="l04029"></a><span class="lineno"> 4029</span>&#160; acc00 = fma(a0.s1, b1.s0, acc00);</div><div class="line"><a name="l04030"></a><span class="lineno"> 4030</span>&#160; acc00 = fma(a0.s2, b2.s0, acc00);</div><div class="line"><a name="l04031"></a><span class="lineno"> 4031</span>&#160; acc00 = fma(a0.s3, b3.s0, acc00);</div><div class="line"><a name="l04032"></a><span class="lineno"> 4032</span>&#160; acc00 = fma(a0.s4, b4.s0, acc00);</div><div class="line"><a name="l04033"></a><span class="lineno"> 4033</span>&#160; acc00 = fma(a0.s5, b5.s0, acc00);</div><div class="line"><a name="l04034"></a><span class="lineno"> 4034</span>&#160; acc00 = fma(a0.s6, b6.s0, acc00);</div><div class="line"><a name="l04035"></a><span class="lineno"> 4035</span>&#160; acc00 = fma(a0.s7, b7.s0, acc00);</div><div class="line"><a name="l04036"></a><span class="lineno"> 4036</span>&#160;</div><div class="line"><a name="l04037"></a><span class="lineno"> 4037</span>&#160; acc01 = fma(a0.s0, b0.s1, acc01);</div><div class="line"><a name="l04038"></a><span class="lineno"> 4038</span>&#160; acc01 = fma(a0.s1, b1.s1, acc01);</div><div class="line"><a name="l04039"></a><span class="lineno"> 4039</span>&#160; acc01 = fma(a0.s2, b2.s1, acc01);</div><div class="line"><a name="l04040"></a><span class="lineno"> 4040</span>&#160; acc01 = fma(a0.s3, b3.s1, acc01);</div><div class="line"><a name="l04041"></a><span class="lineno"> 4041</span>&#160; acc01 = fma(a0.s4, b4.s1, acc01);</div><div class="line"><a name="l04042"></a><span class="lineno"> 4042</span>&#160; acc01 = fma(a0.s5, b5.s1, acc01);</div><div class="line"><a name="l04043"></a><span class="lineno"> 4043</span>&#160; acc01 = fma(a0.s6, b6.s1, acc01);</div><div class="line"><a name="l04044"></a><span class="lineno"> 4044</span>&#160; acc01 = fma(a0.s7, b7.s1, acc01);</div><div class="line"><a name="l04045"></a><span class="lineno"> 4045</span>&#160;</div><div class="line"><a name="l04046"></a><span class="lineno"> 4046</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 1</span></div><div class="line"><a name="l04047"></a><span class="lineno"> 4047</span>&#160;<span class="preprocessor">#if defined(REINTERPRET_INPUT_AS_3D)</span></div><div class="line"><a name="l04048"></a><span class="lineno"> 4048</span>&#160; a0 = vload8(0, (__global <span class="keywordtype">float</span> *)(src0_ptr + src_addr.s0 + 1 * src0_stride_y + zin.s1));</div><div class="line"><a name="l04049"></a><span class="lineno"> 4049</span>&#160;<span class="preprocessor">#else // defined(REINTERPRET_INPUT_AS_3D)</span></div><div class="line"><a name="l04050"></a><span class="lineno"> 4050</span>&#160; a0 = vload8(0, (__global <span class="keywordtype">float</span> *)(src0_ptr + src_addr.s0 + 1 * src0_stride_y));</div><div class="line"><a name="l04051"></a><span class="lineno"> 4051</span>&#160;<span class="preprocessor">#endif // defined(REINTERPRET_INPUT_AS_3D)</span></div><div class="line"><a name="l04052"></a><span class="lineno"> 4052</span>&#160; acc10 = fma(a0.s0, b0.s0, acc10);</div><div class="line"><a name="l04053"></a><span class="lineno"> 4053</span>&#160; acc10 = fma(a0.s1, b1.s0, acc10);</div><div class="line"><a name="l04054"></a><span class="lineno"> 4054</span>&#160; acc10 = fma(a0.s2, b2.s0, acc10);</div><div class="line"><a name="l04055"></a><span class="lineno"> 4055</span>&#160; acc10 = fma(a0.s3, b3.s0, acc10);</div><div class="line"><a name="l04056"></a><span class="lineno"> 4056</span>&#160; acc10 = fma(a0.s4, b4.s0, acc10);</div><div class="line"><a name="l04057"></a><span class="lineno"> 4057</span>&#160; acc10 = fma(a0.s5, b5.s0, acc10);</div><div class="line"><a name="l04058"></a><span class="lineno"> 4058</span>&#160; acc10 = fma(a0.s6, b6.s0, acc10);</div><div class="line"><a name="l04059"></a><span class="lineno"> 4059</span>&#160; acc10 = fma(a0.s7, b7.s0, acc10);</div><div class="line"><a name="l04060"></a><span class="lineno"> 4060</span>&#160;</div><div class="line"><a name="l04061"></a><span class="lineno"> 4061</span>&#160; acc11 = fma(a0.s0, b0.s1, acc11);</div><div class="line"><a name="l04062"></a><span class="lineno"> 4062</span>&#160; acc11 = fma(a0.s1, b1.s1, acc11);</div><div class="line"><a name="l04063"></a><span class="lineno"> 4063</span>&#160; acc11 = fma(a0.s2, b2.s1, acc11);</div><div class="line"><a name="l04064"></a><span class="lineno"> 4064</span>&#160; acc11 = fma(a0.s3, b3.s1, acc11);</div><div class="line"><a name="l04065"></a><span class="lineno"> 4065</span>&#160; acc11 = fma(a0.s4, b4.s1, acc11);</div><div class="line"><a name="l04066"></a><span class="lineno"> 4066</span>&#160; acc11 = fma(a0.s5, b5.s1, acc11);</div><div class="line"><a name="l04067"></a><span class="lineno"> 4067</span>&#160; acc11 = fma(a0.s6, b6.s1, acc11);</div><div class="line"><a name="l04068"></a><span class="lineno"> 4068</span>&#160; acc11 = fma(a0.s7, b7.s1, acc11);</div><div class="line"><a name="l04069"></a><span class="lineno"> 4069</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 1</span></div><div class="line"><a name="l04070"></a><span class="lineno"> 4070</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 2</span></div><div class="line"><a name="l04071"></a><span class="lineno"> 4071</span>&#160;<span class="preprocessor">#if defined(REINTERPRET_INPUT_AS_3D)</span></div><div class="line"><a name="l04072"></a><span class="lineno"> 4072</span>&#160; a0 = vload8(0, (__global <span class="keywordtype">float</span> *)(src0_ptr + src_addr.s0 + 2 * src0_stride_y + zin.s2));</div><div class="line"><a name="l04073"></a><span class="lineno"> 4073</span>&#160;<span class="preprocessor">#else // defined(REINTERPRET_INPUT_AS_3D)</span></div><div class="line"><a name="l04074"></a><span class="lineno"> 4074</span>&#160; a0 = vload8(0, (__global <span class="keywordtype">float</span> *)(src0_ptr + src_addr.s0 + 2 * src0_stride_y));</div><div class="line"><a name="l04075"></a><span class="lineno"> 4075</span>&#160;<span class="preprocessor">#endif // defined(REINTERPRET_INPUT_AS_3D)</span></div><div class="line"><a name="l04076"></a><span class="lineno"> 4076</span>&#160; acc20 = fma(a0.s0, b0.s0, acc20);</div><div class="line"><a name="l04077"></a><span class="lineno"> 4077</span>&#160; acc20 = fma(a0.s1, b1.s0, acc20);</div><div class="line"><a name="l04078"></a><span class="lineno"> 4078</span>&#160; acc20 = fma(a0.s2, b2.s0, acc20);</div><div class="line"><a name="l04079"></a><span class="lineno"> 4079</span>&#160; acc20 = fma(a0.s3, b3.s0, acc20);</div><div class="line"><a name="l04080"></a><span class="lineno"> 4080</span>&#160; acc20 = fma(a0.s4, b4.s0, acc20);</div><div class="line"><a name="l04081"></a><span class="lineno"> 4081</span>&#160; acc20 = fma(a0.s5, b5.s0, acc20);</div><div class="line"><a name="l04082"></a><span class="lineno"> 4082</span>&#160; acc20 = fma(a0.s6, b6.s0, acc20);</div><div class="line"><a name="l04083"></a><span class="lineno"> 4083</span>&#160; acc20 = fma(a0.s7, b7.s0, acc20);</div><div class="line"><a name="l04084"></a><span class="lineno"> 4084</span>&#160;</div><div class="line"><a name="l04085"></a><span class="lineno"> 4085</span>&#160; acc21 = fma(a0.s0, b0.s1, acc21);</div><div class="line"><a name="l04086"></a><span class="lineno"> 4086</span>&#160; acc21 = fma(a0.s1, b1.s1, acc21);</div><div class="line"><a name="l04087"></a><span class="lineno"> 4087</span>&#160; acc21 = fma(a0.s2, b2.s1, acc21);</div><div class="line"><a name="l04088"></a><span class="lineno"> 4088</span>&#160; acc21 = fma(a0.s3, b3.s1, acc21);</div><div class="line"><a name="l04089"></a><span class="lineno"> 4089</span>&#160; acc21 = fma(a0.s4, b4.s1, acc21);</div><div class="line"><a name="l04090"></a><span class="lineno"> 4090</span>&#160; acc21 = fma(a0.s5, b5.s1, acc21);</div><div class="line"><a name="l04091"></a><span class="lineno"> 4091</span>&#160; acc21 = fma(a0.s6, b6.s1, acc21);</div><div class="line"><a name="l04092"></a><span class="lineno"> 4092</span>&#160; acc21 = fma(a0.s7, b7.s1, acc21);</div><div class="line"><a name="l04093"></a><span class="lineno"> 4093</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 2</span></div><div class="line"><a name="l04094"></a><span class="lineno"> 4094</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 3</span></div><div class="line"><a name="l04095"></a><span class="lineno"> 4095</span>&#160;<span class="preprocessor">#if defined(REINTERPRET_INPUT_AS_3D)</span></div><div class="line"><a name="l04096"></a><span class="lineno"> 4096</span>&#160; a0 = vload8(0, (__global <span class="keywordtype">float</span> *)(src0_ptr + src_addr.s0 + 3 * src0_stride_y + zin.s3));</div><div class="line"><a name="l04097"></a><span class="lineno"> 4097</span>&#160;<span class="preprocessor">#else // defined(REINTERPRET_INPUT_AS_3D)</span></div><div class="line"><a name="l04098"></a><span class="lineno"> 4098</span>&#160; a0 = vload8(0, (__global <span class="keywordtype">float</span> *)(src0_ptr + src_addr.s0 + 3 * src0_stride_y));</div><div class="line"><a name="l04099"></a><span class="lineno"> 4099</span>&#160;<span class="preprocessor">#endif // defined(REINTERPRET_INPUT_AS_3D)</span></div><div class="line"><a name="l04100"></a><span class="lineno"> 4100</span>&#160; acc30 = fma(a0.s0, b0.s0, acc30);</div><div class="line"><a name="l04101"></a><span class="lineno"> 4101</span>&#160; acc30 = fma(a0.s1, b1.s0, acc30);</div><div class="line"><a name="l04102"></a><span class="lineno"> 4102</span>&#160; acc30 = fma(a0.s2, b2.s0, acc30);</div><div class="line"><a name="l04103"></a><span class="lineno"> 4103</span>&#160; acc30 = fma(a0.s3, b3.s0, acc30);</div><div class="line"><a name="l04104"></a><span class="lineno"> 4104</span>&#160; acc30 = fma(a0.s4, b4.s0, acc30);</div><div class="line"><a name="l04105"></a><span class="lineno"> 4105</span>&#160; acc30 = fma(a0.s5, b5.s0, acc30);</div><div class="line"><a name="l04106"></a><span class="lineno"> 4106</span>&#160; acc30 = fma(a0.s6, b6.s0, acc30);</div><div class="line"><a name="l04107"></a><span class="lineno"> 4107</span>&#160; acc30 = fma(a0.s7, b7.s0, acc30);</div><div class="line"><a name="l04108"></a><span class="lineno"> 4108</span>&#160;</div><div class="line"><a name="l04109"></a><span class="lineno"> 4109</span>&#160; acc31 = fma(a0.s0, b0.s1, acc31);</div><div class="line"><a name="l04110"></a><span class="lineno"> 4110</span>&#160; acc31 = fma(a0.s1, b1.s1, acc31);</div><div class="line"><a name="l04111"></a><span class="lineno"> 4111</span>&#160; acc31 = fma(a0.s2, b2.s1, acc31);</div><div class="line"><a name="l04112"></a><span class="lineno"> 4112</span>&#160; acc31 = fma(a0.s3, b3.s1, acc31);</div><div class="line"><a name="l04113"></a><span class="lineno"> 4113</span>&#160; acc31 = fma(a0.s4, b4.s1, acc31);</div><div class="line"><a name="l04114"></a><span class="lineno"> 4114</span>&#160; acc31 = fma(a0.s5, b5.s1, acc31);</div><div class="line"><a name="l04115"></a><span class="lineno"> 4115</span>&#160; acc31 = fma(a0.s6, b6.s1, acc31);</div><div class="line"><a name="l04116"></a><span class="lineno"> 4116</span>&#160; acc31 = fma(a0.s7, b7.s1, acc31);</div><div class="line"><a name="l04117"></a><span class="lineno"> 4117</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 3</span></div><div class="line"><a name="l04118"></a><span class="lineno"> 4118</span>&#160;</div><div class="line"><a name="l04119"></a><span class="lineno"> 4119</span>&#160; src_addr.s0 += <span class="keyword">sizeof</span>(float) * 8;</div><div class="line"><a name="l04120"></a><span class="lineno"> 4120</span>&#160; }</div><div class="line"><a name="l04121"></a><span class="lineno"> 4121</span>&#160; <span class="comment">// float size increment</span></div><div class="line"><a name="l04122"></a><span class="lineno"> 4122</span>&#160; <span class="keywordflow">for</span>(; i &lt; (int)COLS_A; ++i)</div><div class="line"><a name="l04123"></a><span class="lineno"> 4123</span>&#160; {</div><div class="line"><a name="l04124"></a><span class="lineno"> 4124</span>&#160;<span class="preprocessor">#if defined(REINTERPRET_INPUT_AS_3D)</span></div><div class="line"><a name="l04125"></a><span class="lineno"> 4125</span>&#160; <span class="comment">// Load values from matrix A</span></div><div class="line"><a name="l04126"></a><span class="lineno"> 4126</span>&#160; <span class="keywordtype">float</span> a0 = *((__global <span class="keywordtype">float</span> *)(src0_ptr + src_addr.s0 + 0 * src0_stride_y + zin.s0));</div><div class="line"><a name="l04127"></a><span class="lineno"> 4127</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 1</span></div><div class="line"><a name="l04128"></a><span class="lineno"> 4128</span>&#160; <span class="keywordtype">float</span> a1 = *((__global <span class="keywordtype">float</span> *)(src0_ptr + src_addr.s0 + 1 * src0_stride_y + zin.s1));</div><div class="line"><a name="l04129"></a><span class="lineno"> 4129</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 1</span></div><div class="line"><a name="l04130"></a><span class="lineno"> 4130</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 2</span></div><div class="line"><a name="l04131"></a><span class="lineno"> 4131</span>&#160; <span class="keywordtype">float</span> a2 = *((__global <span class="keywordtype">float</span> *)(src0_ptr + src_addr.s0 + 2 * src0_stride_y + zin.s2));</div><div class="line"><a name="l04132"></a><span class="lineno"> 4132</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 2</span></div><div class="line"><a name="l04133"></a><span class="lineno"> 4133</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 3</span></div><div class="line"><a name="l04134"></a><span class="lineno"> 4134</span>&#160; <span class="keywordtype">float</span> a3 = *((__global <span class="keywordtype">float</span> *)(src0_ptr + src_addr.s0 + 3 * src0_stride_y + zin.s3));</div><div class="line"><a name="l04135"></a><span class="lineno"> 4135</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 3</span></div><div class="line"><a name="l04136"></a><span class="lineno"> 4136</span>&#160;<span class="preprocessor">#else // defined(REINTERPRET_INPUT_AS_3D)</span></div><div class="line"><a name="l04137"></a><span class="lineno"> 4137</span>&#160; <span class="comment">// Load values from matrix A</span></div><div class="line"><a name="l04138"></a><span class="lineno"> 4138</span>&#160; <span class="keywordtype">float</span> a0 = *((__global <span class="keywordtype">float</span> *)(src0_ptr + src_addr.s0 + 0 * src0_stride_y));</div><div class="line"><a name="l04139"></a><span class="lineno"> 4139</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 1</span></div><div class="line"><a name="l04140"></a><span class="lineno"> 4140</span>&#160; <span class="keywordtype">float</span> a1 = *((__global <span class="keywordtype">float</span> *)(src0_ptr + src_addr.s0 + 1 * src0_stride_y));</div><div class="line"><a name="l04141"></a><span class="lineno"> 4141</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 1</span></div><div class="line"><a name="l04142"></a><span class="lineno"> 4142</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 2</span></div><div class="line"><a name="l04143"></a><span class="lineno"> 4143</span>&#160; <span class="keywordtype">float</span> a2 = *((__global <span class="keywordtype">float</span> *)(src0_ptr + src_addr.s0 + 2 * src0_stride_y));</div><div class="line"><a name="l04144"></a><span class="lineno"> 4144</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 2</span></div><div class="line"><a name="l04145"></a><span class="lineno"> 4145</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 3</span></div><div class="line"><a name="l04146"></a><span class="lineno"> 4146</span>&#160; <span class="keywordtype">float</span> a3 = *((__global <span class="keywordtype">float</span> *)(src0_ptr + src_addr.s0 + 3 * src0_stride_y));</div><div class="line"><a name="l04147"></a><span class="lineno"> 4147</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 3</span></div><div class="line"><a name="l04148"></a><span class="lineno"> 4148</span>&#160;<span class="preprocessor">#endif // defined(REINTERPRET_INPUT_AS_3D)</span></div><div class="line"><a name="l04149"></a><span class="lineno"> 4149</span>&#160;</div><div class="line"><a name="l04150"></a><span class="lineno"> 4150</span>&#160; <span class="comment">// Load values from matrix B</span></div><div class="line"><a name="l04151"></a><span class="lineno"> 4151</span>&#160; float2 b0 = vload2(0, (__global <span class="keywordtype">float</span> *)(src1_ptr + src_addr.s1));</div><div class="line"><a name="l04152"></a><span class="lineno"> 4152</span>&#160; src_addr.s1 += src1_stride_y;</div><div class="line"><a name="l04153"></a><span class="lineno"> 4153</span>&#160;</div><div class="line"><a name="l04154"></a><span class="lineno"> 4154</span>&#160; <span class="comment">// Multiply and accumulate</span></div><div class="line"><a name="l04155"></a><span class="lineno"> 4155</span>&#160; acc00 = fma(a0, b0.s0, acc00);</div><div class="line"><a name="l04156"></a><span class="lineno"> 4156</span>&#160; acc01 = fma(a0, b0.s1, acc01);</div><div class="line"><a name="l04157"></a><span class="lineno"> 4157</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 1</span></div><div class="line"><a name="l04158"></a><span class="lineno"> 4158</span>&#160; acc10 = fma(a1, b0.s0, acc10);</div><div class="line"><a name="l04159"></a><span class="lineno"> 4159</span>&#160; acc11 = fma(a1, b0.s1, acc11);</div><div class="line"><a name="l04160"></a><span class="lineno"> 4160</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 1</span></div><div class="line"><a name="l04161"></a><span class="lineno"> 4161</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 2</span></div><div class="line"><a name="l04162"></a><span class="lineno"> 4162</span>&#160; acc20 = fma(a2, b0.s0, acc20);</div><div class="line"><a name="l04163"></a><span class="lineno"> 4163</span>&#160; acc21 = fma(a2, b0.s1, acc21);</div><div class="line"><a name="l04164"></a><span class="lineno"> 4164</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 2</span></div><div class="line"><a name="l04165"></a><span class="lineno"> 4165</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 3</span></div><div class="line"><a name="l04166"></a><span class="lineno"> 4166</span>&#160; acc30 = fma(a3, b0.s0, acc30);</div><div class="line"><a name="l04167"></a><span class="lineno"> 4167</span>&#160; acc31 = fma(a3, b0.s1, acc31);</div><div class="line"><a name="l04168"></a><span class="lineno"> 4168</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 3</span></div><div class="line"><a name="l04169"></a><span class="lineno"> 4169</span>&#160;</div><div class="line"><a name="l04170"></a><span class="lineno"> 4170</span>&#160; src_addr.s0 += <span class="keyword">sizeof</span>(float);</div><div class="line"><a name="l04171"></a><span class="lineno"> 4171</span>&#160; }</div><div class="line"><a name="l04172"></a><span class="lineno"> 4172</span>&#160;</div><div class="line"><a name="l04173"></a><span class="lineno"> 4173</span>&#160; <span class="comment">// Multiply by the weight of matrix-matrix product and store the result</span></div><div class="line"><a name="l04174"></a><span class="lineno"> 4174</span>&#160;<span class="preprocessor">#if defined(ALPHA)</span></div><div class="line"><a name="l04175"></a><span class="lineno"> 4175</span>&#160; acc00 = acc00 * ALPHA;</div><div class="line"><a name="l04176"></a><span class="lineno"> 4176</span>&#160; acc01 = acc01 * ALPHA;</div><div class="line"><a name="l04177"></a><span class="lineno"> 4177</span>&#160;<span class="preprocessor">#endif // defined(ALPHA)</span></div><div class="line"><a name="l04178"></a><span class="lineno"> 4178</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 1 &amp;&amp; defined(ALPHA)</span></div><div class="line"><a name="l04179"></a><span class="lineno"> 4179</span>&#160; acc10 = acc10 * ALPHA;</div><div class="line"><a name="l04180"></a><span class="lineno"> 4180</span>&#160; acc11 = acc11 * ALPHA;</div><div class="line"><a name="l04181"></a><span class="lineno"> 4181</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 1 &amp;&amp; defined(ALPHA)</span></div><div class="line"><a name="l04182"></a><span class="lineno"> 4182</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 2 &amp;&amp; defined(ALPHA)</span></div><div class="line"><a name="l04183"></a><span class="lineno"> 4183</span>&#160; acc20 = acc20 * ALPHA;</div><div class="line"><a name="l04184"></a><span class="lineno"> 4184</span>&#160; acc21 = acc21 * ALPHA;</div><div class="line"><a name="l04185"></a><span class="lineno"> 4185</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 2 &amp;&amp; defined(ALPHA)</span></div><div class="line"><a name="l04186"></a><span class="lineno"> 4186</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 3 &amp;&amp; defined(ALPHA)</span></div><div class="line"><a name="l04187"></a><span class="lineno"> 4187</span>&#160; acc30 = acc30 * ALPHA;</div><div class="line"><a name="l04188"></a><span class="lineno"> 4188</span>&#160; acc31 = acc31 * ALPHA;</div><div class="line"><a name="l04189"></a><span class="lineno"> 4189</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 3 &amp;&amp; defined(ALPHA)</span></div><div class="line"><a name="l04190"></a><span class="lineno"> 4190</span>&#160;</div><div class="line"><a name="l04191"></a><span class="lineno"> 4191</span>&#160; <span class="keywordtype">int</span> z = get_global_id(2);</div><div class="line"><a name="l04192"></a><span class="lineno"> 4192</span>&#160;</div><div class="line"><a name="l04193"></a><span class="lineno"> 4193</span>&#160; <span class="comment">// Compute destination address</span></div><div class="line"><a name="l04194"></a><span class="lineno"> 4194</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="l04195"></a><span class="lineno"> 4195</span>&#160;</div><div class="line"><a name="l04196"></a><span class="lineno"> 4196</span>&#160; <span class="comment">// Compute dst address</span></div><div class="line"><a name="l04197"></a><span class="lineno"> 4197</span>&#160; __global uchar *dst_addr = <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>(&amp;<a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#adbf67dcee294e673cf796f1ed8aeb6a4">dst</a>, 0, 0);</div><div class="line"><a name="l04198"></a><span class="lineno"> 4198</span>&#160;</div><div class="line"><a name="l04199"></a><span class="lineno"> 4199</span>&#160;<span class="preprocessor">#if defined(ADD_VEC_C)</span></div><div class="line"><a name="l04200"></a><span class="lineno"> 4200</span>&#160; __global <span class="keywordtype">float</span> *src2_addr = (__global <span class="keywordtype">float</span> *)(src2_ptr + src2_offset_first_element_in_bytes + get_global_id(0) * src2_step_x);</div><div class="line"><a name="l04201"></a><span class="lineno"> 4201</span>&#160; float2 c0 = vload2(0, src2_addr);</div><div class="line"><a name="l04202"></a><span class="lineno"> 4202</span>&#160;</div><div class="line"><a name="l04203"></a><span class="lineno"> 4203</span>&#160; acc00 += c0.s0;</div><div class="line"><a name="l04204"></a><span class="lineno"> 4204</span>&#160; acc01 += c0.s1;</div><div class="line"><a name="l04205"></a><span class="lineno"> 4205</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 1</span></div><div class="line"><a name="l04206"></a><span class="lineno"> 4206</span>&#160; acc10 += c0.s0;</div><div class="line"><a name="l04207"></a><span class="lineno"> 4207</span>&#160; acc11 += c0.s1;</div><div class="line"><a name="l04208"></a><span class="lineno"> 4208</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 1</span></div><div class="line"><a name="l04209"></a><span class="lineno"> 4209</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 2</span></div><div class="line"><a name="l04210"></a><span class="lineno"> 4210</span>&#160; acc20 += c0.s0;</div><div class="line"><a name="l04211"></a><span class="lineno"> 4211</span>&#160; acc21 += c0.s1;</div><div class="line"><a name="l04212"></a><span class="lineno"> 4212</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 2</span></div><div class="line"><a name="l04213"></a><span class="lineno"> 4213</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 3</span></div><div class="line"><a name="l04214"></a><span class="lineno"> 4214</span>&#160; acc30 += c0.s0;</div><div class="line"><a name="l04215"></a><span class="lineno"> 4215</span>&#160; acc31 += c0.s1;</div><div class="line"><a name="l04216"></a><span class="lineno"> 4216</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 3</span></div><div class="line"><a name="l04217"></a><span class="lineno"> 4217</span>&#160;<span class="preprocessor">#endif </span><span class="comment">/* defined(ADD_VEC_C) */</span><span class="preprocessor"></span></div><div class="line"><a name="l04218"></a><span class="lineno"> 4218</span>&#160;</div><div class="line"><a name="l04219"></a><span class="lineno"> 4219</span>&#160;<span class="preprocessor">#if defined(REINTERPRET_OUTPUT_AS_3D)</span></div><div class="line"><a name="l04220"></a><span class="lineno"> 4220</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="l04221"></a><span class="lineno"> 4221</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="l04222"></a><span class="lineno"> 4222</span>&#160; <span class="comment">//</span></div><div class="line"><a name="l04223"></a><span class="lineno"> 4223</span>&#160; <span class="comment">// | |</span></div><div class="line"><a name="l04224"></a><span class="lineno"> 4224</span>&#160; <span class="comment">// | plane0 |</span></div><div class="line"><a name="l04225"></a><span class="lineno"> 4225</span>&#160; <span class="comment">// | |</span></div><div class="line"><a name="l04226"></a><span class="lineno"> 4226</span>&#160; <span class="comment">// |__________________|</span></div><div class="line"><a name="l04227"></a><span class="lineno"> 4227</span>&#160; <span class="comment">// |******************|</span></div><div class="line"><a name="l04228"></a><span class="lineno"> 4228</span>&#160; <span class="comment">// | cross_plane_pad |</span></div><div class="line"><a name="l04229"></a><span class="lineno"> 4229</span>&#160; <span class="comment">// |******************|</span></div><div class="line"><a name="l04230"></a><span class="lineno"> 4230</span>&#160; <span class="comment">// | |</span></div><div class="line"><a name="l04231"></a><span class="lineno"> 4231</span>&#160; <span class="comment">// | plane1 |</span></div><div class="line"><a name="l04232"></a><span class="lineno"> 4232</span>&#160; <span class="comment">// | |</span></div><div class="line"><a name="l04233"></a><span class="lineno"> 4233</span>&#160; <span class="comment">// |__________________|</span></div><div class="line"><a name="l04234"></a><span class="lineno"> 4234</span>&#160;</div><div class="line"><a name="l04235"></a><span class="lineno"> 4235</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="l04236"></a><span class="lineno"> 4236</span>&#160; uint4 zout = ((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="l04237"></a><span class="lineno"> 4237</span>&#160; zout = min(DEPTH_GEMM3D - 1, zout);</div><div class="line"><a name="l04238"></a><span class="lineno"> 4238</span>&#160;</div><div class="line"><a name="l04239"></a><span class="lineno"> 4239</span>&#160; <span class="comment">// Add offset due to the cross plane paddings</span></div><div class="line"><a name="l04240"></a><span class="lineno"> 4240</span>&#160; zout *= (dst_cross_plane_pad * dst_stride_y);</div><div class="line"><a name="l04241"></a><span class="lineno"> 4241</span>&#160;</div><div class="line"><a name="l04242"></a><span class="lineno"> 4242</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="l04243"></a><span class="lineno"> 4243</span>&#160; <span class="comment">// multiply dst_stride_z by DEPTH_GEMM3D</span></div><div class="line"><a name="l04244"></a><span class="lineno"> 4244</span>&#160; dst_addr += z * dst_stride_z * DEPTH_GEMM3D;</div><div class="line"><a name="l04245"></a><span class="lineno"> 4245</span>&#160;</div><div class="line"><a name="l04246"></a><span class="lineno"> 4246</span>&#160; <span class="comment">// Store the output block</span></div><div class="line"><a name="l04247"></a><span class="lineno"> 4247</span>&#160; vstore2((float2)(acc00, acc01), 0, (__global <span class="keywordtype">float</span> *)(dst_addr + 0 * dst_stride_y + zout.s0));</div><div class="line"><a name="l04248"></a><span class="lineno"> 4248</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 1</span></div><div class="line"><a name="l04249"></a><span class="lineno"> 4249</span>&#160; vstore2((float2)(acc10, acc11), 0, (__global <span class="keywordtype">float</span> *)(dst_addr + 1 * dst_stride_y + zout.s1));</div><div class="line"><a name="l04250"></a><span class="lineno"> 4250</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 1</span></div><div class="line"><a name="l04251"></a><span class="lineno"> 4251</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 2</span></div><div class="line"><a name="l04252"></a><span class="lineno"> 4252</span>&#160; vstore2((float2)(acc20, acc21), 0, (__global <span class="keywordtype">float</span> *)(dst_addr + 2 * dst_stride_y + zout.s2));</div><div class="line"><a name="l04253"></a><span class="lineno"> 4253</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 2</span></div><div class="line"><a name="l04254"></a><span class="lineno"> 4254</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 3</span></div><div class="line"><a name="l04255"></a><span class="lineno"> 4255</span>&#160; vstore2((float2)(acc30, acc31), 0, (__global <span class="keywordtype">float</span> *)(dst_addr + 3 * dst_stride_y + zout.s3));</div><div class="line"><a name="l04256"></a><span class="lineno"> 4256</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 3</span></div><div class="line"><a name="l04257"></a><span class="lineno"> 4257</span>&#160;</div><div class="line"><a name="l04258"></a><span class="lineno"> 4258</span>&#160;<span class="preprocessor">#else // defined(REINTERPRET_OUTPUT_AS_3D)</span></div><div class="line"><a name="l04259"></a><span class="lineno"> 4259</span>&#160; <span class="comment">// Add offset for batched GEMM</span></div><div class="line"><a name="l04260"></a><span class="lineno"> 4260</span>&#160; dst_addr += z * dst_stride_z;</div><div class="line"><a name="l04261"></a><span class="lineno"> 4261</span>&#160;</div><div class="line"><a name="l04262"></a><span class="lineno"> 4262</span>&#160; <span class="comment">// Store the output block</span></div><div class="line"><a name="l04263"></a><span class="lineno"> 4263</span>&#160; vstore2((float2)(acc00, acc01), 0, (__global <span class="keywordtype">float</span> *)(dst_addr + 0 * dst_stride_y));</div><div class="line"><a name="l04264"></a><span class="lineno"> 4264</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 1</span></div><div class="line"><a name="l04265"></a><span class="lineno"> 4265</span>&#160; vstore2((float2)(acc10, acc11), 0, (__global <span class="keywordtype">float</span> *)(dst_addr + 1 * dst_stride_y));</div><div class="line"><a name="l04266"></a><span class="lineno"> 4266</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 1</span></div><div class="line"><a name="l04267"></a><span class="lineno"> 4267</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 2</span></div><div class="line"><a name="l04268"></a><span class="lineno"> 4268</span>&#160; vstore2((float2)(acc20, acc21), 0, (__global <span class="keywordtype">float</span> *)(dst_addr + 2 * dst_stride_y));</div><div class="line"><a name="l04269"></a><span class="lineno"> 4269</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 2</span></div><div class="line"><a name="l04270"></a><span class="lineno"> 4270</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 3</span></div><div class="line"><a name="l04271"></a><span class="lineno"> 4271</span>&#160; vstore2((float2)(acc30, acc31), 0, (__global <span class="keywordtype">float</span> *)(dst_addr + 3 * dst_stride_y));</div><div class="line"><a name="l04272"></a><span class="lineno"> 4272</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 3</span></div><div class="line"><a name="l04273"></a><span class="lineno"> 4273</span>&#160;<span class="preprocessor">#endif // defined(REINTERPRET_OUTPUT_AS_3D)</span></div><div class="line"><a name="l04274"></a><span class="lineno"> 4274</span>&#160;}</div><div class="line"><a name="l04275"></a><span class="lineno"> 4275</span>&#160;</div><div class="line"><a name="l04276"></a><span class="lineno"> 4276</span>&#160;<span class="preprocessor">#if defined(ARM_COMPUTE_OPENCL_FP16_ENABLED)</span></div><div class="line"><a name="l04277"></a><span class="lineno"> 4277</span>&#160;</div><div class="line"><a name="l04326"></a><span class="lineno"> 4326</span>&#160;__kernel <span class="keywordtype">void</span> gemm_mm_floating_point_f16_bifrost_acc32(<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="l04327"></a><span class="lineno"> 4327</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="l04328"></a><span class="lineno"> 4328</span>&#160;#<span class="keywordflow">if</span> defined(ADD_VEC_C)</div><div class="line"><a name="l04329"></a><span class="lineno"> 4329</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a40a6eb9f2a7712f08d6bb8ff6c9e6ca7">VECTOR_DECLARATION</a>(src2),</div><div class="line"><a name="l04330"></a><span class="lineno"> 4330</span>&#160;#endif <span class="comment">/* defined(ADD_VEC_C) */</span></div><div class="line"><a name="l04331"></a><span class="lineno"> 4331</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="l04332"></a><span class="lineno"> 4332</span>&#160; uint src0_stride_z,</div><div class="line"><a name="l04333"></a><span class="lineno"> 4333</span>&#160; uint src1_stride_z,</div><div class="line"><a name="l04334"></a><span class="lineno"> 4334</span>&#160; uint dst_stride_z</div><div class="line"><a name="l04335"></a><span class="lineno"> 4335</span>&#160;#<span class="keywordflow">if</span> defined(REINTERPRET_INPUT_AS_3D)</div><div class="line"><a name="l04336"></a><span class="lineno"> 4336</span>&#160; ,</div><div class="line"><a name="l04337"></a><span class="lineno"> 4337</span>&#160; uint src_cross_plane_pad</div><div class="line"><a name="l04338"></a><span class="lineno"> 4338</span>&#160;#endif <span class="comment">// REINTERPRET_INPUT_AS_3D</span></div><div class="line"><a name="l04339"></a><span class="lineno"> 4339</span>&#160;#<span class="keywordflow">if</span> defined(REINTERPRET_OUTPUT_AS_3D)</div><div class="line"><a name="l04340"></a><span class="lineno"> 4340</span>&#160; ,</div><div class="line"><a name="l04341"></a><span class="lineno"> 4341</span>&#160; uint dst_cross_plane_pad</div><div class="line"><a name="l04342"></a><span class="lineno"> 4342</span>&#160;#endif <span class="comment">// REINTERPRET_OUTPUT_AS_3D</span></div><div class="line"><a name="l04343"></a><span class="lineno"> 4343</span>&#160; )</div><div class="line"><a name="l04344"></a><span class="lineno"> 4344</span>&#160;{</div><div class="line"><a name="l04345"></a><span class="lineno"> 4345</span>&#160; <span class="keywordtype">int</span> idx = get_global_id(0) * NUM_ELEMS_PROCESSED_PER_THREAD_X;</div><div class="line"><a name="l04346"></a><span class="lineno"> 4346</span>&#160;</div><div class="line"><a name="l04347"></a><span class="lineno"> 4347</span>&#160; <span class="comment">// Compute starting address for matrix A and Matrix B</span></div><div class="line"><a name="l04348"></a><span class="lineno"> 4348</span>&#160; int2 src_addr = ((int2)(src0_offset_first_element_in_bytes, src1_offset_first_element_in_bytes));</div><div class="line"><a name="l04349"></a><span class="lineno"> 4349</span>&#160;</div><div class="line"><a name="l04350"></a><span class="lineno"> 4350</span>&#160; <span class="comment">// Update address for the matrix A</span></div><div class="line"><a name="l04351"></a><span class="lineno"> 4351</span>&#160; src_addr.s0 += get_global_id(1) * src0_stride_y * NUM_ELEMS_PROCESSED_PER_THREAD_Y;</div><div class="line"><a name="l04352"></a><span class="lineno"> 4352</span>&#160;</div><div class="line"><a name="l04353"></a><span class="lineno"> 4353</span>&#160; <span class="comment">// Update address for the matrix B</span></div><div class="line"><a name="l04354"></a><span class="lineno"> 4354</span>&#160; src_addr.s1 += idx * <span class="keyword">sizeof</span>(<a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a>);</div><div class="line"><a name="l04355"></a><span class="lineno"> 4355</span>&#160;</div><div class="line"><a name="l04356"></a><span class="lineno"> 4356</span>&#160;<span class="preprocessor">#if defined(REINTERPRET_INPUT_AS_3D)</span></div><div class="line"><a name="l04357"></a><span class="lineno"> 4357</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="l04358"></a><span class="lineno"> 4358</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="l04359"></a><span class="lineno"> 4359</span>&#160; <span class="comment">//</span></div><div class="line"><a name="l04360"></a><span class="lineno"> 4360</span>&#160; <span class="comment">// | |</span></div><div class="line"><a name="l04361"></a><span class="lineno"> 4361</span>&#160; <span class="comment">// | plane0 |</span></div><div class="line"><a name="l04362"></a><span class="lineno"> 4362</span>&#160; <span class="comment">// | |</span></div><div class="line"><a name="l04363"></a><span class="lineno"> 4363</span>&#160; <span class="comment">// |__________________|</span></div><div class="line"><a name="l04364"></a><span class="lineno"> 4364</span>&#160; <span class="comment">// |******************|</span></div><div class="line"><a name="l04365"></a><span class="lineno"> 4365</span>&#160; <span class="comment">// | cross_plane_pad |</span></div><div class="line"><a name="l04366"></a><span class="lineno"> 4366</span>&#160; <span class="comment">// |******************|</span></div><div class="line"><a name="l04367"></a><span class="lineno"> 4367</span>&#160; <span class="comment">// | |</span></div><div class="line"><a name="l04368"></a><span class="lineno"> 4368</span>&#160; <span class="comment">// | plane1 |</span></div><div class="line"><a name="l04369"></a><span class="lineno"> 4369</span>&#160; <span class="comment">// | |</span></div><div class="line"><a name="l04370"></a><span class="lineno"> 4370</span>&#160; <span class="comment">// |__________________|</span></div><div class="line"><a name="l04371"></a><span class="lineno"> 4371</span>&#160;</div><div class="line"><a name="l04372"></a><span class="lineno"> 4372</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="l04373"></a><span class="lineno"> 4373</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="l04374"></a><span class="lineno"> 4374</span>&#160; zin = min(DEPTH_GEMM3D - 1, zin);</div><div class="line"><a name="l04375"></a><span class="lineno"> 4375</span>&#160;</div><div class="line"><a name="l04376"></a><span class="lineno"> 4376</span>&#160; <span class="comment">// Add offset due to the cross plane paddings</span></div><div class="line"><a name="l04377"></a><span class="lineno"> 4377</span>&#160; zin *= (src_cross_plane_pad * src0_stride_y);</div><div class="line"><a name="l04378"></a><span class="lineno"> 4378</span>&#160;</div><div class="line"><a name="l04379"></a><span class="lineno"> 4379</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="l04380"></a><span class="lineno"> 4380</span>&#160; <span class="comment">// multiply src0_stride_z by DEPTH_GEMM3D</span></div><div class="line"><a name="l04381"></a><span class="lineno"> 4381</span>&#160; src_addr.s0 += get_global_id(2) * src0_stride_z * DEPTH_GEMM3D;</div><div class="line"><a name="l04382"></a><span class="lineno"> 4382</span>&#160;</div><div class="line"><a name="l04383"></a><span class="lineno"> 4383</span>&#160;<span class="preprocessor">#else // defined(REINTERPRET_INPUT_AS_3D)</span></div><div class="line"><a name="l04384"></a><span class="lineno"> 4384</span>&#160;</div><div class="line"><a name="l04385"></a><span class="lineno"> 4385</span>&#160; <span class="comment">// Add offset for batched GEMM</span></div><div class="line"><a name="l04386"></a><span class="lineno"> 4386</span>&#160; src_addr.s0 += get_global_id(2) * src0_stride_z;</div><div class="line"><a name="l04387"></a><span class="lineno"> 4387</span>&#160;</div><div class="line"><a name="l04388"></a><span class="lineno"> 4388</span>&#160;<span class="preprocessor">#endif // defined(REINTERPRET_INPUT_AS_3D)</span></div><div class="line"><a name="l04389"></a><span class="lineno"> 4389</span>&#160;</div><div class="line"><a name="l04390"></a><span class="lineno"> 4390</span>&#160;<span class="preprocessor">#if defined(MATRIX_B_DEPTH)</span></div><div class="line"><a name="l04391"></a><span class="lineno"> 4391</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="l04392"></a><span class="lineno"> 4392</span>&#160; src_addr.s1 += (get_global_id(2) % MATRIX_B_DEPTH) * src1_stride_z;</div><div class="line"><a name="l04393"></a><span class="lineno"> 4393</span>&#160;<span class="preprocessor">#else // defined(MATRIX_B_DEPTH)</span></div><div class="line"><a name="l04394"></a><span class="lineno"> 4394</span>&#160; src_addr.s1 += get_global_id(2) * src1_stride_z;</div><div class="line"><a name="l04395"></a><span class="lineno"> 4395</span>&#160;<span class="preprocessor">#endif // defined(MATRIX_B_DEPTH)</span></div><div class="line"><a name="l04396"></a><span class="lineno"> 4396</span>&#160;</div><div class="line"><a name="l04397"></a><span class="lineno"> 4397</span>&#160; float8 acc0 = 0.0h;</div><div class="line"><a name="l04398"></a><span class="lineno"> 4398</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 1</span></div><div class="line"><a name="l04399"></a><span class="lineno"> 4399</span>&#160; float8 acc1 = 0.0h;</div><div class="line"><a name="l04400"></a><span class="lineno"> 4400</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 1</span></div><div class="line"><a name="l04401"></a><span class="lineno"> 4401</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 2</span></div><div class="line"><a name="l04402"></a><span class="lineno"> 4402</span>&#160; float8 acc2 = 0.0h;</div><div class="line"><a name="l04403"></a><span class="lineno"> 4403</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 2</span></div><div class="line"><a name="l04404"></a><span class="lineno"> 4404</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 3</span></div><div class="line"><a name="l04405"></a><span class="lineno"> 4405</span>&#160; float8 acc3 = 0.0h;</div><div class="line"><a name="l04406"></a><span class="lineno"> 4406</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 3</span></div><div class="line"><a name="l04407"></a><span class="lineno"> 4407</span>&#160;</div><div class="line"><a name="l04408"></a><span class="lineno"> 4408</span>&#160; <span class="keywordtype">int</span> i = 0;</div><div class="line"><a name="l04409"></a><span class="lineno"> 4409</span>&#160; <span class="keywordflow">for</span>(; i &lt;= ((int)COLS_A - 4); i += 4)</div><div class="line"><a name="l04410"></a><span class="lineno"> 4410</span>&#160; {</div><div class="line"><a name="l04411"></a><span class="lineno"> 4411</span>&#160;<span class="preprocessor">#if defined(REINTERPRET_INPUT_AS_3D)</span></div><div class="line"><a name="l04412"></a><span class="lineno"> 4412</span>&#160; <span class="comment">// Load values from matrix A</span></div><div class="line"><a name="l04413"></a><span class="lineno"> 4413</span>&#160; half4 a0 = vload4(0, (__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)(src0_ptr + src_addr.s0 + 0 * src0_stride_y + zin.s0));</div><div class="line"><a name="l04414"></a><span class="lineno"> 4414</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 1</span></div><div class="line"><a name="l04415"></a><span class="lineno"> 4415</span>&#160; half4 a1 = vload4(0, (__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)(src0_ptr + src_addr.s0 + 1 * src0_stride_y + zin.s1));</div><div class="line"><a name="l04416"></a><span class="lineno"> 4416</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 1</span></div><div class="line"><a name="l04417"></a><span class="lineno"> 4417</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 2</span></div><div class="line"><a name="l04418"></a><span class="lineno"> 4418</span>&#160; half4 a2 = vload4(0, (__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)(src0_ptr + src_addr.s0 + 2 * src0_stride_y + zin.s2));</div><div class="line"><a name="l04419"></a><span class="lineno"> 4419</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 2</span></div><div class="line"><a name="l04420"></a><span class="lineno"> 4420</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 3</span></div><div class="line"><a name="l04421"></a><span class="lineno"> 4421</span>&#160; half4 a3 = vload4(0, (__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)(src0_ptr + src_addr.s0 + 3 * src0_stride_y + zin.s3));</div><div class="line"><a name="l04422"></a><span class="lineno"> 4422</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 3</span></div><div class="line"><a name="l04423"></a><span class="lineno"> 4423</span>&#160;<span class="preprocessor">#else // defined(REINTERPRET_INPUT_AS_3D)</span></div><div class="line"><a name="l04424"></a><span class="lineno"> 4424</span>&#160; <span class="comment">// Load values from matrix A</span></div><div class="line"><a name="l04425"></a><span class="lineno"> 4425</span>&#160; half4 a0 = vload4(0, (__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)(src0_ptr + src_addr.s0 + 0 * src0_stride_y));</div><div class="line"><a name="l04426"></a><span class="lineno"> 4426</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 1</span></div><div class="line"><a name="l04427"></a><span class="lineno"> 4427</span>&#160; half4 a1 = vload4(0, (__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)(src0_ptr + src_addr.s0 + 1 * src0_stride_y));</div><div class="line"><a name="l04428"></a><span class="lineno"> 4428</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 1</span></div><div class="line"><a name="l04429"></a><span class="lineno"> 4429</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 2</span></div><div class="line"><a name="l04430"></a><span class="lineno"> 4430</span>&#160; half4 a2 = vload4(0, (__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)(src0_ptr + src_addr.s0 + 2 * src0_stride_y));</div><div class="line"><a name="l04431"></a><span class="lineno"> 4431</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 2</span></div><div class="line"><a name="l04432"></a><span class="lineno"> 4432</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 3</span></div><div class="line"><a name="l04433"></a><span class="lineno"> 4433</span>&#160; half4 a3 = vload4(0, (__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)(src0_ptr + src_addr.s0 + 3 * src0_stride_y));</div><div class="line"><a name="l04434"></a><span class="lineno"> 4434</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 3</span></div><div class="line"><a name="l04435"></a><span class="lineno"> 4435</span>&#160;<span class="preprocessor">#endif // defined(REINTERPRET_INPUT_AS_3D)</span></div><div class="line"><a name="l04436"></a><span class="lineno"> 4436</span>&#160;</div><div class="line"><a name="l04437"></a><span class="lineno"> 4437</span>&#160; <span class="comment">// Load values from matrix B</span></div><div class="line"><a name="l04438"></a><span class="lineno"> 4438</span>&#160; float8 b0 = convert_float8(vload8(0, (__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)(src1_ptr + src_addr.s1)));</div><div class="line"><a name="l04439"></a><span class="lineno"> 4439</span>&#160; src_addr.s1 += src1_stride_y;</div><div class="line"><a name="l04440"></a><span class="lineno"> 4440</span>&#160;</div><div class="line"><a name="l04441"></a><span class="lineno"> 4441</span>&#160; <span class="comment">// Accumulate</span></div><div class="line"><a name="l04442"></a><span class="lineno"> 4442</span>&#160; acc0 = fma(b0, (float8)a0.s0, acc0);</div><div class="line"><a name="l04443"></a><span class="lineno"> 4443</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 1</span></div><div class="line"><a name="l04444"></a><span class="lineno"> 4444</span>&#160; acc1 = fma(b0, (float8)a1.s0, acc1);</div><div class="line"><a name="l04445"></a><span class="lineno"> 4445</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 1</span></div><div class="line"><a name="l04446"></a><span class="lineno"> 4446</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 2</span></div><div class="line"><a name="l04447"></a><span class="lineno"> 4447</span>&#160; acc2 = fma(b0, (float8)a2.s0, acc2);</div><div class="line"><a name="l04448"></a><span class="lineno"> 4448</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 2</span></div><div class="line"><a name="l04449"></a><span class="lineno"> 4449</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 3</span></div><div class="line"><a name="l04450"></a><span class="lineno"> 4450</span>&#160; acc3 = fma(b0, (float8)a3.s0, acc3);</div><div class="line"><a name="l04451"></a><span class="lineno"> 4451</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 3</span></div><div class="line"><a name="l04452"></a><span class="lineno"> 4452</span>&#160;</div><div class="line"><a name="l04453"></a><span class="lineno"> 4453</span>&#160; b0 = convert_float8(vload8(0, (__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)(src1_ptr + src_addr.s1)));</div><div class="line"><a name="l04454"></a><span class="lineno"> 4454</span>&#160; src_addr.s1 += src1_stride_y;</div><div class="line"><a name="l04455"></a><span class="lineno"> 4455</span>&#160; acc0 = fma(b0, (float8)a0.s1, acc0);</div><div class="line"><a name="l04456"></a><span class="lineno"> 4456</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 1</span></div><div class="line"><a name="l04457"></a><span class="lineno"> 4457</span>&#160; acc1 = fma(b0, (float8)a1.s1, acc1);</div><div class="line"><a name="l04458"></a><span class="lineno"> 4458</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 1</span></div><div class="line"><a name="l04459"></a><span class="lineno"> 4459</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 2</span></div><div class="line"><a name="l04460"></a><span class="lineno"> 4460</span>&#160; acc2 = fma(b0, (float8)a2.s1, acc2);</div><div class="line"><a name="l04461"></a><span class="lineno"> 4461</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 2</span></div><div class="line"><a name="l04462"></a><span class="lineno"> 4462</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 3</span></div><div class="line"><a name="l04463"></a><span class="lineno"> 4463</span>&#160; acc3 = fma(b0, (float8)a3.s1, acc3);</div><div class="line"><a name="l04464"></a><span class="lineno"> 4464</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 3</span></div><div class="line"><a name="l04465"></a><span class="lineno"> 4465</span>&#160;</div><div class="line"><a name="l04466"></a><span class="lineno"> 4466</span>&#160; b0 = convert_float8(vload8(0, (__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)(src1_ptr + src_addr.s1)));</div><div class="line"><a name="l04467"></a><span class="lineno"> 4467</span>&#160; src_addr.s1 += src1_stride_y;</div><div class="line"><a name="l04468"></a><span class="lineno"> 4468</span>&#160; acc0 = fma(b0, (float8)a0.s2, acc0);</div><div class="line"><a name="l04469"></a><span class="lineno"> 4469</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 1</span></div><div class="line"><a name="l04470"></a><span class="lineno"> 4470</span>&#160; acc1 = fma(b0, (float8)a1.s2, acc1);</div><div class="line"><a name="l04471"></a><span class="lineno"> 4471</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 1</span></div><div class="line"><a name="l04472"></a><span class="lineno"> 4472</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 2</span></div><div class="line"><a name="l04473"></a><span class="lineno"> 4473</span>&#160; acc2 = fma(b0, (float8)a2.s2, acc2);</div><div class="line"><a name="l04474"></a><span class="lineno"> 4474</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 2</span></div><div class="line"><a name="l04475"></a><span class="lineno"> 4475</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 3</span></div><div class="line"><a name="l04476"></a><span class="lineno"> 4476</span>&#160; acc3 = fma(b0, (float8)a3.s2, acc3);</div><div class="line"><a name="l04477"></a><span class="lineno"> 4477</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 3</span></div><div class="line"><a name="l04478"></a><span class="lineno"> 4478</span>&#160;</div><div class="line"><a name="l04479"></a><span class="lineno"> 4479</span>&#160; b0 = convert_float8(vload8(0, (__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)(src1_ptr + src_addr.s1)));</div><div class="line"><a name="l04480"></a><span class="lineno"> 4480</span>&#160; src_addr.s1 += src1_stride_y;</div><div class="line"><a name="l04481"></a><span class="lineno"> 4481</span>&#160; acc0 = fma(b0, (float8)a0.s3, acc0);</div><div class="line"><a name="l04482"></a><span class="lineno"> 4482</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 1</span></div><div class="line"><a name="l04483"></a><span class="lineno"> 4483</span>&#160; acc1 = fma(b0, (float8)a1.s3, acc1);</div><div class="line"><a name="l04484"></a><span class="lineno"> 4484</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 1</span></div><div class="line"><a name="l04485"></a><span class="lineno"> 4485</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 2</span></div><div class="line"><a name="l04486"></a><span class="lineno"> 4486</span>&#160; acc2 = fma(b0, (float8)a2.s3, acc2);</div><div class="line"><a name="l04487"></a><span class="lineno"> 4487</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 2</span></div><div class="line"><a name="l04488"></a><span class="lineno"> 4488</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 3</span></div><div class="line"><a name="l04489"></a><span class="lineno"> 4489</span>&#160; acc3 = fma(b0, (float8)a3.s3, acc3);</div><div class="line"><a name="l04490"></a><span class="lineno"> 4490</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 3</span></div><div class="line"><a name="l04491"></a><span class="lineno"> 4491</span>&#160;</div><div class="line"><a name="l04492"></a><span class="lineno"> 4492</span>&#160; src_addr.s0 += 4 * <span class="keyword">sizeof</span>(<a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a>);</div><div class="line"><a name="l04493"></a><span class="lineno"> 4493</span>&#160; }</div><div class="line"><a name="l04494"></a><span class="lineno"> 4494</span>&#160;</div><div class="line"><a name="l04495"></a><span class="lineno"> 4495</span>&#160; <span class="keywordflow">for</span>(; i &lt; (int)COLS_A; ++i)</div><div class="line"><a name="l04496"></a><span class="lineno"> 4496</span>&#160; {</div><div class="line"><a name="l04497"></a><span class="lineno"> 4497</span>&#160;<span class="preprocessor">#if defined(REINTERPRET_INPUT_AS_3D)</span></div><div class="line"><a name="l04498"></a><span class="lineno"> 4498</span>&#160; <span class="comment">// Load values from matrix A</span></div><div class="line"><a name="l04499"></a><span class="lineno"> 4499</span>&#160; <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> a0 = *((__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)(src0_ptr + src_addr.s0 + 0 * src0_stride_y + zin.s0));</div><div class="line"><a name="l04500"></a><span class="lineno"> 4500</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 1</span></div><div class="line"><a name="l04501"></a><span class="lineno"> 4501</span>&#160; <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> a1 = *((__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)(src0_ptr + src_addr.s0 + 1 * src0_stride_y + zin.s1));</div><div class="line"><a name="l04502"></a><span class="lineno"> 4502</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 1</span></div><div class="line"><a name="l04503"></a><span class="lineno"> 4503</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 2</span></div><div class="line"><a name="l04504"></a><span class="lineno"> 4504</span>&#160; <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> a2 = *((__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)(src0_ptr + src_addr.s0 + 2 * src0_stride_y + zin.s2));</div><div class="line"><a name="l04505"></a><span class="lineno"> 4505</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 2</span></div><div class="line"><a name="l04506"></a><span class="lineno"> 4506</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 3</span></div><div class="line"><a name="l04507"></a><span class="lineno"> 4507</span>&#160; <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> a3 = *((__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)(src0_ptr + src_addr.s0 + 3 * src0_stride_y + zin.s3));</div><div class="line"><a name="l04508"></a><span class="lineno"> 4508</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 3</span></div><div class="line"><a name="l04509"></a><span class="lineno"> 4509</span>&#160;<span class="preprocessor">#else // defined(REINTERPRET_INPUT_AS_3D)</span></div><div class="line"><a name="l04510"></a><span class="lineno"> 4510</span>&#160; <span class="comment">// Load values from matrix A</span></div><div class="line"><a name="l04511"></a><span class="lineno"> 4511</span>&#160; <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> a0 = *((__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)(src0_ptr + src_addr.s0 + 0 * src0_stride_y));</div><div class="line"><a name="l04512"></a><span class="lineno"> 4512</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 1</span></div><div class="line"><a name="l04513"></a><span class="lineno"> 4513</span>&#160; <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> a1 = *((__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)(src0_ptr + src_addr.s0 + 1 * src0_stride_y));</div><div class="line"><a name="l04514"></a><span class="lineno"> 4514</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 1</span></div><div class="line"><a name="l04515"></a><span class="lineno"> 4515</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 2</span></div><div class="line"><a name="l04516"></a><span class="lineno"> 4516</span>&#160; <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> a2 = *((__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)(src0_ptr + src_addr.s0 + 2 * src0_stride_y));</div><div class="line"><a name="l04517"></a><span class="lineno"> 4517</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 2</span></div><div class="line"><a name="l04518"></a><span class="lineno"> 4518</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 3</span></div><div class="line"><a name="l04519"></a><span class="lineno"> 4519</span>&#160; <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> a3 = *((__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)(src0_ptr + src_addr.s0 + 3 * src0_stride_y));</div><div class="line"><a name="l04520"></a><span class="lineno"> 4520</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 3</span></div><div class="line"><a name="l04521"></a><span class="lineno"> 4521</span>&#160;<span class="preprocessor">#endif // defined(REINTERPRET_INPUT_AS_3D)</span></div><div class="line"><a name="l04522"></a><span class="lineno"> 4522</span>&#160;</div><div class="line"><a name="l04523"></a><span class="lineno"> 4523</span>&#160; <span class="comment">// Load values from matrix B</span></div><div class="line"><a name="l04524"></a><span class="lineno"> 4524</span>&#160; float8 b0 = convert_float8(vload8(0, (__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)(src1_ptr + src_addr.s1)));</div><div class="line"><a name="l04525"></a><span class="lineno"> 4525</span>&#160;</div><div class="line"><a name="l04526"></a><span class="lineno"> 4526</span>&#160; src_addr += (int2)(<span class="keyword">sizeof</span>(<a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a>), src1_stride_y);</div><div class="line"><a name="l04527"></a><span class="lineno"> 4527</span>&#160;</div><div class="line"><a name="l04528"></a><span class="lineno"> 4528</span>&#160; <span class="comment">// Accumulate</span></div><div class="line"><a name="l04529"></a><span class="lineno"> 4529</span>&#160; acc0 = fma(b0, (float8)a0, acc0); <span class="comment">// b0 * (half8)a0;</span></div><div class="line"><a name="l04530"></a><span class="lineno"> 4530</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 1</span></div><div class="line"><a name="l04531"></a><span class="lineno"> 4531</span>&#160; acc1 = fma(b0, (float8)a1, acc1); <span class="comment">// b0 * (half8)a1;</span></div><div class="line"><a name="l04532"></a><span class="lineno"> 4532</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 1</span></div><div class="line"><a name="l04533"></a><span class="lineno"> 4533</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 2</span></div><div class="line"><a name="l04534"></a><span class="lineno"> 4534</span>&#160; acc2 = fma(b0, (float8)a2, acc2); <span class="comment">// b0 * (half8)a2;</span></div><div class="line"><a name="l04535"></a><span class="lineno"> 4535</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 2</span></div><div class="line"><a name="l04536"></a><span class="lineno"> 4536</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 3</span></div><div class="line"><a name="l04537"></a><span class="lineno"> 4537</span>&#160; acc3 = fma(b0, (float8)a3, acc3); <span class="comment">// b0 * (half8)a3;</span></div><div class="line"><a name="l04538"></a><span class="lineno"> 4538</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 3</span></div><div class="line"><a name="l04539"></a><span class="lineno"> 4539</span>&#160; }</div><div class="line"><a name="l04540"></a><span class="lineno"> 4540</span>&#160;</div><div class="line"><a name="l04541"></a><span class="lineno"> 4541</span>&#160; <span class="comment">// Multiply by the weight of matrix-matrix product and store the result</span></div><div class="line"><a name="l04542"></a><span class="lineno"> 4542</span>&#160;<span class="preprocessor">#if defined(ALPHA)</span></div><div class="line"><a name="l04543"></a><span class="lineno"> 4543</span>&#160; half8 hacc0 = convert_half8(acc0) * (half8)ALPHA;</div><div class="line"><a name="l04544"></a><span class="lineno"> 4544</span>&#160;<span class="preprocessor">#else //defined(ALPHA)</span></div><div class="line"><a name="l04545"></a><span class="lineno"> 4545</span>&#160; half8 hacc0 = convert_half8(acc0);</div><div class="line"><a name="l04546"></a><span class="lineno"> 4546</span>&#160;<span class="preprocessor">#endif // defined(ALPHA)</span></div><div class="line"><a name="l04547"></a><span class="lineno"> 4547</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 1</span></div><div class="line"><a name="l04548"></a><span class="lineno"> 4548</span>&#160;<span class="preprocessor">#if defined(ALPHA)</span></div><div class="line"><a name="l04549"></a><span class="lineno"> 4549</span>&#160; half8 hacc1 = convert_half8(acc1) * (half8)ALPHA;</div><div class="line"><a name="l04550"></a><span class="lineno"> 4550</span>&#160;<span class="preprocessor">#else //defined(ALPHA)</span></div><div class="line"><a name="l04551"></a><span class="lineno"> 4551</span>&#160; half8 hacc1 = convert_half8(acc1);</div><div class="line"><a name="l04552"></a><span class="lineno"> 4552</span>&#160;<span class="preprocessor">#endif //defined(ALPHA)</span></div><div class="line"><a name="l04553"></a><span class="lineno"> 4553</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y</span></div><div class="line"><a name="l04554"></a><span class="lineno"> 4554</span>&#160;</div><div class="line"><a name="l04555"></a><span class="lineno"> 4555</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 2</span></div><div class="line"><a name="l04556"></a><span class="lineno"> 4556</span>&#160;<span class="preprocessor">#if defined(ALPHA)</span></div><div class="line"><a name="l04557"></a><span class="lineno"> 4557</span>&#160; half8 hacc2 = convert_half8(acc2) * (half8)ALPHA;</div><div class="line"><a name="l04558"></a><span class="lineno"> 4558</span>&#160;<span class="preprocessor">#else //defined(ALPHA)</span></div><div class="line"><a name="l04559"></a><span class="lineno"> 4559</span>&#160; half8 hacc2 = convert_half8(acc2);</div><div class="line"><a name="l04560"></a><span class="lineno"> 4560</span>&#160;<span class="preprocessor">#endif //defined(ALPHA)</span></div><div class="line"><a name="l04561"></a><span class="lineno"> 4561</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 2</span></div><div class="line"><a name="l04562"></a><span class="lineno"> 4562</span>&#160;</div><div class="line"><a name="l04563"></a><span class="lineno"> 4563</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 3</span></div><div class="line"><a name="l04564"></a><span class="lineno"> 4564</span>&#160;<span class="preprocessor">#if defined(ALPHA)</span></div><div class="line"><a name="l04565"></a><span class="lineno"> 4565</span>&#160; half8 hacc3 = convert_half8(acc3) * (half8)ALPHA;</div><div class="line"><a name="l04566"></a><span class="lineno"> 4566</span>&#160;<span class="preprocessor">#else //defined(ALPHA)</span></div><div class="line"><a name="l04567"></a><span class="lineno"> 4567</span>&#160; half8 hacc3 = convert_half8(acc3);</div><div class="line"><a name="l04568"></a><span class="lineno"> 4568</span>&#160;<span class="preprocessor">#endif // defined(ALPHA)</span></div><div class="line"><a name="l04569"></a><span class="lineno"> 4569</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 3</span></div><div class="line"><a name="l04570"></a><span class="lineno"> 4570</span>&#160;</div><div class="line"><a name="l04571"></a><span class="lineno"> 4571</span>&#160;<span class="preprocessor">#if defined(ADD_VEC_C)</span></div><div class="line"><a name="l04572"></a><span class="lineno"> 4572</span>&#160; <span class="comment">// *INDENT-OFF*</span></div><div class="line"><a name="l04573"></a><span class="lineno"> 4573</span>&#160; <span class="comment">// clang-format off</span></div><div class="line"><a name="l04574"></a><span class="lineno"> 4574</span>&#160; __global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *src2_addr = (__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)(src2_ptr + src2_offset_first_element_in_bytes + get_global_id(0) * src2_step_x);</div><div class="line"><a name="l04575"></a><span class="lineno"> 4575</span>&#160; half8 c0 = vload8(0, src2_addr);</div><div class="line"><a name="l04576"></a><span class="lineno"> 4576</span>&#160; <span class="comment">// clang-format on</span></div><div class="line"><a name="l04577"></a><span class="lineno"> 4577</span>&#160; <span class="comment">// *INDENT-ON*</span></div><div class="line"><a name="l04578"></a><span class="lineno"> 4578</span>&#160;</div><div class="line"><a name="l04579"></a><span class="lineno"> 4579</span>&#160; hacc0 += c0;</div><div class="line"><a name="l04580"></a><span class="lineno"> 4580</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 1</span></div><div class="line"><a name="l04581"></a><span class="lineno"> 4581</span>&#160; hacc1 += c0;</div><div class="line"><a name="l04582"></a><span class="lineno"> 4582</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 1</span></div><div class="line"><a name="l04583"></a><span class="lineno"> 4583</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 2</span></div><div class="line"><a name="l04584"></a><span class="lineno"> 4584</span>&#160; hacc2 += c0;</div><div class="line"><a name="l04585"></a><span class="lineno"> 4585</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 2</span></div><div class="line"><a name="l04586"></a><span class="lineno"> 4586</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 3</span></div><div class="line"><a name="l04587"></a><span class="lineno"> 4587</span>&#160; hacc3 += c0;</div><div class="line"><a name="l04588"></a><span class="lineno"> 4588</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 3</span></div><div class="line"><a name="l04589"></a><span class="lineno"> 4589</span>&#160;<span class="preprocessor">#endif </span><span class="comment">/* defined(ADD_VEC_C) */</span><span class="preprocessor"></span></div><div class="line"><a name="l04590"></a><span class="lineno"> 4590</span>&#160;</div><div class="line"><a name="l04591"></a><span class="lineno"> 4591</span>&#160; <span class="keywordtype">int</span> z = get_global_id(2);</div><div class="line"><a name="l04592"></a><span class="lineno"> 4592</span>&#160;</div><div class="line"><a name="l04593"></a><span class="lineno"> 4593</span>&#160; <span class="comment">// Compute destination address</span></div><div class="line"><a name="l04594"></a><span class="lineno"> 4594</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="l04595"></a><span class="lineno"> 4595</span>&#160;</div><div class="line"><a name="l04596"></a><span class="lineno"> 4596</span>&#160; <span class="comment">// Compute dst address</span></div><div class="line"><a name="l04597"></a><span class="lineno"> 4597</span>&#160; __global uchar *dst_addr = <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>(&amp;<a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#adbf67dcee294e673cf796f1ed8aeb6a4">dst</a>, 0, 0);</div><div class="line"><a name="l04598"></a><span class="lineno"> 4598</span>&#160;</div><div class="line"><a name="l04599"></a><span class="lineno"> 4599</span>&#160;<span class="preprocessor">#if defined(REINTERPRET_OUTPUT_AS_3D)</span></div><div class="line"><a name="l04600"></a><span class="lineno"> 4600</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="l04601"></a><span class="lineno"> 4601</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="l04602"></a><span class="lineno"> 4602</span>&#160; <span class="comment">//</span></div><div class="line"><a name="l04603"></a><span class="lineno"> 4603</span>&#160; <span class="comment">// | |</span></div><div class="line"><a name="l04604"></a><span class="lineno"> 4604</span>&#160; <span class="comment">// | plane0 |</span></div><div class="line"><a name="l04605"></a><span class="lineno"> 4605</span>&#160; <span class="comment">// | |</span></div><div class="line"><a name="l04606"></a><span class="lineno"> 4606</span>&#160; <span class="comment">// |__________________|</span></div><div class="line"><a name="l04607"></a><span class="lineno"> 4607</span>&#160; <span class="comment">// |******************|</span></div><div class="line"><a name="l04608"></a><span class="lineno"> 4608</span>&#160; <span class="comment">// | cross_plane_pad |</span></div><div class="line"><a name="l04609"></a><span class="lineno"> 4609</span>&#160; <span class="comment">// |******************|</span></div><div class="line"><a name="l04610"></a><span class="lineno"> 4610</span>&#160; <span class="comment">// | |</span></div><div class="line"><a name="l04611"></a><span class="lineno"> 4611</span>&#160; <span class="comment">// | plane1 |</span></div><div class="line"><a name="l04612"></a><span class="lineno"> 4612</span>&#160; <span class="comment">// | |</span></div><div class="line"><a name="l04613"></a><span class="lineno"> 4613</span>&#160; <span class="comment">// |__________________|</span></div><div class="line"><a name="l04614"></a><span class="lineno"> 4614</span>&#160;</div><div class="line"><a name="l04615"></a><span class="lineno"> 4615</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="l04616"></a><span class="lineno"> 4616</span>&#160; uint4 zout = ((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="l04617"></a><span class="lineno"> 4617</span>&#160; zout = min(DEPTH_GEMM3D - 1, zout);</div><div class="line"><a name="l04618"></a><span class="lineno"> 4618</span>&#160;</div><div class="line"><a name="l04619"></a><span class="lineno"> 4619</span>&#160; <span class="comment">// Add offset due to the cross plane paddings</span></div><div class="line"><a name="l04620"></a><span class="lineno"> 4620</span>&#160; zout *= (dst_cross_plane_pad * dst_stride_y);</div><div class="line"><a name="l04621"></a><span class="lineno"> 4621</span>&#160;</div><div class="line"><a name="l04622"></a><span class="lineno"> 4622</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="l04623"></a><span class="lineno"> 4623</span>&#160; <span class="comment">// multiply dst_stride_z by DEPTH_GEMM3D</span></div><div class="line"><a name="l04624"></a><span class="lineno"> 4624</span>&#160; dst_addr += z * dst_stride_z * DEPTH_GEMM3D;</div><div class="line"><a name="l04625"></a><span class="lineno"> 4625</span>&#160;</div><div class="line"><a name="l04626"></a><span class="lineno"> 4626</span>&#160; <span class="comment">// Store the output block</span></div><div class="line"><a name="l04627"></a><span class="lineno"> 4627</span>&#160; vstore8(hacc0, 0, (__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)(dst_addr + 0 * dst_stride_y + zout.s0));</div><div class="line"><a name="l04628"></a><span class="lineno"> 4628</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 1</span></div><div class="line"><a name="l04629"></a><span class="lineno"> 4629</span>&#160; vstore8(hacc1, 0, (__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)(dst_addr + 1 * dst_stride_y + zout.s1));</div><div class="line"><a name="l04630"></a><span class="lineno"> 4630</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 1</span></div><div class="line"><a name="l04631"></a><span class="lineno"> 4631</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 2</span></div><div class="line"><a name="l04632"></a><span class="lineno"> 4632</span>&#160; vstore8(hacc2, 0, (__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)(dst_addr + 2 * dst_stride_y + zout.s2));</div><div class="line"><a name="l04633"></a><span class="lineno"> 4633</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 2</span></div><div class="line"><a name="l04634"></a><span class="lineno"> 4634</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 3</span></div><div class="line"><a name="l04635"></a><span class="lineno"> 4635</span>&#160; vstore8(hacc3, 0, (__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)(dst_addr + 3 * dst_stride_y + zout.s3));</div><div class="line"><a name="l04636"></a><span class="lineno"> 4636</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 3</span></div><div class="line"><a name="l04637"></a><span class="lineno"> 4637</span>&#160;</div><div class="line"><a name="l04638"></a><span class="lineno"> 4638</span>&#160;<span class="preprocessor">#else // defined(REINTERPRET_OUTPUT_AS_3D)</span></div><div class="line"><a name="l04639"></a><span class="lineno"> 4639</span>&#160; <span class="comment">// Add offset for batched GEMM</span></div><div class="line"><a name="l04640"></a><span class="lineno"> 4640</span>&#160; dst_addr += z * dst_stride_z;</div><div class="line"><a name="l04641"></a><span class="lineno"> 4641</span>&#160;</div><div class="line"><a name="l04642"></a><span class="lineno"> 4642</span>&#160; <span class="comment">// Store the output block</span></div><div class="line"><a name="l04643"></a><span class="lineno"> 4643</span>&#160; vstore8(hacc0, 0, (__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)(dst_addr + 0 * dst_stride_y));</div><div class="line"><a name="l04644"></a><span class="lineno"> 4644</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 1</span></div><div class="line"><a name="l04645"></a><span class="lineno"> 4645</span>&#160; vstore8(hacc1, 0, (__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)(dst_addr + 1 * dst_stride_y));</div><div class="line"><a name="l04646"></a><span class="lineno"> 4646</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 1</span></div><div class="line"><a name="l04647"></a><span class="lineno"> 4647</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 2</span></div><div class="line"><a name="l04648"></a><span class="lineno"> 4648</span>&#160; vstore8(hacc2, 0, (__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)(dst_addr + 2 * dst_stride_y));</div><div class="line"><a name="l04649"></a><span class="lineno"> 4649</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 2</span></div><div class="line"><a name="l04650"></a><span class="lineno"> 4650</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 3</span></div><div class="line"><a name="l04651"></a><span class="lineno"> 4651</span>&#160; vstore8(hacc3, 0, (__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)(dst_addr + 3 * dst_stride_y));</div><div class="line"><a name="l04652"></a><span class="lineno"> 4652</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 3</span></div><div class="line"><a name="l04653"></a><span class="lineno"> 4653</span>&#160;<span class="preprocessor">#endif // REINTERPRET_OUTPUT_AS_3D</span></div><div class="line"><a name="l04654"></a><span class="lineno"> 4654</span>&#160;}</div><div class="line"><a name="l04655"></a><span class="lineno"> 4655</span>&#160;</div><div class="line"><a name="l04705"></a><span class="lineno"> 4705</span>&#160;__kernel <span class="keywordtype">void</span> gemm_mm_floating_point_f16_bifrost(<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="l04706"></a><span class="lineno"> 4706</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="l04707"></a><span class="lineno"> 4707</span>&#160;#<span class="keywordflow">if</span> defined(ADD_VEC_C)</div><div class="line"><a name="l04708"></a><span class="lineno"> 4708</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a40a6eb9f2a7712f08d6bb8ff6c9e6ca7">VECTOR_DECLARATION</a>(src2),</div><div class="line"><a name="l04709"></a><span class="lineno"> 4709</span>&#160;#endif <span class="comment">/* defined(ADD_VEC_C) */</span></div><div class="line"><a name="l04710"></a><span class="lineno"> 4710</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="l04711"></a><span class="lineno"> 4711</span>&#160; uint src0_stride_z,</div><div class="line"><a name="l04712"></a><span class="lineno"> 4712</span>&#160; uint src1_stride_z,</div><div class="line"><a name="l04713"></a><span class="lineno"> 4713</span>&#160; uint dst_stride_z</div><div class="line"><a name="l04714"></a><span class="lineno"> 4714</span>&#160;#<span class="keywordflow">if</span> defined(REINTERPRET_INPUT_AS_3D)</div><div class="line"><a name="l04715"></a><span class="lineno"> 4715</span>&#160; ,</div><div class="line"><a name="l04716"></a><span class="lineno"> 4716</span>&#160; uint src_cross_plane_pad</div><div class="line"><a name="l04717"></a><span class="lineno"> 4717</span>&#160;#endif <span class="comment">// REINTERPRET_INPUT_AS_3D</span></div><div class="line"><a name="l04718"></a><span class="lineno"> 4718</span>&#160;#<span class="keywordflow">if</span> defined(REINTERPRET_OUTPUT_AS_3D)</div><div class="line"><a name="l04719"></a><span class="lineno"> 4719</span>&#160; ,</div><div class="line"><a name="l04720"></a><span class="lineno"> 4720</span>&#160; uint dst_cross_plane_pad</div><div class="line"><a name="l04721"></a><span class="lineno"> 4721</span>&#160;#endif <span class="comment">// REINTERPRET_OUTPUT_AS_3D</span></div><div class="line"><a name="l04722"></a><span class="lineno"> 4722</span>&#160; )</div><div class="line"><a name="l04723"></a><span class="lineno"> 4723</span>&#160;{</div><div class="line"><a name="l04724"></a><span class="lineno"> 4724</span>&#160; <span class="keywordtype">int</span> idx = get_global_id(0) * NUM_ELEMS_PROCESSED_PER_THREAD_X;</div><div class="line"><a name="l04725"></a><span class="lineno"> 4725</span>&#160;</div><div class="line"><a name="l04726"></a><span class="lineno"> 4726</span>&#160; <span class="comment">// Compute starting address for matrix A and Matrix B</span></div><div class="line"><a name="l04727"></a><span class="lineno"> 4727</span>&#160; int2 src_addr = ((int2)(src0_offset_first_element_in_bytes, src1_offset_first_element_in_bytes));</div><div class="line"><a name="l04728"></a><span class="lineno"> 4728</span>&#160;</div><div class="line"><a name="l04729"></a><span class="lineno"> 4729</span>&#160; <span class="comment">// Update address for the matrix A</span></div><div class="line"><a name="l04730"></a><span class="lineno"> 4730</span>&#160; src_addr.s0 += get_global_id(1) * src0_stride_y * NUM_ELEMS_PROCESSED_PER_THREAD_Y;</div><div class="line"><a name="l04731"></a><span class="lineno"> 4731</span>&#160;</div><div class="line"><a name="l04732"></a><span class="lineno"> 4732</span>&#160; <span class="comment">// Update address for the matrix B</span></div><div class="line"><a name="l04733"></a><span class="lineno"> 4733</span>&#160; src_addr.s1 += idx * <span class="keyword">sizeof</span>(<a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a>);</div><div class="line"><a name="l04734"></a><span class="lineno"> 4734</span>&#160;</div><div class="line"><a name="l04735"></a><span class="lineno"> 4735</span>&#160;<span class="preprocessor">#if defined(REINTERPRET_INPUT_AS_3D)</span></div><div class="line"><a name="l04736"></a><span class="lineno"> 4736</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="l04737"></a><span class="lineno"> 4737</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="l04738"></a><span class="lineno"> 4738</span>&#160; <span class="comment">//</span></div><div class="line"><a name="l04739"></a><span class="lineno"> 4739</span>&#160; <span class="comment">// | |</span></div><div class="line"><a name="l04740"></a><span class="lineno"> 4740</span>&#160; <span class="comment">// | plane0 |</span></div><div class="line"><a name="l04741"></a><span class="lineno"> 4741</span>&#160; <span class="comment">// | |</span></div><div class="line"><a name="l04742"></a><span class="lineno"> 4742</span>&#160; <span class="comment">// |__________________|</span></div><div class="line"><a name="l04743"></a><span class="lineno"> 4743</span>&#160; <span class="comment">// |******************|</span></div><div class="line"><a name="l04744"></a><span class="lineno"> 4744</span>&#160; <span class="comment">// | cross_plane_pad |</span></div><div class="line"><a name="l04745"></a><span class="lineno"> 4745</span>&#160; <span class="comment">// |******************|</span></div><div class="line"><a name="l04746"></a><span class="lineno"> 4746</span>&#160; <span class="comment">// | |</span></div><div class="line"><a name="l04747"></a><span class="lineno"> 4747</span>&#160; <span class="comment">// | plane1 |</span></div><div class="line"><a name="l04748"></a><span class="lineno"> 4748</span>&#160; <span class="comment">// | |</span></div><div class="line"><a name="l04749"></a><span class="lineno"> 4749</span>&#160; <span class="comment">// |__________________|</span></div><div class="line"><a name="l04750"></a><span class="lineno"> 4750</span>&#160;</div><div class="line"><a name="l04751"></a><span class="lineno"> 4751</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="l04752"></a><span class="lineno"> 4752</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="l04753"></a><span class="lineno"> 4753</span>&#160; zin = min(DEPTH_GEMM3D - 1, zin);</div><div class="line"><a name="l04754"></a><span class="lineno"> 4754</span>&#160;</div><div class="line"><a name="l04755"></a><span class="lineno"> 4755</span>&#160; <span class="comment">// Add offset due to the cross plane paddings</span></div><div class="line"><a name="l04756"></a><span class="lineno"> 4756</span>&#160; zin *= (src_cross_plane_pad * src0_stride_y);</div><div class="line"><a name="l04757"></a><span class="lineno"> 4757</span>&#160;</div><div class="line"><a name="l04758"></a><span class="lineno"> 4758</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="l04759"></a><span class="lineno"> 4759</span>&#160; <span class="comment">// multiply src0_stride_z by DEPTH_GEMM3D</span></div><div class="line"><a name="l04760"></a><span class="lineno"> 4760</span>&#160; src_addr.s0 += get_global_id(2) * src0_stride_z * DEPTH_GEMM3D;</div><div class="line"><a name="l04761"></a><span class="lineno"> 4761</span>&#160;</div><div class="line"><a name="l04762"></a><span class="lineno"> 4762</span>&#160;<span class="preprocessor">#else // defined(REINTERPRET_INPUT_AS_3D)</span></div><div class="line"><a name="l04763"></a><span class="lineno"> 4763</span>&#160;</div><div class="line"><a name="l04764"></a><span class="lineno"> 4764</span>&#160; <span class="comment">// Add offset for batched GEMM</span></div><div class="line"><a name="l04765"></a><span class="lineno"> 4765</span>&#160; src_addr.s0 += get_global_id(2) * src0_stride_z;</div><div class="line"><a name="l04766"></a><span class="lineno"> 4766</span>&#160;</div><div class="line"><a name="l04767"></a><span class="lineno"> 4767</span>&#160;<span class="preprocessor">#endif // defined(REINTERPRET_INPUT_AS_3D)</span></div><div class="line"><a name="l04768"></a><span class="lineno"> 4768</span>&#160;</div><div class="line"><a name="l04769"></a><span class="lineno"> 4769</span>&#160;<span class="preprocessor">#if defined(MATRIX_B_DEPTH)</span></div><div class="line"><a name="l04770"></a><span class="lineno"> 4770</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="l04771"></a><span class="lineno"> 4771</span>&#160; src_addr.s1 += (get_global_id(2) % MATRIX_B_DEPTH) * src1_stride_z;</div><div class="line"><a name="l04772"></a><span class="lineno"> 4772</span>&#160;<span class="preprocessor">#else // defined(MATRIX_B_DEPTH)</span></div><div class="line"><a name="l04773"></a><span class="lineno"> 4773</span>&#160; src_addr.s1 += get_global_id(2) * src1_stride_z;</div><div class="line"><a name="l04774"></a><span class="lineno"> 4774</span>&#160;<span class="preprocessor">#endif // defined(MATRIX_B_DEPTH)</span></div><div class="line"><a name="l04775"></a><span class="lineno"> 4775</span>&#160;</div><div class="line"><a name="l04776"></a><span class="lineno"> 4776</span>&#160; half8 acc0 = 0.0h;</div><div class="line"><a name="l04777"></a><span class="lineno"> 4777</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 1</span></div><div class="line"><a name="l04778"></a><span class="lineno"> 4778</span>&#160; half8 acc1 = 0.0h;</div><div class="line"><a name="l04779"></a><span class="lineno"> 4779</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 1</span></div><div class="line"><a name="l04780"></a><span class="lineno"> 4780</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 2</span></div><div class="line"><a name="l04781"></a><span class="lineno"> 4781</span>&#160; half8 acc2 = 0.0h;</div><div class="line"><a name="l04782"></a><span class="lineno"> 4782</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 2</span></div><div class="line"><a name="l04783"></a><span class="lineno"> 4783</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 3</span></div><div class="line"><a name="l04784"></a><span class="lineno"> 4784</span>&#160; half8 acc3 = 0.0h;</div><div class="line"><a name="l04785"></a><span class="lineno"> 4785</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 3</span></div><div class="line"><a name="l04786"></a><span class="lineno"> 4786</span>&#160;</div><div class="line"><a name="l04787"></a><span class="lineno"> 4787</span>&#160; <span class="keywordtype">int</span> i = 0;</div><div class="line"><a name="l04788"></a><span class="lineno"> 4788</span>&#160; <span class="keywordflow">for</span>(; i &lt;= ((int)COLS_A - 4); i += 4)</div><div class="line"><a name="l04789"></a><span class="lineno"> 4789</span>&#160; {</div><div class="line"><a name="l04790"></a><span class="lineno"> 4790</span>&#160;<span class="preprocessor">#if defined(REINTERPRET_INPUT_AS_3D)</span></div><div class="line"><a name="l04791"></a><span class="lineno"> 4791</span>&#160; <span class="comment">// Load values from matrix A</span></div><div class="line"><a name="l04792"></a><span class="lineno"> 4792</span>&#160; half4 a0 = vload4(0, (__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)(src0_ptr + src_addr.s0 + 0 * src0_stride_y + zin.s0));</div><div class="line"><a name="l04793"></a><span class="lineno"> 4793</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 1</span></div><div class="line"><a name="l04794"></a><span class="lineno"> 4794</span>&#160; half4 a1 = vload4(0, (__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)(src0_ptr + src_addr.s0 + 1 * src0_stride_y + zin.s1));</div><div class="line"><a name="l04795"></a><span class="lineno"> 4795</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 1</span></div><div class="line"><a name="l04796"></a><span class="lineno"> 4796</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 2</span></div><div class="line"><a name="l04797"></a><span class="lineno"> 4797</span>&#160; half4 a2 = vload4(0, (__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)(src0_ptr + src_addr.s0 + 2 * src0_stride_y + zin.s2));</div><div class="line"><a name="l04798"></a><span class="lineno"> 4798</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 2</span></div><div class="line"><a name="l04799"></a><span class="lineno"> 4799</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 3</span></div><div class="line"><a name="l04800"></a><span class="lineno"> 4800</span>&#160; half4 a3 = vload4(0, (__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)(src0_ptr + src_addr.s0 + 3 * src0_stride_y + zin.s3));</div><div class="line"><a name="l04801"></a><span class="lineno"> 4801</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 3</span></div><div class="line"><a name="l04802"></a><span class="lineno"> 4802</span>&#160;<span class="preprocessor">#else // defined(REINTERPRET_INPUT_AS_3D)</span></div><div class="line"><a name="l04803"></a><span class="lineno"> 4803</span>&#160; <span class="comment">// Load values from matrix A</span></div><div class="line"><a name="l04804"></a><span class="lineno"> 4804</span>&#160; half4 a0 = vload4(0, (__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)(src0_ptr + src_addr.s0 + 0 * src0_stride_y));</div><div class="line"><a name="l04805"></a><span class="lineno"> 4805</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 1</span></div><div class="line"><a name="l04806"></a><span class="lineno"> 4806</span>&#160; half4 a1 = vload4(0, (__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)(src0_ptr + src_addr.s0 + 1 * src0_stride_y));</div><div class="line"><a name="l04807"></a><span class="lineno"> 4807</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 1</span></div><div class="line"><a name="l04808"></a><span class="lineno"> 4808</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 2</span></div><div class="line"><a name="l04809"></a><span class="lineno"> 4809</span>&#160; half4 a2 = vload4(0, (__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)(src0_ptr + src_addr.s0 + 2 * src0_stride_y));</div><div class="line"><a name="l04810"></a><span class="lineno"> 4810</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 2</span></div><div class="line"><a name="l04811"></a><span class="lineno"> 4811</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 3</span></div><div class="line"><a name="l04812"></a><span class="lineno"> 4812</span>&#160; half4 a3 = vload4(0, (__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)(src0_ptr + src_addr.s0 + 3 * src0_stride_y));</div><div class="line"><a name="l04813"></a><span class="lineno"> 4813</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 3</span></div><div class="line"><a name="l04814"></a><span class="lineno"> 4814</span>&#160;<span class="preprocessor">#endif // defined(REINTERPRET_INPUT_AS_3D)</span></div><div class="line"><a name="l04815"></a><span class="lineno"> 4815</span>&#160;</div><div class="line"><a name="l04816"></a><span class="lineno"> 4816</span>&#160; <span class="comment">// Load values from matrix B</span></div><div class="line"><a name="l04817"></a><span class="lineno"> 4817</span>&#160; half8 b0 = vload8(0, (__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)(src1_ptr + src_addr.s1));</div><div class="line"><a name="l04818"></a><span class="lineno"> 4818</span>&#160; src_addr.s1 += src1_stride_y;</div><div class="line"><a name="l04819"></a><span class="lineno"> 4819</span>&#160;</div><div class="line"><a name="l04820"></a><span class="lineno"> 4820</span>&#160; <span class="comment">// Accumulate</span></div><div class="line"><a name="l04821"></a><span class="lineno"> 4821</span>&#160; acc0 = fma(b0, (half8)a0.s0, acc0);</div><div class="line"><a name="l04822"></a><span class="lineno"> 4822</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 1</span></div><div class="line"><a name="l04823"></a><span class="lineno"> 4823</span>&#160; acc1 = fma(b0, (half8)a1.s0, acc1);</div><div class="line"><a name="l04824"></a><span class="lineno"> 4824</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 1</span></div><div class="line"><a name="l04825"></a><span class="lineno"> 4825</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 2</span></div><div class="line"><a name="l04826"></a><span class="lineno"> 4826</span>&#160; acc2 = fma(b0, (half8)a2.s0, acc2);</div><div class="line"><a name="l04827"></a><span class="lineno"> 4827</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 2</span></div><div class="line"><a name="l04828"></a><span class="lineno"> 4828</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 3</span></div><div class="line"><a name="l04829"></a><span class="lineno"> 4829</span>&#160; acc3 = fma(b0, (half8)a3.s0, acc3);</div><div class="line"><a name="l04830"></a><span class="lineno"> 4830</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 3</span></div><div class="line"><a name="l04831"></a><span class="lineno"> 4831</span>&#160;</div><div class="line"><a name="l04832"></a><span class="lineno"> 4832</span>&#160; b0 = vload8(0, (__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)(src1_ptr + src_addr.s1));</div><div class="line"><a name="l04833"></a><span class="lineno"> 4833</span>&#160; src_addr.s1 += src1_stride_y;</div><div class="line"><a name="l04834"></a><span class="lineno"> 4834</span>&#160; acc0 = fma(b0, (half8)a0.s1, acc0);</div><div class="line"><a name="l04835"></a><span class="lineno"> 4835</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 1</span></div><div class="line"><a name="l04836"></a><span class="lineno"> 4836</span>&#160; acc1 = fma(b0, (half8)a1.s1, acc1);</div><div class="line"><a name="l04837"></a><span class="lineno"> 4837</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 1</span></div><div class="line"><a name="l04838"></a><span class="lineno"> 4838</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 2</span></div><div class="line"><a name="l04839"></a><span class="lineno"> 4839</span>&#160; acc2 = fma(b0, (half8)a2.s1, acc2);</div><div class="line"><a name="l04840"></a><span class="lineno"> 4840</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 2</span></div><div class="line"><a name="l04841"></a><span class="lineno"> 4841</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 3</span></div><div class="line"><a name="l04842"></a><span class="lineno"> 4842</span>&#160; acc3 = fma(b0, (half8)a3.s1, acc3);</div><div class="line"><a name="l04843"></a><span class="lineno"> 4843</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 3</span></div><div class="line"><a name="l04844"></a><span class="lineno"> 4844</span>&#160;</div><div class="line"><a name="l04845"></a><span class="lineno"> 4845</span>&#160; b0 = vload8(0, (__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)(src1_ptr + src_addr.s1));</div><div class="line"><a name="l04846"></a><span class="lineno"> 4846</span>&#160; src_addr.s1 += src1_stride_y;</div><div class="line"><a name="l04847"></a><span class="lineno"> 4847</span>&#160; acc0 = fma(b0, (half8)a0.s2, acc0);</div><div class="line"><a name="l04848"></a><span class="lineno"> 4848</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 1</span></div><div class="line"><a name="l04849"></a><span class="lineno"> 4849</span>&#160; acc1 = fma(b0, (half8)a1.s2, acc1);</div><div class="line"><a name="l04850"></a><span class="lineno"> 4850</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 1</span></div><div class="line"><a name="l04851"></a><span class="lineno"> 4851</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 2</span></div><div class="line"><a name="l04852"></a><span class="lineno"> 4852</span>&#160; acc2 = fma(b0, (half8)a2.s2, acc2);</div><div class="line"><a name="l04853"></a><span class="lineno"> 4853</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 2</span></div><div class="line"><a name="l04854"></a><span class="lineno"> 4854</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 3</span></div><div class="line"><a name="l04855"></a><span class="lineno"> 4855</span>&#160; acc3 = fma(b0, (half8)a3.s2, acc3);</div><div class="line"><a name="l04856"></a><span class="lineno"> 4856</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 3</span></div><div class="line"><a name="l04857"></a><span class="lineno"> 4857</span>&#160;</div><div class="line"><a name="l04858"></a><span class="lineno"> 4858</span>&#160; b0 = vload8(0, (__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)(src1_ptr + src_addr.s1));</div><div class="line"><a name="l04859"></a><span class="lineno"> 4859</span>&#160; src_addr.s1 += src1_stride_y;</div><div class="line"><a name="l04860"></a><span class="lineno"> 4860</span>&#160; acc0 = fma(b0, (half8)a0.s3, acc0);</div><div class="line"><a name="l04861"></a><span class="lineno"> 4861</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 1</span></div><div class="line"><a name="l04862"></a><span class="lineno"> 4862</span>&#160; acc1 = fma(b0, (half8)a1.s3, acc1);</div><div class="line"><a name="l04863"></a><span class="lineno"> 4863</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 1</span></div><div class="line"><a name="l04864"></a><span class="lineno"> 4864</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 2</span></div><div class="line"><a name="l04865"></a><span class="lineno"> 4865</span>&#160; acc2 = fma(b0, (half8)a2.s3, acc2);</div><div class="line"><a name="l04866"></a><span class="lineno"> 4866</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 2</span></div><div class="line"><a name="l04867"></a><span class="lineno"> 4867</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 3</span></div><div class="line"><a name="l04868"></a><span class="lineno"> 4868</span>&#160; acc3 = fma(b0, (half8)a3.s3, acc3);</div><div class="line"><a name="l04869"></a><span class="lineno"> 4869</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 3</span></div><div class="line"><a name="l04870"></a><span class="lineno"> 4870</span>&#160;</div><div class="line"><a name="l04871"></a><span class="lineno"> 4871</span>&#160; src_addr.s0 += 4 * <span class="keyword">sizeof</span>(<a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a>);</div><div class="line"><a name="l04872"></a><span class="lineno"> 4872</span>&#160; }</div><div class="line"><a name="l04873"></a><span class="lineno"> 4873</span>&#160;</div><div class="line"><a name="l04874"></a><span class="lineno"> 4874</span>&#160; <span class="keywordflow">for</span>(; i &lt; (int)COLS_A; ++i)</div><div class="line"><a name="l04875"></a><span class="lineno"> 4875</span>&#160; {</div><div class="line"><a name="l04876"></a><span class="lineno"> 4876</span>&#160;<span class="preprocessor">#if defined(REINTERPRET_INPUT_AS_3D)</span></div><div class="line"><a name="l04877"></a><span class="lineno"> 4877</span>&#160; <span class="comment">// Load values from matrix A</span></div><div class="line"><a name="l04878"></a><span class="lineno"> 4878</span>&#160; <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> a0 = *((__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)(src0_ptr + src_addr.s0 + 0 * src0_stride_y + zin.s0));</div><div class="line"><a name="l04879"></a><span class="lineno"> 4879</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 1</span></div><div class="line"><a name="l04880"></a><span class="lineno"> 4880</span>&#160; <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> a1 = *((__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)(src0_ptr + src_addr.s0 + 1 * src0_stride_y + zin.s1));</div><div class="line"><a name="l04881"></a><span class="lineno"> 4881</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 1</span></div><div class="line"><a name="l04882"></a><span class="lineno"> 4882</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 2</span></div><div class="line"><a name="l04883"></a><span class="lineno"> 4883</span>&#160; <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> a2 = *((__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)(src0_ptr + src_addr.s0 + 2 * src0_stride_y + zin.s2));</div><div class="line"><a name="l04884"></a><span class="lineno"> 4884</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 2</span></div><div class="line"><a name="l04885"></a><span class="lineno"> 4885</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 3</span></div><div class="line"><a name="l04886"></a><span class="lineno"> 4886</span>&#160; <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> a3 = *((__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)(src0_ptr + src_addr.s0 + 3 * src0_stride_y + zin.s3));</div><div class="line"><a name="l04887"></a><span class="lineno"> 4887</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 3</span></div><div class="line"><a name="l04888"></a><span class="lineno"> 4888</span>&#160;<span class="preprocessor">#else // defined(REINTERPRET_INPUT_AS_3D)</span></div><div class="line"><a name="l04889"></a><span class="lineno"> 4889</span>&#160; <span class="comment">// Load values from matrix A</span></div><div class="line"><a name="l04890"></a><span class="lineno"> 4890</span>&#160; <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> a0 = *((__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)(src0_ptr + src_addr.s0 + 0 * src0_stride_y));</div><div class="line"><a name="l04891"></a><span class="lineno"> 4891</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 1</span></div><div class="line"><a name="l04892"></a><span class="lineno"> 4892</span>&#160; <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> a1 = *((__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)(src0_ptr + src_addr.s0 + 1 * src0_stride_y));</div><div class="line"><a name="l04893"></a><span class="lineno"> 4893</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 1</span></div><div class="line"><a name="l04894"></a><span class="lineno"> 4894</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 2</span></div><div class="line"><a name="l04895"></a><span class="lineno"> 4895</span>&#160; <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> a2 = *((__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)(src0_ptr + src_addr.s0 + 2 * src0_stride_y));</div><div class="line"><a name="l04896"></a><span class="lineno"> 4896</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 2</span></div><div class="line"><a name="l04897"></a><span class="lineno"> 4897</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 3</span></div><div class="line"><a name="l04898"></a><span class="lineno"> 4898</span>&#160; <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> a3 = *((__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)(src0_ptr + src_addr.s0 + 3 * src0_stride_y));</div><div class="line"><a name="l04899"></a><span class="lineno"> 4899</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 3</span></div><div class="line"><a name="l04900"></a><span class="lineno"> 4900</span>&#160;<span class="preprocessor">#endif // defined(REINTERPRET_INPUT_AS_3D)</span></div><div class="line"><a name="l04901"></a><span class="lineno"> 4901</span>&#160;</div><div class="line"><a name="l04902"></a><span class="lineno"> 4902</span>&#160; <span class="comment">// Load values from matrix B</span></div><div class="line"><a name="l04903"></a><span class="lineno"> 4903</span>&#160; half8 b0 = vload8(0, (__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)(src1_ptr + src_addr.s1));</div><div class="line"><a name="l04904"></a><span class="lineno"> 4904</span>&#160;</div><div class="line"><a name="l04905"></a><span class="lineno"> 4905</span>&#160; src_addr += (int2)(<span class="keyword">sizeof</span>(<a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a>), src1_stride_y);</div><div class="line"><a name="l04906"></a><span class="lineno"> 4906</span>&#160;</div><div class="line"><a name="l04907"></a><span class="lineno"> 4907</span>&#160; <span class="comment">// Accumulate</span></div><div class="line"><a name="l04908"></a><span class="lineno"> 4908</span>&#160; acc0 = fma(b0, (half8)a0, acc0); <span class="comment">// b0 * (half8)a0;</span></div><div class="line"><a name="l04909"></a><span class="lineno"> 4909</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 1</span></div><div class="line"><a name="l04910"></a><span class="lineno"> 4910</span>&#160; acc1 = fma(b0, (half8)a1, acc1); <span class="comment">// b0 * (half8)a1;</span></div><div class="line"><a name="l04911"></a><span class="lineno"> 4911</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 1</span></div><div class="line"><a name="l04912"></a><span class="lineno"> 4912</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 2</span></div><div class="line"><a name="l04913"></a><span class="lineno"> 4913</span>&#160; acc2 = fma(b0, (half8)a2, acc2); <span class="comment">// b0 * (half8)a2;</span></div><div class="line"><a name="l04914"></a><span class="lineno"> 4914</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 2</span></div><div class="line"><a name="l04915"></a><span class="lineno"> 4915</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 3</span></div><div class="line"><a name="l04916"></a><span class="lineno"> 4916</span>&#160; acc3 = fma(b0, (half8)a3, acc3); <span class="comment">// b0 * (half8)a3;</span></div><div class="line"><a name="l04917"></a><span class="lineno"> 4917</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 3</span></div><div class="line"><a name="l04918"></a><span class="lineno"> 4918</span>&#160; }</div><div class="line"><a name="l04919"></a><span class="lineno"> 4919</span>&#160;</div><div class="line"><a name="l04920"></a><span class="lineno"> 4920</span>&#160; <span class="comment">// Multiply by the weight of matrix-matrix product and store the result</span></div><div class="line"><a name="l04921"></a><span class="lineno"> 4921</span>&#160;<span class="preprocessor">#if defined(ALPHA)</span></div><div class="line"><a name="l04922"></a><span class="lineno"> 4922</span>&#160; acc0 = acc0 * (half8)ALPHA;</div><div class="line"><a name="l04923"></a><span class="lineno"> 4923</span>&#160;<span class="preprocessor">#endif // defined(ALPHA)</span></div><div class="line"><a name="l04924"></a><span class="lineno"> 4924</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 1 &amp;&amp; defined(ALPHA)</span></div><div class="line"><a name="l04925"></a><span class="lineno"> 4925</span>&#160; acc1 = acc1 * (half8)ALPHA;</div><div class="line"><a name="l04926"></a><span class="lineno"> 4926</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 1 &amp;&amp; defined(ALPHA)</span></div><div class="line"><a name="l04927"></a><span class="lineno"> 4927</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 2 &amp;&amp; defined(ALPHA)</span></div><div class="line"><a name="l04928"></a><span class="lineno"> 4928</span>&#160; acc2 = acc2 * (half8)ALPHA;</div><div class="line"><a name="l04929"></a><span class="lineno"> 4929</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 2 &amp;&amp; defined(ALPHA)</span></div><div class="line"><a name="l04930"></a><span class="lineno"> 4930</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 3 &amp;&amp; defined(ALPHA)</span></div><div class="line"><a name="l04931"></a><span class="lineno"> 4931</span>&#160; acc3 = acc3 * (half8)ALPHA;</div><div class="line"><a name="l04932"></a><span class="lineno"> 4932</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 3 &amp;&amp; defined(ALPHA)</span></div><div class="line"><a name="l04933"></a><span class="lineno"> 4933</span>&#160;</div><div class="line"><a name="l04934"></a><span class="lineno"> 4934</span>&#160;<span class="preprocessor">#if defined(ADD_VEC_C)</span></div><div class="line"><a name="l04935"></a><span class="lineno"> 4935</span>&#160; <span class="comment">// *INDENT-OFF*</span></div><div class="line"><a name="l04936"></a><span class="lineno"> 4936</span>&#160; <span class="comment">// clang-format off</span></div><div class="line"><a name="l04937"></a><span class="lineno"> 4937</span>&#160; __global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *src2_addr = (__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)(src2_ptr + src2_offset_first_element_in_bytes + get_global_id(0) * src2_step_x);</div><div class="line"><a name="l04938"></a><span class="lineno"> 4938</span>&#160; half8 c0 = vload8(0, src2_addr);</div><div class="line"><a name="l04939"></a><span class="lineno"> 4939</span>&#160; <span class="comment">// clang-format on</span></div><div class="line"><a name="l04940"></a><span class="lineno"> 4940</span>&#160; <span class="comment">// *INDENT-ON*</span></div><div class="line"><a name="l04941"></a><span class="lineno"> 4941</span>&#160;</div><div class="line"><a name="l04942"></a><span class="lineno"> 4942</span>&#160; acc0 += c0;</div><div class="line"><a name="l04943"></a><span class="lineno"> 4943</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 1</span></div><div class="line"><a name="l04944"></a><span class="lineno"> 4944</span>&#160; acc1 += c0;</div><div class="line"><a name="l04945"></a><span class="lineno"> 4945</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 1</span></div><div class="line"><a name="l04946"></a><span class="lineno"> 4946</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 2</span></div><div class="line"><a name="l04947"></a><span class="lineno"> 4947</span>&#160; acc2 += c0;</div><div class="line"><a name="l04948"></a><span class="lineno"> 4948</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 2</span></div><div class="line"><a name="l04949"></a><span class="lineno"> 4949</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 3</span></div><div class="line"><a name="l04950"></a><span class="lineno"> 4950</span>&#160; acc3 += c0;</div><div class="line"><a name="l04951"></a><span class="lineno"> 4951</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 3</span></div><div class="line"><a name="l04952"></a><span class="lineno"> 4952</span>&#160;<span class="preprocessor">#endif </span><span class="comment">/* defined(ADD_VEC_C) */</span><span class="preprocessor"></span></div><div class="line"><a name="l04953"></a><span class="lineno"> 4953</span>&#160;</div><div class="line"><a name="l04954"></a><span class="lineno"> 4954</span>&#160; <span class="keywordtype">int</span> z = get_global_id(2);</div><div class="line"><a name="l04955"></a><span class="lineno"> 4955</span>&#160;</div><div class="line"><a name="l04956"></a><span class="lineno"> 4956</span>&#160; <span class="comment">// Compute destination address</span></div><div class="line"><a name="l04957"></a><span class="lineno"> 4957</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="l04958"></a><span class="lineno"> 4958</span>&#160;</div><div class="line"><a name="l04959"></a><span class="lineno"> 4959</span>&#160; <span class="comment">// Compute dst address</span></div><div class="line"><a name="l04960"></a><span class="lineno"> 4960</span>&#160; __global uchar *dst_addr = <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>(&amp;<a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#adbf67dcee294e673cf796f1ed8aeb6a4">dst</a>, 0, 0);</div><div class="line"><a name="l04961"></a><span class="lineno"> 4961</span>&#160;</div><div class="line"><a name="l04962"></a><span class="lineno"> 4962</span>&#160;<span class="preprocessor">#if defined(REINTERPRET_OUTPUT_AS_3D)</span></div><div class="line"><a name="l04963"></a><span class="lineno"> 4963</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="l04964"></a><span class="lineno"> 4964</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="l04965"></a><span class="lineno"> 4965</span>&#160; <span class="comment">//</span></div><div class="line"><a name="l04966"></a><span class="lineno"> 4966</span>&#160; <span class="comment">// | |</span></div><div class="line"><a name="l04967"></a><span class="lineno"> 4967</span>&#160; <span class="comment">// | plane0 |</span></div><div class="line"><a name="l04968"></a><span class="lineno"> 4968</span>&#160; <span class="comment">// | |</span></div><div class="line"><a name="l04969"></a><span class="lineno"> 4969</span>&#160; <span class="comment">// |__________________|</span></div><div class="line"><a name="l04970"></a><span class="lineno"> 4970</span>&#160; <span class="comment">// |******************|</span></div><div class="line"><a name="l04971"></a><span class="lineno"> 4971</span>&#160; <span class="comment">// | cross_plane_pad |</span></div><div class="line"><a name="l04972"></a><span class="lineno"> 4972</span>&#160; <span class="comment">// |******************|</span></div><div class="line"><a name="l04973"></a><span class="lineno"> 4973</span>&#160; <span class="comment">// | |</span></div><div class="line"><a name="l04974"></a><span class="lineno"> 4974</span>&#160; <span class="comment">// | plane1 |</span></div><div class="line"><a name="l04975"></a><span class="lineno"> 4975</span>&#160; <span class="comment">// | |</span></div><div class="line"><a name="l04976"></a><span class="lineno"> 4976</span>&#160; <span class="comment">// |__________________|</span></div><div class="line"><a name="l04977"></a><span class="lineno"> 4977</span>&#160;</div><div class="line"><a name="l04978"></a><span class="lineno"> 4978</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="l04979"></a><span class="lineno"> 4979</span>&#160; uint4 zout = ((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="l04980"></a><span class="lineno"> 4980</span>&#160; zout = min(DEPTH_GEMM3D - 1, zout);</div><div class="line"><a name="l04981"></a><span class="lineno"> 4981</span>&#160;</div><div class="line"><a name="l04982"></a><span class="lineno"> 4982</span>&#160; <span class="comment">// Add offset due to the cross plane paddings</span></div><div class="line"><a name="l04983"></a><span class="lineno"> 4983</span>&#160; zout *= (dst_cross_plane_pad * dst_stride_y);</div><div class="line"><a name="l04984"></a><span class="lineno"> 4984</span>&#160;</div><div class="line"><a name="l04985"></a><span class="lineno"> 4985</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="l04986"></a><span class="lineno"> 4986</span>&#160; <span class="comment">// multiply dst_stride_z by DEPTH_GEMM3D</span></div><div class="line"><a name="l04987"></a><span class="lineno"> 4987</span>&#160; dst_addr += z * dst_stride_z * DEPTH_GEMM3D;</div><div class="line"><a name="l04988"></a><span class="lineno"> 4988</span>&#160;</div><div class="line"><a name="l04989"></a><span class="lineno"> 4989</span>&#160; <span class="comment">// Store the output block</span></div><div class="line"><a name="l04990"></a><span class="lineno"> 4990</span>&#160; vstore8(acc0, 0, (__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)(dst_addr + 0 * dst_stride_y + zout.s0));</div><div class="line"><a name="l04991"></a><span class="lineno"> 4991</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 1</span></div><div class="line"><a name="l04992"></a><span class="lineno"> 4992</span>&#160; vstore8(acc1, 0, (__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)(dst_addr + 1 * dst_stride_y + zout.s1));</div><div class="line"><a name="l04993"></a><span class="lineno"> 4993</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 1</span></div><div class="line"><a name="l04994"></a><span class="lineno"> 4994</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 2</span></div><div class="line"><a name="l04995"></a><span class="lineno"> 4995</span>&#160; vstore8(acc2, 0, (__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)(dst_addr + 2 * dst_stride_y + zout.s2));</div><div class="line"><a name="l04996"></a><span class="lineno"> 4996</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 2</span></div><div class="line"><a name="l04997"></a><span class="lineno"> 4997</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 3</span></div><div class="line"><a name="l04998"></a><span class="lineno"> 4998</span>&#160; vstore8(acc3, 0, (__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)(dst_addr + 3 * dst_stride_y + zout.s3));</div><div class="line"><a name="l04999"></a><span class="lineno"> 4999</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 3</span></div><div class="line"><a name="l05000"></a><span class="lineno"> 5000</span>&#160;</div><div class="line"><a name="l05001"></a><span class="lineno"> 5001</span>&#160;<span class="preprocessor">#else // defined(REINTERPRET_OUTPUT_AS_3D)</span></div><div class="line"><a name="l05002"></a><span class="lineno"> 5002</span>&#160; <span class="comment">// Add offset for batched GEMM</span></div><div class="line"><a name="l05003"></a><span class="lineno"> 5003</span>&#160; dst_addr += z * dst_stride_z;</div><div class="line"><a name="l05004"></a><span class="lineno"> 5004</span>&#160;</div><div class="line"><a name="l05005"></a><span class="lineno"> 5005</span>&#160; <span class="comment">// Store the output block</span></div><div class="line"><a name="l05006"></a><span class="lineno"> 5006</span>&#160; vstore8(acc0, 0, (__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)(dst_addr + 0 * dst_stride_y));</div><div class="line"><a name="l05007"></a><span class="lineno"> 5007</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 1</span></div><div class="line"><a name="l05008"></a><span class="lineno"> 5008</span>&#160; vstore8(acc1, 0, (__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)(dst_addr + 1 * dst_stride_y));</div><div class="line"><a name="l05009"></a><span class="lineno"> 5009</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 1</span></div><div class="line"><a name="l05010"></a><span class="lineno"> 5010</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 2</span></div><div class="line"><a name="l05011"></a><span class="lineno"> 5011</span>&#160; vstore8(acc2, 0, (__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)(dst_addr + 2 * dst_stride_y));</div><div class="line"><a name="l05012"></a><span class="lineno"> 5012</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 2</span></div><div class="line"><a name="l05013"></a><span class="lineno"> 5013</span>&#160;<span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 3</span></div><div class="line"><a name="l05014"></a><span class="lineno"> 5014</span>&#160; vstore8(acc3, 0, (__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)(dst_addr + 3 * dst_stride_y));</div><div class="line"><a name="l05015"></a><span class="lineno"> 5015</span>&#160;<span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y &gt; 3</span></div><div class="line"><a name="l05016"></a><span class="lineno"> 5016</span>&#160;<span class="preprocessor">#endif // REINTERPRET_OUTPUT_AS_3D</span></div><div class="line"><a name="l05017"></a><span class="lineno"> 5017</span>&#160;}</div><div class="line"><a name="l05018"></a><span class="lineno"> 5018</span>&#160;<span class="preprocessor">#endif // defined(ARM_COMPUTE_OPENCL_FP16_ENABLED)</span></div><div class="line"><a name="l05019"></a><span class="lineno"> 5019</span>&#160;</div><div class="line"><a name="l05020"></a><span class="lineno"> 5020</span>&#160;<span class="preprocessor">#endif // defined(COLS_A) &amp;&amp; defined(NUM_ELEMS_PROCESSED_PER_THREAD_X) &amp;&amp; (NUM_ELEMS_PROCESSED_PER_THREAD_Y)</span></div><div class="line"><a name="l05021"></a><span class="lineno"> 5021</span>&#160;</div><div class="line"><a name="l05022"></a><span class="lineno"> 5022</span>&#160;<span class="preprocessor">#if defined(BETA)</span></div><div class="line"><a name="l05023"></a><span class="lineno"> 5023</span>&#160;</div><div class="line"><a name="l05044"></a><span class="lineno"> 5044</span>&#160;__kernel <span class="keywordtype">void</span> gemm_ma_f32(<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#ae9d2dc29c2789c253406f9b304cc75a8">src</a>),</div><div class="line"><a name="l05045"></a><span class="lineno"> 5045</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="l05046"></a><span class="lineno"> 5046</span>&#160;{</div><div class="line"><a name="l05047"></a><span class="lineno"> 5047</span>&#160; <span class="comment">// Compute source and destination addresses</span></div><div class="line"><a name="l05048"></a><span class="lineno"> 5048</span>&#160; <a class="code" href="struct_tensor3_d.xhtml">Tensor3D</a> <a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#ae9d2dc29c2789c253406f9b304cc75a8">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#ae9d2dc29c2789c253406f9b304cc75a8">src</a>);</div><div class="line"><a name="l05049"></a><span class="lineno"> 5049</span>&#160; <a class="code" href="struct_tensor3_d.xhtml">Tensor3D</a> <a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#adbf67dcee294e673cf796f1ed8aeb6a4">dst</a> = <a class="code" href="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#adbf67dcee294e673cf796f1ed8aeb6a4">dst</a>);</div><div class="line"><a name="l05050"></a><span class="lineno"> 5050</span>&#160;</div><div class="line"><a name="l05051"></a><span class="lineno"> 5051</span>&#160; <span class="comment">// Load values from A x B</span></div><div class="line"><a name="l05052"></a><span class="lineno"> 5052</span>&#160; float4 alpha_ab = vload4(0, (__global <span class="keywordtype">float</span> *)<a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#adbf67dcee294e673cf796f1ed8aeb6a4">dst</a>.ptr);</div><div class="line"><a name="l05053"></a><span class="lineno"> 5053</span>&#160;</div><div class="line"><a name="l05054"></a><span class="lineno"> 5054</span>&#160; <span class="comment">// Load values from Matrix C</span></div><div class="line"><a name="l05055"></a><span class="lineno"> 5055</span>&#160; float4 c = vload4(0, (__global <span class="keywordtype">float</span> *)<a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#ae9d2dc29c2789c253406f9b304cc75a8">src</a>.ptr);</div><div class="line"><a name="l05056"></a><span class="lineno"> 5056</span>&#160;</div><div class="line"><a name="l05057"></a><span class="lineno"> 5057</span>&#160; <span class="comment">// Computes alpha * axb + beta * c</span></div><div class="line"><a name="l05058"></a><span class="lineno"> 5058</span>&#160; float4 out = alpha_ab + (float4)BETA * c;</div><div class="line"><a name="l05059"></a><span class="lineno"> 5059</span>&#160;</div><div class="line"><a name="l05060"></a><span class="lineno"> 5060</span>&#160; <span class="comment">// Store final result in axb matrix</span></div><div class="line"><a name="l05061"></a><span class="lineno"> 5061</span>&#160; vstore4(out, 0, (__global <span class="keywordtype">float</span> *)<a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#adbf67dcee294e673cf796f1ed8aeb6a4">dst</a>.ptr);</div><div class="line"><a name="l05062"></a><span class="lineno"> 5062</span>&#160;}</div><div class="line"><a name="l05063"></a><span class="lineno"> 5063</span>&#160;</div><div class="line"><a name="l05064"></a><span class="lineno"> 5064</span>&#160;<span class="preprocessor">#if defined(ARM_COMPUTE_OPENCL_FP16_ENABLED)</span></div><div class="line"><a name="l05065"></a><span class="lineno"> 5065</span>&#160;</div><div class="line"><a name="l05086"></a><span class="lineno"> 5086</span>&#160;__kernel <span class="keywordtype">void</span> gemm_ma_f16(<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#ae9d2dc29c2789c253406f9b304cc75a8">src</a>),</div><div class="line"><a name="l05087"></a><span class="lineno"> 5087</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="l05088"></a><span class="lineno"> 5088</span>&#160;{</div><div class="line"><a name="l05089"></a><span class="lineno"> 5089</span>&#160; <span class="comment">// Compute source and destination addresses</span></div><div class="line"><a name="l05090"></a><span class="lineno"> 5090</span>&#160; <a class="code" href="struct_tensor3_d.xhtml">Tensor3D</a> <a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#ae9d2dc29c2789c253406f9b304cc75a8">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#ae9d2dc29c2789c253406f9b304cc75a8">src</a>);</div><div class="line"><a name="l05091"></a><span class="lineno"> 5091</span>&#160; <a class="code" href="struct_tensor3_d.xhtml">Tensor3D</a> <a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#adbf67dcee294e673cf796f1ed8aeb6a4">dst</a> = <a class="code" href="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#adbf67dcee294e673cf796f1ed8aeb6a4">dst</a>);</div><div class="line"><a name="l05092"></a><span class="lineno"> 5092</span>&#160;</div><div class="line"><a name="l05093"></a><span class="lineno"> 5093</span>&#160; <span class="comment">// Load values from A x B</span></div><div class="line"><a name="l05094"></a><span class="lineno"> 5094</span>&#160; half8 alpha_ab = vload8(0, (__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)<a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#adbf67dcee294e673cf796f1ed8aeb6a4">dst</a>.ptr);</div><div class="line"><a name="l05095"></a><span class="lineno"> 5095</span>&#160;</div><div class="line"><a name="l05096"></a><span class="lineno"> 5096</span>&#160; <span class="comment">// Load values from Matrix C</span></div><div class="line"><a name="l05097"></a><span class="lineno"> 5097</span>&#160; half8 c = vload8(0, (__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)<a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#ae9d2dc29c2789c253406f9b304cc75a8">src</a>.ptr);</div><div class="line"><a name="l05098"></a><span class="lineno"> 5098</span>&#160;</div><div class="line"><a name="l05099"></a><span class="lineno"> 5099</span>&#160; <span class="comment">// Computes alpha * axb + beta * c</span></div><div class="line"><a name="l05100"></a><span class="lineno"> 5100</span>&#160; half8 out = alpha_ab + (half8)BETA * c;</div><div class="line"><a name="l05101"></a><span class="lineno"> 5101</span>&#160;</div><div class="line"><a name="l05102"></a><span class="lineno"> 5102</span>&#160; <span class="comment">// Store final result in axb matrix</span></div><div class="line"><a name="l05103"></a><span class="lineno"> 5103</span>&#160; vstore8(out, 0, (__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)<a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#adbf67dcee294e673cf796f1ed8aeb6a4">dst</a>.ptr);</div><div class="line"><a name="l05104"></a><span class="lineno"> 5104</span>&#160;}</div><div class="line"><a name="l05105"></a><span class="lineno"> 5105</span>&#160;<span class="preprocessor">#endif // defined(ARM_COMPUTE_OPENCL_FP16_ENABLED)</span></div><div class="line"><a name="l05106"></a><span class="lineno"> 5106</span>&#160;<span class="preprocessor">#endif // defined(BETA)</span></div><div class="line"><a name="l05107"></a><span class="lineno"> 5107</span>&#160;</div><div class="line"><a name="l05108"></a><span class="lineno"> 5108</span>&#160;<span class="preprocessor">#if defined(WIDTH_VECTOR_A)</span></div><div class="line"><a name="l05109"></a><span class="lineno"> 5109</span>&#160;</div><div class="line"><a name="l05136"></a><span class="lineno"> 5136</span>&#160;__kernel <span class="keywordtype">void</span> gemm_lc_vm_f32(<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="l05137"></a><span class="lineno"> 5137</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a6b83038822d1ae7ab619b684ed3b7fc0">TENSOR3D_DECLARATION</a>(src1),</div><div class="line"><a name="l05138"></a><span class="lineno"> 5138</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="l05139"></a><span class="lineno"> 5139</span>&#160;{</div><div class="line"><a name="l05140"></a><span class="lineno"> 5140</span>&#160; <span class="keywordtype">int</span> idx = get_global_id(0) * 4;</div><div class="line"><a name="l05141"></a><span class="lineno"> 5141</span>&#160; <span class="keywordtype">int</span> idy = get_global_id(1);</div><div class="line"><a name="l05142"></a><span class="lineno"> 5142</span>&#160;</div><div class="line"><a name="l05143"></a><span class="lineno"> 5143</span>&#160; <span class="comment">// Compute the address for the vector A and matrix B</span></div><div class="line"><a name="l05144"></a><span class="lineno"> 5144</span>&#160; int2 src_addr = ((int2)(src0_offset_first_element_in_bytes + src0_stride_y * idy, src1_offset_first_element_in_bytes + src1_stride_z * idy));</div><div class="line"><a name="l05145"></a><span class="lineno"> 5145</span>&#160; src_addr.s1 += idx * <span class="keyword">sizeof</span>(float);</div><div class="line"><a name="l05146"></a><span class="lineno"> 5146</span>&#160;</div><div class="line"><a name="l05147"></a><span class="lineno"> 5147</span>&#160; <span class="keywordtype">int</span> end_row_vec_a = src_addr.s0 + (WIDTH_VECTOR_A * <span class="keyword">sizeof</span>(float));</div><div class="line"><a name="l05148"></a><span class="lineno"> 5148</span>&#160;</div><div class="line"><a name="l05149"></a><span class="lineno"> 5149</span>&#160; float4 acc = 0.0f;</div><div class="line"><a name="l05150"></a><span class="lineno"> 5150</span>&#160;</div><div class="line"><a name="l05151"></a><span class="lineno"> 5151</span>&#160; <span class="keywordflow">for</span>(; src_addr.s0 &lt;= (end_row_vec_a - 2 * (int)<span class="keyword">sizeof</span>(<span class="keywordtype">float</span>)); src_addr += (int2)(2 * <span class="keyword">sizeof</span>(<span class="keywordtype">float</span>), 2 * src1_stride_y))</div><div class="line"><a name="l05152"></a><span class="lineno"> 5152</span>&#160; {</div><div class="line"><a name="l05153"></a><span class="lineno"> 5153</span>&#160; float2 a0 = vload2(0, (__global <span class="keywordtype">float</span> *)(src0_ptr + src_addr.s0));</div><div class="line"><a name="l05154"></a><span class="lineno"> 5154</span>&#160; float4 b0 = vload4(0, (__global <span class="keywordtype">float</span> *)(src1_ptr + src_addr.s1));</div><div class="line"><a name="l05155"></a><span class="lineno"> 5155</span>&#160; float4 b1 = vload4(0, (__global <span class="keywordtype">float</span> *)(src1_ptr + src_addr.s1 + src1_stride_y));</div><div class="line"><a name="l05156"></a><span class="lineno"> 5156</span>&#160;</div><div class="line"><a name="l05157"></a><span class="lineno"> 5157</span>&#160; acc += b0 * (float4)a0.s0;</div><div class="line"><a name="l05158"></a><span class="lineno"> 5158</span>&#160; acc += b1 * (float4)a0.s1;</div><div class="line"><a name="l05159"></a><span class="lineno"> 5159</span>&#160; }</div><div class="line"><a name="l05160"></a><span class="lineno"> 5160</span>&#160;</div><div class="line"><a name="l05161"></a><span class="lineno"> 5161</span>&#160; <span class="keywordflow">for</span>(; src_addr.s0 &lt; end_row_vec_a; src_addr += (int2)(<span class="keyword">sizeof</span>(<span class="keywordtype">float</span>), src1_stride_y))</div><div class="line"><a name="l05162"></a><span class="lineno"> 5162</span>&#160; {</div><div class="line"><a name="l05163"></a><span class="lineno"> 5163</span>&#160; <span class="keywordtype">float</span> a0 = *((__global <span class="keywordtype">float</span> *)(src0_ptr + src_addr.s0));</div><div class="line"><a name="l05164"></a><span class="lineno"> 5164</span>&#160; float4 b0 = vload4(0, (__global <span class="keywordtype">float</span> *)(src1_ptr + src_addr.s1));</div><div class="line"><a name="l05165"></a><span class="lineno"> 5165</span>&#160;</div><div class="line"><a name="l05166"></a><span class="lineno"> 5166</span>&#160; acc += b0 * (float4)a0;</div><div class="line"><a name="l05167"></a><span class="lineno"> 5167</span>&#160; }</div><div class="line"><a name="l05168"></a><span class="lineno"> 5168</span>&#160;</div><div class="line"><a name="l05169"></a><span class="lineno"> 5169</span>&#160; <span class="comment">// Compute destination address</span></div><div class="line"><a name="l05170"></a><span class="lineno"> 5170</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="l05171"></a><span class="lineno"> 5171</span>&#160;</div><div class="line"><a name="l05172"></a><span class="lineno"> 5172</span>&#160; vstore4(acc, 0, (__global <span class="keywordtype">float</span> *)(<a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>(&amp;<a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#adbf67dcee294e673cf796f1ed8aeb6a4">dst</a>, 0, 0)));</div><div class="line"><a name="l05173"></a><span class="lineno"> 5173</span>&#160;}</div><div class="line"><a name="l05174"></a><span class="lineno"> 5174</span>&#160;<span class="preprocessor">#endif // defined(WIDTH_VECTOR_A)</span></div><div class="line"><a name="l05175"></a><span class="lineno"> 5175</span>&#160;</div><div class="line"><a name="l05192"></a><span class="lineno"> 5192</span>&#160;<span class="preprocessor">#if defined(DATA_TYPE) &amp;&amp; defined(VECTOR_SIZE)</span></div><div class="line"><a name="l05193"></a><span class="lineno"> 5193</span>&#160;__kernel <span class="keywordtype">void</span> gemm_accumulate_biases(</div><div class="line"><a name="l05194"></a><span class="lineno"> 5194</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a22f42fcf2077d951271df83b55c1a71a">IMAGE_DECLARATION</a>(accum),</div><div class="line"><a name="l05195"></a><span class="lineno"> 5195</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="l05196"></a><span class="lineno"> 5196</span>&#160;{</div><div class="line"><a name="l05197"></a><span class="lineno"> 5197</span>&#160; <a class="code" href="struct_image.xhtml">Image</a> accum = <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#aebe814363556c244be043b13e7969197">CONVERT_TO_IMAGE_STRUCT</a>(accum);</div><div class="line"><a name="l05198"></a><span class="lineno"> 5198</span>&#160; <a class="code" href="struct_vector.xhtml">Vector</a> biases = <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a527bfdf5eeb306f1cf01c4a8e29f38e0">CONVERT_TO_VECTOR_STRUCT</a>(biases);</div><div class="line"><a name="l05199"></a><span class="lineno"> 5199</span>&#160;</div><div class="line"><a name="l05200"></a><span class="lineno"> 5200</span>&#160; <span class="comment">// Vector size, i.e. number of vector elements.</span></div><div class="line"><a name="l05201"></a><span class="lineno"> 5201</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, <a class="code" href="softmax__layer_8cl.xhtml#a7c78836761fa3b5b124efea237dac70f">VECTOR_SIZE</a>)</div><div class="line"><a name="l05202"></a><span class="lineno"> 5202</span>&#160; accum_value = <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a287e2fc366c312b468382c95bb90f91f">VLOAD</a>(<a class="code" href="softmax__layer_8cl.xhtml#a7c78836761fa3b5b124efea237dac70f">VECTOR_SIZE</a>)(0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)accum.<a class="code" href="struct_image.xhtml#acf52c23cbd7424606c10a606524e3e32">ptr</a>);</div><div class="line"><a name="l05203"></a><span class="lineno"> 5203</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, <a class="code" href="softmax__layer_8cl.xhtml#a7c78836761fa3b5b124efea237dac70f">VECTOR_SIZE</a>)</div><div class="line"><a name="l05204"></a><span class="lineno"> 5204</span>&#160; biases_value = <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a287e2fc366c312b468382c95bb90f91f">VLOAD</a>(<a class="code" href="softmax__layer_8cl.xhtml#a7c78836761fa3b5b124efea237dac70f">VECTOR_SIZE</a>)(0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)biases.<a class="code" href="struct_vector.xhtml#acf52c23cbd7424606c10a606524e3e32">ptr</a>);</div><div class="line"><a name="l05205"></a><span class="lineno"> 5205</span>&#160; accum_value = biases_value + accum_value;</div><div class="line"><a name="l05206"></a><span class="lineno"> 5206</span>&#160; <span class="comment">// Store result in the accumulate buffer</span></div><div class="line"><a name="l05207"></a><span class="lineno"> 5207</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#acb282042d1edeeaa3cc979a206f78b54">VSTORE</a>(<a class="code" href="softmax__layer_8cl.xhtml#a7c78836761fa3b5b124efea237dac70f">VECTOR_SIZE</a>)</div><div class="line"><a name="l05208"></a><span class="lineno"> 5208</span>&#160; (accum_value, 0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)accum.<a class="code" href="struct_image.xhtml#acf52c23cbd7424606c10a606524e3e32">ptr</a>);</div><div class="line"><a name="l05209"></a><span class="lineno"> 5209</span>&#160;}</div><div class="line"><a name="l05210"></a><span class="lineno"> 5210</span>&#160;<span class="preprocessor">#endif // defined(DATA_TYPE) &amp;&amp; defined(VECTOR_SIZE)</span></div><div class="ttc" id="struct_vector_xhtml"><div class="ttname"><a href="struct_vector.xhtml">Vector</a></div><div class="ttdoc">Structure to hold Vector information.</div><div class="ttdef"><b>Definition:</b> <a href="src_2core_2_c_l_2cl__kernels_2_helpers_8h_source.xhtml#l00143">helpers.h:143</a></div></div>
<div class="ttc" id="src_2core_2_c_l_2cl__kernels_2_helpers_8h_xhtml_a009469e4d9b8fce3b6d5e97d2077827d"><div class="ttname"><a href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a></div><div class="ttdeci">__global uchar * offset(const Image *img, int x, int y)</div><div class="ttdoc">Get the pointer position of a Image.</div><div class="ttdef"><b>Definition:</b> <a href="src_2core_2_c_l_2cl__kernels_2_helpers_8h_source.xhtml#l00312">helpers.h:312</a></div></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="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#l00113">helpers.h:113</a></div></div>
<div class="ttc" id="convolution3x3_8cl_xhtml_afb8c72ce35c4a1f4a2588d6573e54aa1"><div class="ttname"><a href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a></div><div class="ttdeci">#define DATA_TYPE</div><div class="ttdef"><b>Definition:</b> <a href="convolution3x3_8cl_source.xhtml#l00027">convolution3x3.cl:27</a></div></div>
<div class="ttc" id="namespacearm__compute_xhtml_a73e2825fd61d349c5ca2f5313e3c8ea1"><div class="ttname"><a href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">arm_compute::half</a></div><div class="ttdeci">half_float::half half</div><div class="ttdoc">16-bit floating point type</div><div class="ttdef"><b>Definition:</b> <a href="arm__compute_2core_2_types_8h_source.xhtml#l00044">Types.h:44</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#l00077">helpers.h:77</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#l00160">helpers.h:160</a></div></div>
<div class="ttc" id="src_2core_2_c_l_2cl__kernels_2_helpers_8h_xhtml_a527bfdf5eeb306f1cf01c4a8e29f38e0"><div class="ttname"><a href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a527bfdf5eeb306f1cf01c4a8e29f38e0">CONVERT_TO_VECTOR_STRUCT</a></div><div class="ttdeci">#define CONVERT_TO_VECTOR_STRUCT(name)</div><div class="ttdef"><b>Definition:</b> <a href="src_2core_2_c_l_2cl__kernels_2_helpers_8h_source.xhtml#l00107">helpers.h:107</a></div></div>
<div class="ttc" id="softmax__layer_8cl_xhtml_a7c78836761fa3b5b124efea237dac70f"><div class="ttname"><a href="softmax__layer_8cl.xhtml#a7c78836761fa3b5b124efea237dac70f">VECTOR_SIZE</a></div><div class="ttdeci">#define VECTOR_SIZE</div><div class="ttdef"><b>Definition:</b> <a href="softmax__layer_8cl_source.xhtml#l00060">softmax_layer.cl:60</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#l00071">helpers.h:71</a></div></div>
<div class="ttc" id="struct_vector_xhtml_acf52c23cbd7424606c10a606524e3e32"><div class="ttname"><a href="struct_vector.xhtml#acf52c23cbd7424606c10a606524e3e32">Vector::ptr</a></div><div class="ttdeci">__global uchar * ptr</div><div class="ttdoc">Pointer to the starting postion of the buffer.</div><div class="ttdef"><b>Definition:</b> <a href="src_2core_2_c_l_2cl__kernels_2_helpers_8h_source.xhtml#l00145">helpers.h:145</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="namespacearm__compute_1_1test_1_1validation_xhtml_ae9d2dc29c2789c253406f9b304cc75a8"><div class="ttname"><a href="namespacearm__compute_1_1test_1_1validation.xhtml#ae9d2dc29c2789c253406f9b304cc75a8">arm_compute::test::validation::src</a></div><div class="ttdeci">cast configure &amp; src</div><div class="ttdef"><b>Definition:</b> <a href="_cast_8cpp_source.xhtml#l00169">Cast.cpp:169</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#l00151">helpers.h:151</a></div></div>
<div class="ttc" id="struct_image_xhtml_acf52c23cbd7424606c10a606524e3e32"><div class="ttname"><a href="struct_image.xhtml#acf52c23cbd7424606c10a606524e3e32">Image::ptr</a></div><div class="ttdeci">__global uchar * ptr</div><div class="ttdoc">Pointer to the starting postion of the buffer.</div><div class="ttdef"><b>Definition:</b> <a href="src_2core_2_c_l_2cl__kernels_2_helpers_8h_source.xhtml#l00153">helpers.h:153</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#l00128">helpers.h:128</a></div></div>
<div class="ttc" id="src_2core_2_c_l_2cl__kernels_2_helpers_8h_xhtml"><div class="ttname"><a href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml">helpers.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#l00051">helpers.h:51</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#l00048">helpers.h:48</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#l00085">helpers.h:85</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#l00057">helpers.h:57</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="gemm_8cl.xhtml">gemm.cl</a></li>
<li class="footer">Generated on Thu Feb 28 2019 12:24:49 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>