Google Git
Sign in
android / platform / external / ComputeLibrary / e8521c234 / . / documentation / direct__convolution3x3_8cl_source.xhtml
blob: d2eb1026ba70ffdf4c6fa11706f6d0de6d9260ca [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/direct_convolution3x3.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">20.02.1</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('direct__convolution3x3_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">direct_convolution3x3.cl</div> </div>
</div><!--header-->
<div class="contents">
<a href="direct__convolution3x3_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) 2016-2018 ARM Limited.</span></div><div class="line"><a name="l00003"></a><span class="lineno"> 3</span>&#160;<span class="comment"> *</span></div><div class="line"><a name="l00004"></a><span class="lineno"> 4</span>&#160;<span class="comment"> * SPDX-License-Identifier: MIT</span></div><div class="line"><a name="l00005"></a><span class="lineno"> 5</span>&#160;<span class="comment"> *</span></div><div class="line"><a name="l00006"></a><span class="lineno"> 6</span>&#160;<span class="comment"> * Permission is hereby granted, free of charge, to any person obtaining a copy</span></div><div class="line"><a name="l00007"></a><span class="lineno"> 7</span>&#160;<span class="comment"> * of this software and associated documentation files (the &quot;Software&quot;), to</span></div><div class="line"><a name="l00008"></a><span class="lineno"> 8</span>&#160;<span class="comment"> * deal in the Software without restriction, including without limitation the</span></div><div class="line"><a name="l00009"></a><span class="lineno"> 9</span>&#160;<span class="comment"> * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or</span></div><div class="line"><a name="l00010"></a><span class="lineno"> 10</span>&#160;<span class="comment"> * sell copies of the Software, and to permit persons to whom the Software is</span></div><div class="line"><a name="l00011"></a><span class="lineno"> 11</span>&#160;<span class="comment"> * furnished to do so, subject to the following conditions:</span></div><div class="line"><a name="l00012"></a><span class="lineno"> 12</span>&#160;<span class="comment"> *</span></div><div class="line"><a name="l00013"></a><span class="lineno"> 13</span>&#160;<span class="comment"> * The above copyright notice and this permission notice shall be included in all</span></div><div class="line"><a name="l00014"></a><span class="lineno"> 14</span>&#160;<span class="comment"> * copies or substantial portions of the Software.</span></div><div class="line"><a name="l00015"></a><span class="lineno"> 15</span>&#160;<span class="comment"> *</span></div><div class="line"><a name="l00016"></a><span class="lineno"> 16</span>&#160;<span class="comment"> * THE SOFTWARE IS PROVIDED &quot;AS IS&quot;, WITHOUT WARRANTY OF ANY KIND, EXPRESS OR</span></div><div class="line"><a name="l00017"></a><span class="lineno"> 17</span>&#160;<span class="comment"> * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,</span></div><div class="line"><a name="l00018"></a><span class="lineno"> 18</span>&#160;<span class="comment"> * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE</span></div><div class="line"><a name="l00019"></a><span class="lineno"> 19</span>&#160;<span class="comment"> * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER</span></div><div class="line"><a name="l00020"></a><span class="lineno"> 20</span>&#160;<span class="comment"> * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,</span></div><div class="line"><a name="l00021"></a><span class="lineno"> 21</span>&#160;<span class="comment"> * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE</span></div><div class="line"><a name="l00022"></a><span class="lineno"> 22</span>&#160;<span class="comment"> * SOFTWARE.</span></div><div class="line"><a name="l00023"></a><span class="lineno"> 23</span>&#160;<span class="comment"> */</span></div><div class="line"><a name="l00024"></a><span class="lineno"> 24</span>&#160;<span class="preprocessor">#include &quot;<a class="code" href="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;</div><div class="line"><a name="l00026"></a><span class="lineno"> 26</span>&#160;<span class="preprocessor">#undef CONVERT_SAT</span></div><div class="line"><a name="l00027"></a><span class="lineno"> 27</span>&#160;</div><div class="line"><a name="l00028"></a><span class="lineno"><a class="line" href="direct__convolution3x3_8cl.xhtml#aebbeb1f22eca3a3f4c3e019e8f419f39"> 28</a></span>&#160;<span class="preprocessor">#define ADD_OP(a, b) ((a) + (b))</span></div><div class="line"><a name="l00029"></a><span class="lineno"><a class="line" href="direct__convolution3x3_8cl.xhtml#ad3cc858846806e6b1d3694b9d0a2e6da"> 29</a></span>&#160;<span class="preprocessor">#define MUL_OP(a, b) ((a) * (b))</span></div><div class="line"><a name="l00030"></a><span class="lineno"><a class="line" href="direct__convolution3x3_8cl.xhtml#a1f15728672380ade7a238f5e783d54d2"> 30</a></span>&#160;<span class="preprocessor">#define CONVERT_SAT(a, b) ((a))</span></div><div class="line"><a name="l00031"></a><span class="lineno"> 31</span>&#160;</div><div class="line"><a name="l00032"></a><span class="lineno"> 32</span>&#160;<span class="preprocessor">#if defined(DATA_TYPE) &amp;&amp; defined(STRIDE_X) &amp;&amp; defined(WEIGHTS_DEPTH)</span></div><div class="line"><a name="l00033"></a><span class="lineno"> 33</span>&#160;</div><div class="line"><a name="l00034"></a><span class="lineno"> 34</span>&#160;<span class="preprocessor">#if STRIDE_X == 1</span></div><div class="line"><a name="l00035"></a><span class="lineno"> 35</span>&#160;<span class="preprocessor">#define CONVOLUTION1x3(acc, src_row_ptr, weights_row_ptr) CONVOLUTION1x3_STRIDE1(acc, src_row_ptr, weights_row_ptr)</span></div><div class="line"><a name="l00036"></a><span class="lineno"> 36</span>&#160;<span class="preprocessor">#elif STRIDE_X == 2 </span><span class="comment">/* STRIDE_X == 1 */</span><span class="preprocessor"></span></div><div class="line"><a name="l00037"></a><span class="lineno"> 37</span>&#160;<span class="preprocessor">#define CONVOLUTION1x3(acc, src_row_ptr, weights_row_ptr) CONVOLUTION1x3_STRIDE2(acc, src_row_ptr, weights_row_ptr)</span></div><div class="line"><a name="l00038"></a><span class="lineno"> 38</span>&#160;<span class="preprocessor">#else </span><span class="comment">/* STRIDE_X not equals 1 or 2 */</span><span class="preprocessor"></span></div><div class="line"><a name="l00039"></a><span class="lineno"> 39</span>&#160;<span class="preprocessor">#error &quot;STRIDE_X larger than 2 is not supported&quot;</span></div><div class="line"><a name="l00040"></a><span class="lineno"> 40</span>&#160;<span class="preprocessor">#endif </span><span class="comment">/* STRIDE_X == 2 */</span><span class="preprocessor"></span></div><div class="line"><a name="l00041"></a><span class="lineno"> 41</span>&#160;</div><div class="line"><a name="l00042"></a><span class="lineno"> 42</span>&#160;<span class="preprocessor">#define CONVOLUTION1x3_STRIDE1(acc, src_row_ptr, weights_row_ptr) \</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"> VEC_DATA_TYPE(DATA_TYPE, 3) \</span></div><div class="line"><a name="l00045"></a><span class="lineno"> 45</span>&#160;<span class="preprocessor"> weights_values0 = vload3(0, weights_row_ptr); \</span></div><div class="line"><a name="l00046"></a><span class="lineno"> 46</span>&#160;<span class="preprocessor"> VEC_DATA_TYPE(DATA_TYPE, 8) \</span></div><div class="line"><a name="l00047"></a><span class="lineno"> 47</span>&#160;<span class="preprocessor"> src0 = vload8(0, src_row_ptr); \</span></div><div class="line"><a name="l00048"></a><span class="lineno"> 48</span>&#160;<span class="preprocessor"> VEC_DATA_TYPE(DATA_TYPE, 2) \</span></div><div class="line"><a name="l00049"></a><span class="lineno"> 49</span>&#160;<span class="preprocessor"> src1 = vload2(0, src_row_ptr + 8); \</span></div><div class="line"><a name="l00050"></a><span class="lineno"> 50</span>&#160;<span class="preprocessor"> \</span></div><div class="line"><a name="l00051"></a><span class="lineno"> 51</span>&#160;<span class="preprocessor"> acc = ADD_OP(acc, MUL_OP(src0, (VEC_DATA_TYPE(DATA_TYPE, 8))weights_values0.s0)); \</span></div><div class="line"><a name="l00052"></a><span class="lineno"> 52</span>&#160;<span class="preprocessor"> acc = ADD_OP(acc, MUL_OP((VEC_DATA_TYPE(DATA_TYPE, 8))(src0.s1234, src0.s567, src1.s0), (VEC_DATA_TYPE(DATA_TYPE, 8))weights_values0.s1)); \</span></div><div class="line"><a name="l00053"></a><span class="lineno"> 53</span>&#160;<span class="preprocessor"> acc = ADD_OP(acc, MUL_OP((VEC_DATA_TYPE(DATA_TYPE, 8))(src0.s234, src0.s567, src1.s01), (VEC_DATA_TYPE(DATA_TYPE, 8))weights_values0.s2)); \</span></div><div class="line"><a name="l00054"></a><span class="lineno"> 54</span>&#160;<span class="preprocessor"> })</span></div><div class="line"><a name="l00055"></a><span class="lineno"> 55</span>&#160;</div><div class="line"><a name="l00056"></a><span class="lineno"> 56</span>&#160;<span class="preprocessor">#define CONVOLUTION1x3_STRIDE2(acc, src_row_ptr, weights_row_ptr) \</span></div><div class="line"><a name="l00057"></a><span class="lineno"> 57</span>&#160;<span class="preprocessor"> ({ \</span></div><div class="line"><a name="l00058"></a><span class="lineno"> 58</span>&#160;<span class="preprocessor"> VEC_DATA_TYPE(DATA_TYPE, 3) \</span></div><div class="line"><a name="l00059"></a><span class="lineno"> 59</span>&#160;<span class="preprocessor"> weights_values0 = vload3(0, weights_row_ptr); \</span></div><div class="line"><a name="l00060"></a><span class="lineno"> 60</span>&#160;<span class="preprocessor"> VEC_DATA_TYPE(DATA_TYPE, 16) \</span></div><div class="line"><a name="l00061"></a><span class="lineno"> 61</span>&#160;<span class="preprocessor"> src0 = vload16(0, src_row_ptr); \</span></div><div class="line"><a name="l00062"></a><span class="lineno"> 62</span>&#160;<span class="preprocessor"> DATA_TYPE src1 = *(src_row_ptr + 16); \</span></div><div class="line"><a name="l00063"></a><span class="lineno"> 63</span>&#160;<span class="preprocessor"> \</span></div><div class="line"><a name="l00064"></a><span class="lineno"> 64</span>&#160;<span class="preprocessor"> acc = ADD_OP(acc, MUL_OP(src0.even, (VEC_DATA_TYPE(DATA_TYPE, 8))weights_values0.s0)); \</span></div><div class="line"><a name="l00065"></a><span class="lineno"> 65</span>&#160;<span class="preprocessor"> acc = ADD_OP(acc, MUL_OP((VEC_DATA_TYPE(DATA_TYPE, 8))(src0.s1357, src0.s9BDF), (VEC_DATA_TYPE(DATA_TYPE, 8))weights_values0.s1)); \</span></div><div class="line"><a name="l00066"></a><span class="lineno"> 66</span>&#160;<span class="preprocessor"> acc = ADD_OP(acc, MUL_OP((VEC_DATA_TYPE(DATA_TYPE, 8))(src0.s2468, src0.sACE, src1), (VEC_DATA_TYPE(DATA_TYPE, 8))weights_values0.s2)); \</span></div><div class="line"><a name="l00067"></a><span class="lineno"> 67</span>&#160;<span class="preprocessor"> })</span></div><div class="line"><a name="l00068"></a><span class="lineno"> 68</span>&#160;</div><div class="line"><a name="l00069"></a><span class="lineno"> 69</span>&#160;<span class="preprocessor">#if defined(DATA_LAYOUT_NHWC)</span></div><div class="line"><a name="l00070"></a><span class="lineno"> 70</span>&#160;</div><div class="line"><a name="l00071"></a><span class="lineno"> 71</span>&#160;<span class="preprocessor">#define PTR_TO_VALUE(PTR, DATA_TYPE) *((__global DATA_TYPE *)(PTR))</span></div><div class="line"><a name="l00072"></a><span class="lineno"> 72</span>&#160;</div><div class="line"><a name="l00073"></a><span class="lineno"> 73</span>&#160;<span class="preprocessor">#if STRIDE_X == 1</span></div><div class="line"><a name="l00074"></a><span class="lineno"> 74</span>&#160;<span class="preprocessor">#define CONVOLUTION1x3_NHWC(acc, row_ptr, weights_ptr) CONVOLUTION1x3_STRIDE_NHWC_STRIDE1(acc, row_ptr, weights_ptr)</span></div><div class="line"><a name="l00075"></a><span class="lineno"> 75</span>&#160;<span class="preprocessor">#elif STRIDE_X == 2 </span><span class="comment">/* STRIDE_X == 1 */</span><span class="preprocessor"></span></div><div class="line"><a name="l00076"></a><span class="lineno"> 76</span>&#160;<span class="preprocessor">#define CONVOLUTION1x3_NHWC(acc, row_ptr, weights_ptr) CONVOLUTION1x3_STRIDE_NHWC_STRIDE2(acc, row_ptr, weights_ptr)</span></div><div class="line"><a name="l00077"></a><span class="lineno"> 77</span>&#160;<span class="preprocessor">#else </span><span class="comment">/* STRIDE_X not equals 1 or 2 */</span><span class="preprocessor"></span></div><div class="line"><a name="l00078"></a><span class="lineno"> 78</span>&#160;<span class="preprocessor">#error &quot;STRIDE_X larger than 2 is not supported&quot;</span></div><div class="line"><a name="l00079"></a><span class="lineno"> 79</span>&#160;<span class="preprocessor">#endif </span><span class="comment">/* STRIDE_X == 2 */</span><span class="preprocessor"></span></div><div class="line"><a name="l00080"></a><span class="lineno"> 80</span>&#160;</div><div class="line"><a name="l00081"></a><span class="lineno"> 81</span>&#160;<span class="preprocessor">#define CONVOLUTION1x3_STRIDE_NHWC_STRIDE1(acc, row_ptr, weights_ptr) \</span></div><div class="line"><a name="l00082"></a><span class="lineno"> 82</span>&#160;<span class="preprocessor"> { \</span></div><div class="line"><a name="l00083"></a><span class="lineno"> 83</span>&#160;<span class="preprocessor"> VEC_DATA_TYPE(DATA_TYPE, 8) \</span></div><div class="line"><a name="l00084"></a><span class="lineno"> 84</span>&#160;<span class="preprocessor"> src0 = (VEC_DATA_TYPE(DATA_TYPE, 8))( \</span></div><div class="line"><a name="l00085"></a><span class="lineno"> 85</span>&#160;<span class="preprocessor"> PTR_TO_VALUE(row_ptr + 0 * src_stride_y, DATA_TYPE), \</span></div><div class="line"><a name="l00086"></a><span class="lineno"> 86</span>&#160;<span class="preprocessor"> PTR_TO_VALUE(row_ptr + 1 * src_stride_y, DATA_TYPE), \</span></div><div class="line"><a name="l00087"></a><span class="lineno"> 87</span>&#160;<span class="preprocessor"> PTR_TO_VALUE(row_ptr + 2 * src_stride_y, DATA_TYPE), \</span></div><div class="line"><a name="l00088"></a><span class="lineno"> 88</span>&#160;<span class="preprocessor"> PTR_TO_VALUE(row_ptr + 3 * src_stride_y, DATA_TYPE), \</span></div><div class="line"><a name="l00089"></a><span class="lineno"> 89</span>&#160;<span class="preprocessor"> PTR_TO_VALUE(row_ptr + 4 * src_stride_y, DATA_TYPE), \</span></div><div class="line"><a name="l00090"></a><span class="lineno"> 90</span>&#160;<span class="preprocessor"> PTR_TO_VALUE(row_ptr + 5 * src_stride_y, DATA_TYPE), \</span></div><div class="line"><a name="l00091"></a><span class="lineno"> 91</span>&#160;<span class="preprocessor"> PTR_TO_VALUE(row_ptr + 6 * src_stride_y, DATA_TYPE), \</span></div><div class="line"><a name="l00092"></a><span class="lineno"> 92</span>&#160;<span class="preprocessor"> PTR_TO_VALUE(row_ptr + 7 * src_stride_y, DATA_TYPE)); \</span></div><div class="line"><a name="l00093"></a><span class="lineno"> 93</span>&#160;<span class="preprocessor"> VEC_DATA_TYPE(DATA_TYPE, 2) \</span></div><div class="line"><a name="l00094"></a><span class="lineno"> 94</span>&#160;<span class="preprocessor"> src1 = (VEC_DATA_TYPE(DATA_TYPE, 2))( \</span></div><div class="line"><a name="l00095"></a><span class="lineno"> 95</span>&#160;<span class="preprocessor"> PTR_TO_VALUE(row_ptr + 8 * src_stride_y, DATA_TYPE), \</span></div><div class="line"><a name="l00096"></a><span class="lineno"> 96</span>&#160;<span class="preprocessor"> PTR_TO_VALUE(row_ptr + 9 * src_stride_y, DATA_TYPE)); \</span></div><div class="line"><a name="l00097"></a><span class="lineno"> 97</span>&#160;<span class="preprocessor"> VEC_DATA_TYPE(DATA_TYPE, 3) \</span></div><div class="line"><a name="l00098"></a><span class="lineno"> 98</span>&#160;<span class="preprocessor"> weights = (VEC_DATA_TYPE(DATA_TYPE, 3))( \</span></div><div class="line"><a name="l00099"></a><span class="lineno"> 99</span>&#160;<span class="preprocessor"> PTR_TO_VALUE((weights_ptr) + 0 * weights_stride_y, DATA_TYPE), \</span></div><div class="line"><a name="l00100"></a><span class="lineno"> 100</span>&#160;<span class="preprocessor"> PTR_TO_VALUE((weights_ptr) + 1 * weights_stride_y, DATA_TYPE), \</span></div><div class="line"><a name="l00101"></a><span class="lineno"> 101</span>&#160;<span class="preprocessor"> PTR_TO_VALUE((weights_ptr) + 2 * weights_stride_y, DATA_TYPE)); \</span></div><div class="line"><a name="l00102"></a><span class="lineno"> 102</span>&#160;<span class="preprocessor"> acc = ADD_OP(acc, MUL_OP(src0, (VEC_DATA_TYPE(DATA_TYPE, 8))weights.s0)); \</span></div><div class="line"><a name="l00103"></a><span class="lineno"> 103</span>&#160;<span class="preprocessor"> acc = ADD_OP(acc, MUL_OP((VEC_DATA_TYPE(DATA_TYPE, 8))(src0.s1234, src0.s567, src1.s0), (VEC_DATA_TYPE(DATA_TYPE, 8))weights.s1)); \</span></div><div class="line"><a name="l00104"></a><span class="lineno"> 104</span>&#160;<span class="preprocessor"> acc = ADD_OP(acc, MUL_OP((VEC_DATA_TYPE(DATA_TYPE, 8))(src0.s234, src0.s567, src1.s01), (VEC_DATA_TYPE(DATA_TYPE, 8))weights.s2)); \</span></div><div class="line"><a name="l00105"></a><span class="lineno"> 105</span>&#160;<span class="preprocessor"> }</span></div><div class="line"><a name="l00106"></a><span class="lineno"> 106</span>&#160;</div><div class="line"><a name="l00107"></a><span class="lineno"> 107</span>&#160;<span class="preprocessor">#define CONVOLUTION1x3_STRIDE_NHWC_STRIDE2(acc, row_ptr, weights_ptr) \</span></div><div class="line"><a name="l00108"></a><span class="lineno"> 108</span>&#160;<span class="preprocessor"> { \</span></div><div class="line"><a name="l00109"></a><span class="lineno"> 109</span>&#160;<span class="preprocessor"> VEC_DATA_TYPE(DATA_TYPE, 16) \</span></div><div class="line"><a name="l00110"></a><span class="lineno"> 110</span>&#160;<span class="preprocessor"> src0 = (VEC_DATA_TYPE(DATA_TYPE, 16))( \</span></div><div class="line"><a name="l00111"></a><span class="lineno"> 111</span>&#160;<span class="preprocessor"> PTR_TO_VALUE(row_ptr + 0 * src_stride_y, DATA_TYPE), \</span></div><div class="line"><a name="l00112"></a><span class="lineno"> 112</span>&#160;<span class="preprocessor"> PTR_TO_VALUE(row_ptr + 1 * src_stride_y, DATA_TYPE), \</span></div><div class="line"><a name="l00113"></a><span class="lineno"> 113</span>&#160;<span class="preprocessor"> PTR_TO_VALUE(row_ptr + 2 * src_stride_y, DATA_TYPE), \</span></div><div class="line"><a name="l00114"></a><span class="lineno"> 114</span>&#160;<span class="preprocessor"> PTR_TO_VALUE(row_ptr + 3 * src_stride_y, DATA_TYPE), \</span></div><div class="line"><a name="l00115"></a><span class="lineno"> 115</span>&#160;<span class="preprocessor"> PTR_TO_VALUE(row_ptr + 4 * src_stride_y, DATA_TYPE), \</span></div><div class="line"><a name="l00116"></a><span class="lineno"> 116</span>&#160;<span class="preprocessor"> PTR_TO_VALUE(row_ptr + 5 * src_stride_y, DATA_TYPE), \</span></div><div class="line"><a name="l00117"></a><span class="lineno"> 117</span>&#160;<span class="preprocessor"> PTR_TO_VALUE(row_ptr + 6 * src_stride_y, DATA_TYPE), \</span></div><div class="line"><a name="l00118"></a><span class="lineno"> 118</span>&#160;<span class="preprocessor"> PTR_TO_VALUE(row_ptr + 7 * src_stride_y, DATA_TYPE), \</span></div><div class="line"><a name="l00119"></a><span class="lineno"> 119</span>&#160;<span class="preprocessor"> PTR_TO_VALUE(row_ptr + 8 * src_stride_y, DATA_TYPE), \</span></div><div class="line"><a name="l00120"></a><span class="lineno"> 120</span>&#160;<span class="preprocessor"> PTR_TO_VALUE(row_ptr + 9 * src_stride_y, DATA_TYPE), \</span></div><div class="line"><a name="l00121"></a><span class="lineno"> 121</span>&#160;<span class="preprocessor"> PTR_TO_VALUE(row_ptr + 10 * src_stride_y, DATA_TYPE), \</span></div><div class="line"><a name="l00122"></a><span class="lineno"> 122</span>&#160;<span class="preprocessor"> PTR_TO_VALUE(row_ptr + 11 * src_stride_y, DATA_TYPE), \</span></div><div class="line"><a name="l00123"></a><span class="lineno"> 123</span>&#160;<span class="preprocessor"> PTR_TO_VALUE(row_ptr + 12 * src_stride_y, DATA_TYPE), \</span></div><div class="line"><a name="l00124"></a><span class="lineno"> 124</span>&#160;<span class="preprocessor"> PTR_TO_VALUE(row_ptr + 13 * src_stride_y, DATA_TYPE), \</span></div><div class="line"><a name="l00125"></a><span class="lineno"> 125</span>&#160;<span class="preprocessor"> PTR_TO_VALUE(row_ptr + 14 * src_stride_y, DATA_TYPE), \</span></div><div class="line"><a name="l00126"></a><span class="lineno"> 126</span>&#160;<span class="preprocessor"> PTR_TO_VALUE(row_ptr + 15 * src_stride_y, DATA_TYPE)); \</span></div><div class="line"><a name="l00127"></a><span class="lineno"> 127</span>&#160;<span class="preprocessor"> DATA_TYPE src1 = PTR_TO_VALUE(row_ptr + 16 * src_stride_y, DATA_TYPE); \</span></div><div class="line"><a name="l00128"></a><span class="lineno"> 128</span>&#160;<span class="preprocessor"> VEC_DATA_TYPE(DATA_TYPE, 3) \</span></div><div class="line"><a name="l00129"></a><span class="lineno"> 129</span>&#160;<span class="preprocessor"> weights = (VEC_DATA_TYPE(DATA_TYPE, 3))( \</span></div><div class="line"><a name="l00130"></a><span class="lineno"> 130</span>&#160;<span class="preprocessor"> PTR_TO_VALUE((weights_ptr) + 0 * weights_stride_y, DATA_TYPE), \</span></div><div class="line"><a name="l00131"></a><span class="lineno"> 131</span>&#160;<span class="preprocessor"> PTR_TO_VALUE((weights_ptr) + 1 * weights_stride_y, DATA_TYPE), \</span></div><div class="line"><a name="l00132"></a><span class="lineno"> 132</span>&#160;<span class="preprocessor"> PTR_TO_VALUE((weights_ptr) + 2 * weights_stride_y, DATA_TYPE)); \</span></div><div class="line"><a name="l00133"></a><span class="lineno"> 133</span>&#160;<span class="preprocessor"> \</span></div><div class="line"><a name="l00134"></a><span class="lineno"> 134</span>&#160;<span class="preprocessor"> acc = ADD_OP(acc, MUL_OP(src0.s02468ACE, (VEC_DATA_TYPE(DATA_TYPE, 8))weights.s0)); \</span></div><div class="line"><a name="l00135"></a><span class="lineno"> 135</span>&#160;<span class="preprocessor"> acc = ADD_OP(acc, MUL_OP((VEC_DATA_TYPE(DATA_TYPE, 8))(src0.s1357, src0.s9BDF), (VEC_DATA_TYPE(DATA_TYPE, 8))weights.s1)); \</span></div><div class="line"><a name="l00136"></a><span class="lineno"> 136</span>&#160;<span class="preprocessor"> acc = ADD_OP(acc, MUL_OP((VEC_DATA_TYPE(DATA_TYPE, 8))(src0.s2468, src0.sACE, src1), (VEC_DATA_TYPE(DATA_TYPE, 8))weights.s2)); \</span></div><div class="line"><a name="l00137"></a><span class="lineno"> 137</span>&#160;<span class="preprocessor"> }</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">/** This kernel performs a direct convolution to convolve the low three dimensions.</span></div><div class="line"><a name="l00140"></a><span class="lineno"> 140</span>&#160;<span class="comment"> *</span></div><div class="line"><a name="l00141"></a><span class="lineno"> 141</span>&#160;<span class="comment"> * @note This OpenCL kernel works with stride_x = 1 and 2</span></div><div class="line"><a name="l00142"></a><span class="lineno"> 142</span>&#160;<span class="comment"> * @note The data type must be passed at compile time using -DDATA_TYPE: e.g. -DDATA_TYPE=float</span></div><div class="line"><a name="l00143"></a><span class="lineno"> 143</span>&#160;<span class="comment"> * @note The third dimensions of the weights tensors must be passed at compile time using -DWEIGHTS_DEPTH</span></div><div class="line"><a name="l00144"></a><span class="lineno"> 144</span>&#160;<span class="comment"> * @note If biases are used then -DHAS_BIAS has to be passed at compile time</span></div><div class="line"><a name="l00145"></a><span class="lineno"> 145</span>&#160;<span class="comment"> *</span></div><div class="line"><a name="l00146"></a><span class="lineno"> 146</span>&#160;<span class="comment"> * @param[in] src_ptr Pointer to the source tensor. Supported data types: QS8/QS16/F16/F32</span></div><div class="line"><a name="l00147"></a><span class="lineno"> 147</span>&#160;<span class="comment"> * @param[in] src_stride_x Stride of the source tensor in X dimension (in bytes)</span></div><div class="line"><a name="l00148"></a><span class="lineno"> 148</span>&#160;<span class="comment"> * @param[in] src_step_x src_stride_x * number of elements along X processed per workitem(in bytes)</span></div><div class="line"><a name="l00149"></a><span class="lineno"> 149</span>&#160;<span class="comment"> * @param[in] src_stride_y Stride of the source tensor in Y dimension (in bytes)</span></div><div class="line"><a name="l00150"></a><span class="lineno"> 150</span>&#160;<span class="comment"> * @param[in] src_step_y src_stride_y * number of elements along Y processed per workitem(in bytes)</span></div><div class="line"><a name="l00151"></a><span class="lineno"> 151</span>&#160;<span class="comment"> * @param[in] src_stride_z Stride of the source tensor in Z dimension (in bytes)</span></div><div class="line"><a name="l00152"></a><span class="lineno"> 152</span>&#160;<span class="comment"> * @param[in] src_step_z src_stride_z * number of elements along Z processed per workitem(in bytes)</span></div><div class="line"><a name="l00153"></a><span class="lineno"> 153</span>&#160;<span class="comment"> * @param[in] src_offset_first_element_in_bytes The offset of the first element in the source tensor</span></div><div class="line"><a name="l00154"></a><span class="lineno"> 154</span>&#160;<span class="comment"> * @param[out] dst_ptr Pointer to the destination tensor. Supported data types: same as @p src_ptr</span></div><div class="line"><a name="l00155"></a><span class="lineno"> 155</span>&#160;<span class="comment"> * @param[in] dst_stride_x Stride of the destination tensor in X dimension (in bytes)</span></div><div class="line"><a name="l00156"></a><span class="lineno"> 156</span>&#160;<span class="comment"> * @param[in] dst_step_x dst_stride_x * number of elements along X processed per workitem(in bytes)</span></div><div class="line"><a name="l00157"></a><span class="lineno"> 157</span>&#160;<span class="comment"> * @param[in] dst_stride_y Stride of the destination tensor in Y dimension (in bytes)</span></div><div class="line"><a name="l00158"></a><span class="lineno"> 158</span>&#160;<span class="comment"> * @param[in] dst_step_y dst_stride_y * number of elements along Z processed per workitem(in bytes)</span></div><div class="line"><a name="l00159"></a><span class="lineno"> 159</span>&#160;<span class="comment"> * @param[in] dst_stride_z Stride of the destination tensor in Z dimension (in bytes)</span></div><div class="line"><a name="l00160"></a><span class="lineno"> 160</span>&#160;<span class="comment"> * @param[in] dst_step_z dst_stride_z * number of elements along Z processed per workitem(in bytes)</span></div><div class="line"><a name="l00161"></a><span class="lineno"> 161</span>&#160;<span class="comment"> * @param[in] dst_offset_first_element_in_bytes The offset of the first element in the destination tensor</span></div><div class="line"><a name="l00162"></a><span class="lineno"> 162</span>&#160;<span class="comment"> * @param[in] weights_ptr Pointer to the weights tensor. Supported data types: same as @p src_ptr</span></div><div class="line"><a name="l00163"></a><span class="lineno"> 163</span>&#160;<span class="comment"> * @param[in] weights_stride_x Stride of the weights tensor in X dimension (in bytes)</span></div><div class="line"><a name="l00164"></a><span class="lineno"> 164</span>&#160;<span class="comment"> * @param[in] weights_step_x weights_stride_x * number of elements along X processed per workitem(in bytes)</span></div><div class="line"><a name="l00165"></a><span class="lineno"> 165</span>&#160;<span class="comment"> * @param[in] weights_stride_y Stride of the weights tensor in Y dimension (in bytes)</span></div><div class="line"><a name="l00166"></a><span class="lineno"> 166</span>&#160;<span class="comment"> * @param[in] weights_step_y weights_stride_y * number of elements along y processed per workitem(in bytes)</span></div><div class="line"><a name="l00167"></a><span class="lineno"> 167</span>&#160;<span class="comment"> * @param[in] weights_stride_z Stride of the weights tensor in Z dimension (in bytes)</span></div><div class="line"><a name="l00168"></a><span class="lineno"> 168</span>&#160;<span class="comment"> * @param[in] weights_step_z weights_stride_z * number of elements along Z processed per workitem(in bytes)</span></div><div class="line"><a name="l00169"></a><span class="lineno"> 169</span>&#160;<span class="comment"> * @param[in] weights_offset_first_element_in_bytes The offset of the first element in the weights tensor</span></div><div class="line"><a name="l00170"></a><span class="lineno"> 170</span>&#160;<span class="comment"> * @param[in] biases_ptr Pointer to the biases tensor. Same as @p src_ptr</span></div><div class="line"><a name="l00171"></a><span class="lineno"> 171</span>&#160;<span class="comment"> * @param[in] biases_stride_x Stride of the biases tensor in X dimension (in bytes)</span></div><div class="line"><a name="l00172"></a><span class="lineno"> 172</span>&#160;<span class="comment"> * @param[in] biases_step_x biases_stride_x * number of elements along X processed per workitem(in bytes)</span></div><div class="line"><a name="l00173"></a><span class="lineno"> 173</span>&#160;<span class="comment"> * @param[in] biases_offset_first_element_in_bytes The offset of the first element in the biases tensor</span></div><div class="line"><a name="l00174"></a><span class="lineno"> 174</span>&#160;<span class="comment"> * @param[in] weights_stride_w Stride of the weights tensor in the 4th dimension</span></div><div class="line"><a name="l00175"></a><span class="lineno"> 175</span>&#160;<span class="comment"> */</span></div><div class="line"><a name="l00176"></a><span class="lineno"> 176</span>&#160;__kernel <span class="keywordtype">void</span> direct_convolution3x3_nhwc(</div><div class="line"><a name="l00177"></a><span class="lineno"> 177</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#a989ab3e96426615bb98e04e0235088ca">src</a>),</div><div class="line"><a name="l00178"></a><span class="lineno"> 178</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="l00179"></a><span class="lineno"> 179</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#a64a08a9fec5aeee8650e7182b6d171d0">weights</a>),</div><div class="line"><a name="l00180"></a><span class="lineno"> 180</span>&#160;#ifdef HAS_BIAS</div><div class="line"><a name="l00181"></a><span class="lineno"> 181</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="l00182"></a><span class="lineno"> 182</span>&#160;#endif <span class="comment">/* defined(HAS_BIAS) */</span></div><div class="line"><a name="l00183"></a><span class="lineno"> 183</span>&#160; <span class="keywordtype">unsigned</span> <span class="keywordtype">int</span> weights_stride_w)</div><div class="line"><a name="l00184"></a><span class="lineno"> 184</span>&#160;{</div><div class="line"><a name="l00185"></a><span class="lineno"> 185</span>&#160; <a class="code" href="struct_image.xhtml">Image</a> <a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#a989ab3e96426615bb98e04e0235088ca">src</a> = <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#aebe814363556c244be043b13e7969197">CONVERT_TO_IMAGE_STRUCT</a>(<a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#a989ab3e96426615bb98e04e0235088ca">src</a>);</div><div class="line"><a name="l00186"></a><span class="lineno"> 186</span>&#160; <a class="code" href="struct_tensor3_d.xhtml">Tensor3D</a> <a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#a64a08a9fec5aeee8650e7182b6d171d0">weights</a> = <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a79e8e562daa6599317d2d1cd86ef1bf2">CONVERT_TO_TENSOR3D_STRUCT_NO_STEP</a>(<a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#a64a08a9fec5aeee8650e7182b6d171d0">weights</a>);</div><div class="line"><a name="l00187"></a><span class="lineno"> 187</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="l00188"></a><span class="lineno"> 188</span>&#160;</div><div class="line"><a name="l00189"></a><span class="lineno"> 189</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(DATA_TYPE_PROMOTED, 8)</div><div class="line"><a name="l00190"></a><span class="lineno"> 190</span>&#160; values0 = 0;</div><div class="line"><a name="l00191"></a><span class="lineno"> 191</span>&#160; <span class="keyword">const</span> <span class="keywordtype">int</span> id0 = get_global_id(0);</div><div class="line"><a name="l00192"></a><span class="lineno"> 192</span>&#160; <span class="keyword">const</span> <span class="keywordtype">int</span> id1 = get_global_id(1);</div><div class="line"><a name="l00193"></a><span class="lineno"> 193</span>&#160; <span class="keyword">const</span> <span class="keywordtype">int</span> id2 = get_global_id(2);</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; __global uchar *weights_addr = (__global uchar *)<a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a2101b2fe0193ce227ae4e0945e321d85">tensor3D_offset</a>(&amp;<a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#a64a08a9fec5aeee8650e7182b6d171d0">weights</a>, 0, 0, 0);</div><div class="line"><a name="l00196"></a><span class="lineno"> 196</span>&#160; __global uchar *src_addr = (__global uchar *)<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#a989ab3e96426615bb98e04e0235088ca">src</a>, 0, 0) - src_stride_x * id0 + ((id2 * STRIDE_Y) - PAD_TOP) * (int)src_stride_z;</div><div class="line"><a name="l00197"></a><span class="lineno"> 197</span>&#160;</div><div class="line"><a name="l00198"></a><span class="lineno"> 198</span>&#160; weights_addr += id0 * weights_stride_w;</div><div class="line"><a name="l00199"></a><span class="lineno"> 199</span>&#160;</div><div class="line"><a name="l00200"></a><span class="lineno"> 200</span>&#160; <span class="keyword">const</span> <span class="keywordtype">int</span> coordy = ((id2 * STRIDE_Y) - PAD_TOP);</div><div class="line"><a name="l00201"></a><span class="lineno"> 201</span>&#160; <span class="keywordflow">for</span>(<span class="keyword">volatile</span> <span class="keywordtype">int</span> d = 0; d &lt; WEIGHTS_DEPTH; ++d)</div><div class="line"><a name="l00202"></a><span class="lineno"> 202</span>&#160; {</div><div class="line"><a name="l00203"></a><span class="lineno"> 203</span>&#160;<span class="preprocessor">#if PAD_TOP &gt; 0</span></div><div class="line"><a name="l00204"></a><span class="lineno"> 204</span>&#160; <span class="keywordflow">if</span>(coordy &lt; 0) <span class="comment">// special case Z = -1 doesn&#39;t exists</span></div><div class="line"><a name="l00205"></a><span class="lineno"> 205</span>&#160; {</div><div class="line"><a name="l00206"></a><span class="lineno"> 206</span>&#160; <span class="comment">//skip first row and load the two next ones</span></div><div class="line"><a name="l00207"></a><span class="lineno"> 207</span>&#160; CONVOLUTION1x3_NHWC(values0, src_addr + 1 * (<span class="keywordtype">int</span>)src_stride_z, (weights_addr + 1 * (<span class="keywordtype">int</span>)weights_stride_z));</div><div class="line"><a name="l00208"></a><span class="lineno"> 208</span>&#160; CONVOLUTION1x3_NHWC(values0, src_addr + 2 * (<span class="keywordtype">int</span>)src_stride_z, (weights_addr + 2 * (<span class="keywordtype">int</span>)weights_stride_z));</div><div class="line"><a name="l00209"></a><span class="lineno"> 209</span>&#160; }</div><div class="line"><a name="l00210"></a><span class="lineno"> 210</span>&#160; <span class="keywordflow">else</span> <span class="keywordflow">if</span>(coordy == (SRC_HEIGHT - PAD_TOP - 1))</div><div class="line"><a name="l00211"></a><span class="lineno"> 211</span>&#160; {</div><div class="line"><a name="l00212"></a><span class="lineno"> 212</span>&#160; <span class="comment">// special case when computing the last row of the output we must read the last three rows from the input buffer (including padding) but the</span></div><div class="line"><a name="l00213"></a><span class="lineno"> 213</span>&#160; <span class="comment">// Z axis has no padding at all.</span></div><div class="line"><a name="l00214"></a><span class="lineno"> 214</span>&#160; CONVOLUTION1x3_NHWC(values0, src_addr, (weights_addr + 0 * (<span class="keywordtype">int</span>)weights_stride_z));</div><div class="line"><a name="l00215"></a><span class="lineno"> 215</span>&#160; CONVOLUTION1x3_NHWC(values0, src_addr + 1 * (<span class="keywordtype">int</span>)src_stride_z, (weights_addr + 1 * (<span class="keywordtype">int</span>)weights_stride_z));</div><div class="line"><a name="l00216"></a><span class="lineno"> 216</span>&#160; }</div><div class="line"><a name="l00217"></a><span class="lineno"> 217</span>&#160; <span class="keywordflow">else</span></div><div class="line"><a name="l00218"></a><span class="lineno"> 218</span>&#160; {</div><div class="line"><a name="l00219"></a><span class="lineno"> 219</span>&#160; CONVOLUTION1x3_NHWC(values0, src_addr, (weights_addr + 0 * (<span class="keywordtype">int</span>)weights_stride_z));</div><div class="line"><a name="l00220"></a><span class="lineno"> 220</span>&#160; CONVOLUTION1x3_NHWC(values0, src_addr + 1 * (<span class="keywordtype">int</span>)src_stride_z, (weights_addr + 1 * (<span class="keywordtype">int</span>)weights_stride_z));</div><div class="line"><a name="l00221"></a><span class="lineno"> 221</span>&#160; CONVOLUTION1x3_NHWC(values0, src_addr + 2 * (<span class="keywordtype">int</span>)src_stride_z, (weights_addr + 2 * (<span class="keywordtype">int</span>)weights_stride_z));</div><div class="line"><a name="l00222"></a><span class="lineno"> 222</span>&#160; }</div><div class="line"><a name="l00223"></a><span class="lineno"> 223</span>&#160;<span class="preprocessor">#else // PAD_TOP &gt; 0</span></div><div class="line"><a name="l00224"></a><span class="lineno"> 224</span>&#160; CONVOLUTION1x3_NHWC(values0, src_addr, (weights_addr + 0 * (<span class="keywordtype">int</span>)weights_stride_z));</div><div class="line"><a name="l00225"></a><span class="lineno"> 225</span>&#160; CONVOLUTION1x3_NHWC(values0, src_addr + 1 * (<span class="keywordtype">int</span>)src_stride_z, (weights_addr + 1 * (<span class="keywordtype">int</span>)weights_stride_z));</div><div class="line"><a name="l00226"></a><span class="lineno"> 226</span>&#160; CONVOLUTION1x3_NHWC(values0, src_addr + 2 * (<span class="keywordtype">int</span>)src_stride_z, (weights_addr + 2 * (<span class="keywordtype">int</span>)weights_stride_z));</div><div class="line"><a name="l00227"></a><span class="lineno"> 227</span>&#160;<span class="preprocessor">#endif // PAD_TOP &gt; 0</span></div><div class="line"><a name="l00228"></a><span class="lineno"> 228</span>&#160; src_addr += src_stride_x;</div><div class="line"><a name="l00229"></a><span class="lineno"> 229</span>&#160; weights_addr += weights_stride_x;</div><div class="line"><a name="l00230"></a><span class="lineno"> 230</span>&#160; }</div><div class="line"><a name="l00231"></a><span class="lineno"> 231</span>&#160;</div><div class="line"><a name="l00232"></a><span class="lineno"> 232</span>&#160;<span class="preprocessor">#ifdef HAS_BIAS</span></div><div class="line"><a name="l00233"></a><span class="lineno"> 233</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#a64d779f80eeb923e0ab2313433f7b40b">CONVERT_TO_VECTOR_STRUCT_NO_STEP</a>(biases);</div><div class="line"><a name="l00234"></a><span class="lineno"> 234</span>&#160; values0 = <a class="code" href="direct__convolution3x3_8cl.xhtml#aebbeb1f22eca3a3f4c3e019e8f419f39">ADD_OP</a>(values0, (<a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(DATA_TYPE_PROMOTED, 8)) * ((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(<a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a7e4940407322d6f0ccb8b6b86b856019">vector_offset</a>(&amp;biases, id0))));</div><div class="line"><a name="l00235"></a><span class="lineno"> 235</span>&#160;<span class="preprocessor">#endif </span><span class="comment">/* defined(HAS_BIAS) */</span><span class="preprocessor"></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; *((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(<a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#adbf67dcee294e673cf796f1ed8aeb6a4">dst</a>.ptr + 0 * dst_stride_y)) = values0.s0;</div><div class="line"><a name="l00238"></a><span class="lineno"> 238</span>&#160; *((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(<a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#adbf67dcee294e673cf796f1ed8aeb6a4">dst</a>.ptr + 1 * dst_stride_y)) = values0.s1;</div><div class="line"><a name="l00239"></a><span class="lineno"> 239</span>&#160; *((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(<a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#adbf67dcee294e673cf796f1ed8aeb6a4">dst</a>.ptr + 2 * dst_stride_y)) = values0.s2;</div><div class="line"><a name="l00240"></a><span class="lineno"> 240</span>&#160; *((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(<a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#adbf67dcee294e673cf796f1ed8aeb6a4">dst</a>.ptr + 3 * dst_stride_y)) = values0.s3;</div><div class="line"><a name="l00241"></a><span class="lineno"> 241</span>&#160; *((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(<a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#adbf67dcee294e673cf796f1ed8aeb6a4">dst</a>.ptr + 4 * dst_stride_y)) = values0.s4;</div><div class="line"><a name="l00242"></a><span class="lineno"> 242</span>&#160; *((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(<a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#adbf67dcee294e673cf796f1ed8aeb6a4">dst</a>.ptr + 5 * dst_stride_y)) = values0.s5;</div><div class="line"><a name="l00243"></a><span class="lineno"> 243</span>&#160; *((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(<a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#adbf67dcee294e673cf796f1ed8aeb6a4">dst</a>.ptr + 6 * dst_stride_y)) = values0.s6;</div><div class="line"><a name="l00244"></a><span class="lineno"> 244</span>&#160; *((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(<a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#adbf67dcee294e673cf796f1ed8aeb6a4">dst</a>.ptr + 7 * dst_stride_y)) = values0.s7;</div><div class="line"><a name="l00245"></a><span class="lineno"> 245</span>&#160;}</div><div class="line"><a name="l00246"></a><span class="lineno"> 246</span>&#160;<span class="preprocessor">#endif // defined(DATA_LAYOUT_NHWC)</span></div><div class="line"><a name="l00247"></a><span class="lineno"> 247</span>&#160;<span class="comment"></span></div><div class="line"><a name="l00248"></a><span class="lineno"> 248</span>&#160;<span class="comment">/** This kernel performs a direct convolution to convolve the low three dimensions.</span></div><div class="line"><a name="l00249"></a><span class="lineno"> 249</span>&#160;<span class="comment"> *</span></div><div class="line"><a name="l00250"></a><span class="lineno"> 250</span>&#160;<span class="comment"> * @note This OpenCL kernel works with stride_x = 1 and 2</span></div><div class="line"><a name="l00251"></a><span class="lineno"> 251</span>&#160;<span class="comment"> * @note The data type must be passed at compile time using -DDATA_TYPE: e.g. -DDATA_TYPE=float</span></div><div class="line"><a name="l00252"></a><span class="lineno"> 252</span>&#160;<span class="comment"> * @note The third dimensions of the weights tensors must be passed at compile time using -DWEIGHTS_DEPTH</span></div><div class="line"><a name="l00253"></a><span class="lineno"> 253</span>&#160;<span class="comment"> * @note If biases are used then -DHAS_BIAS has to be passed at compile time</span></div><div class="line"><a name="l00254"></a><span class="lineno"> 254</span>&#160;<span class="comment"> *</span></div><div class="line"><a name="l00255"></a><span class="lineno"> 255</span>&#160;<span class="comment"> * @param[in] src_ptr Pointer to the source tensor. Supported data types: F16/F32</span></div><div class="line"><a name="l00256"></a><span class="lineno"> 256</span>&#160;<span class="comment"> * @param[in] src_stride_x Stride of the source tensor in X dimension (in bytes)</span></div><div class="line"><a name="l00257"></a><span class="lineno"> 257</span>&#160;<span class="comment"> * @param[in] src_step_x src_stride_x * number of elements along X processed per workitem(in bytes)</span></div><div class="line"><a name="l00258"></a><span class="lineno"> 258</span>&#160;<span class="comment"> * @param[in] src_stride_y Stride of the source tensor in Y dimension (in bytes)</span></div><div class="line"><a name="l00259"></a><span class="lineno"> 259</span>&#160;<span class="comment"> * @param[in] src_step_y src_stride_y * number of elements along Y processed per workitem(in bytes)</span></div><div class="line"><a name="l00260"></a><span class="lineno"> 260</span>&#160;<span class="comment"> * @param[in] src_stride_z Stride of the source tensor in Z dimension (in bytes)</span></div><div class="line"><a name="l00261"></a><span class="lineno"> 261</span>&#160;<span class="comment"> * @param[in] src_step_z src_stride_z * number of elements along Z processed per workitem(in bytes)</span></div><div class="line"><a name="l00262"></a><span class="lineno"> 262</span>&#160;<span class="comment"> * @param[in] src_offset_first_element_in_bytes The offset of the first element in the source tensor</span></div><div class="line"><a name="l00263"></a><span class="lineno"> 263</span>&#160;<span class="comment"> * @param[out] dst_ptr Pointer to the destination tensor. Supported data types: same as @p src_ptr</span></div><div class="line"><a name="l00264"></a><span class="lineno"> 264</span>&#160;<span class="comment"> * @param[in] dst_stride_x Stride of the destination tensor in X dimension (in bytes)</span></div><div class="line"><a name="l00265"></a><span class="lineno"> 265</span>&#160;<span class="comment"> * @param[in] dst_step_x dst_stride_x * number of elements along X processed per workitem(in bytes)</span></div><div class="line"><a name="l00266"></a><span class="lineno"> 266</span>&#160;<span class="comment"> * @param[in] dst_stride_y Stride of the destination tensor in Y dimension (in bytes)</span></div><div class="line"><a name="l00267"></a><span class="lineno"> 267</span>&#160;<span class="comment"> * @param[in] dst_step_y dst_stride_y * number of elements along Z processed per workitem(in bytes)</span></div><div class="line"><a name="l00268"></a><span class="lineno"> 268</span>&#160;<span class="comment"> * @param[in] dst_stride_z Stride of the destination tensor in Z dimension (in bytes)</span></div><div class="line"><a name="l00269"></a><span class="lineno"> 269</span>&#160;<span class="comment"> * @param[in] dst_step_z dst_stride_z * number of elements along Z processed per workitem(in bytes)</span></div><div class="line"><a name="l00270"></a><span class="lineno"> 270</span>&#160;<span class="comment"> * @param[in] dst_offset_first_element_in_bytes The offset of the first element in the destination tensor</span></div><div class="line"><a name="l00271"></a><span class="lineno"> 271</span>&#160;<span class="comment"> * @param[in] weights_ptr Pointer to the weights tensor. Supported data types: same as @p src_ptr</span></div><div class="line"><a name="l00272"></a><span class="lineno"> 272</span>&#160;<span class="comment"> * @param[in] weights_stride_x Stride of the weights tensor in X dimension (in bytes)</span></div><div class="line"><a name="l00273"></a><span class="lineno"> 273</span>&#160;<span class="comment"> * @param[in] weights_step_x weights_stride_x * number of elements along X processed per workitem(in bytes)</span></div><div class="line"><a name="l00274"></a><span class="lineno"> 274</span>&#160;<span class="comment"> * @param[in] weights_stride_y Stride of the weights tensor in Y dimension (in bytes)</span></div><div class="line"><a name="l00275"></a><span class="lineno"> 275</span>&#160;<span class="comment"> * @param[in] weights_step_y weights_stride_y * number of elements along y processed per workitem(in bytes)</span></div><div class="line"><a name="l00276"></a><span class="lineno"> 276</span>&#160;<span class="comment"> * @param[in] weights_stride_z Stride of the weights tensor in Z dimension (in bytes)</span></div><div class="line"><a name="l00277"></a><span class="lineno"> 277</span>&#160;<span class="comment"> * @param[in] weights_step_z weights_stride_z * number of elements along Z processed per workitem(in bytes)</span></div><div class="line"><a name="l00278"></a><span class="lineno"> 278</span>&#160;<span class="comment"> * @param[in] weights_offset_first_element_in_bytes The offset of the first element in the weights tensor</span></div><div class="line"><a name="l00279"></a><span class="lineno"> 279</span>&#160;<span class="comment"> * @param[in] biases_ptr Pointer to the biases tensor. Same as @p src_ptr</span></div><div class="line"><a name="l00280"></a><span class="lineno"> 280</span>&#160;<span class="comment"> * @param[in] biases_stride_x Stride of the biases tensor in X dimension (in bytes)</span></div><div class="line"><a name="l00281"></a><span class="lineno"> 281</span>&#160;<span class="comment"> * @param[in] biases_step_x biases_stride_x * number of elements along X processed per workitem(in bytes)</span></div><div class="line"><a name="l00282"></a><span class="lineno"> 282</span>&#160;<span class="comment"> * @param[in] biases_offset_first_element_in_bytes The offset of the first element in the biases tensor</span></div><div class="line"><a name="l00283"></a><span class="lineno"> 283</span>&#160;<span class="comment"> * @param[in] weights_stride_w Stride of the weights tensor in the 4th dimension</span></div><div class="line"><a name="l00284"></a><span class="lineno"> 284</span>&#160;<span class="comment"> */</span></div><div class="line"><a name="l00285"></a><span class="lineno"> 285</span>&#160;__kernel <span class="keywordtype">void</span> direct_convolution3x3(</div><div class="line"><a name="l00286"></a><span class="lineno"> 286</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#a989ab3e96426615bb98e04e0235088ca">src</a>),</div><div class="line"><a name="l00287"></a><span class="lineno"> 287</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="l00288"></a><span class="lineno"> 288</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#a64a08a9fec5aeee8650e7182b6d171d0">weights</a>),</div><div class="line"><a name="l00289"></a><span class="lineno"> 289</span>&#160;#ifdef HAS_BIAS</div><div class="line"><a name="l00290"></a><span class="lineno"> 290</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="l00291"></a><span class="lineno"> 291</span>&#160;#endif <span class="comment">/* defined(HAS_BIAS) */</span></div><div class="line"><a name="l00292"></a><span class="lineno"> 292</span>&#160; <span class="keywordtype">unsigned</span> <span class="keywordtype">int</span> weights_stride_w)</div><div class="line"><a name="l00293"></a><span class="lineno"> 293</span>&#160;{</div><div class="line"><a name="l00294"></a><span class="lineno"> 294</span>&#160; <a class="code" href="struct_image.xhtml">Image</a> <a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#a989ab3e96426615bb98e04e0235088ca">src</a> = <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#aebe814363556c244be043b13e7969197">CONVERT_TO_IMAGE_STRUCT</a>(<a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#a989ab3e96426615bb98e04e0235088ca">src</a>);</div><div class="line"><a name="l00295"></a><span class="lineno"> 295</span>&#160; <a class="code" href="struct_tensor3_d.xhtml">Tensor3D</a> <a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#a64a08a9fec5aeee8650e7182b6d171d0">weights</a> = <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a79e8e562daa6599317d2d1cd86ef1bf2">CONVERT_TO_TENSOR3D_STRUCT_NO_STEP</a>(<a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#a64a08a9fec5aeee8650e7182b6d171d0">weights</a>);</div><div class="line"><a name="l00296"></a><span class="lineno"> 296</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="l00297"></a><span class="lineno"> 297</span>&#160;</div><div class="line"><a name="l00298"></a><span class="lineno"> 298</span>&#160; <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(DATA_TYPE_PROMOTED, 8)</div><div class="line"><a name="l00299"></a><span class="lineno"> 299</span>&#160; values0 = 0;</div><div class="line"><a name="l00300"></a><span class="lineno"> 300</span>&#160;</div><div class="line"><a name="l00301"></a><span class="lineno"> 301</span>&#160; __global uchar *weights_addr = (__global uchar *)<a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a2101b2fe0193ce227ae4e0945e321d85">tensor3D_offset</a>(&amp;<a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#a64a08a9fec5aeee8650e7182b6d171d0">weights</a>, 0, 0, 0);</div><div class="line"><a name="l00302"></a><span class="lineno"> 302</span>&#160; __global uchar *src_addr = (__global uchar *)<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#a989ab3e96426615bb98e04e0235088ca">src</a>, 0, 0);</div><div class="line"><a name="l00303"></a><span class="lineno"> 303</span>&#160;</div><div class="line"><a name="l00304"></a><span class="lineno"> 304</span>&#160; const <span class="keywordtype">int</span> kernel_index = get_global_id(2);</div><div class="line"><a name="l00305"></a><span class="lineno"> 305</span>&#160; weights_addr += kernel_index * weights_stride_w;</div><div class="line"><a name="l00306"></a><span class="lineno"> 306</span>&#160;</div><div class="line"><a name="l00307"></a><span class="lineno"> 307</span>&#160; <a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#a0886942393a3ba0dfefaa7516b159784">for</a>(volatile <span class="keywordtype">int</span> d = 0; d &lt; WEIGHTS_DEPTH; ++d)</div><div class="line"><a name="l00308"></a><span class="lineno"> 308</span>&#160; {</div><div class="line"><a name="l00309"></a><span class="lineno"> 309</span>&#160; CONVOLUTION1x3(values0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src_addr + 0 * src_stride_y), (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(weights_addr + 0 * weights_stride_y));</div><div class="line"><a name="l00310"></a><span class="lineno"> 310</span>&#160; CONVOLUTION1x3(values0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src_addr + 1 * src_stride_y), (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(weights_addr + 1 * weights_stride_y));</div><div class="line"><a name="l00311"></a><span class="lineno"> 311</span>&#160; CONVOLUTION1x3(values0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(src_addr + 2 * src_stride_y), (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(weights_addr + 2 * weights_stride_y));</div><div class="line"><a name="l00312"></a><span class="lineno"> 312</span>&#160;</div><div class="line"><a name="l00313"></a><span class="lineno"> 313</span>&#160; src_addr += src_stride_z;</div><div class="line"><a name="l00314"></a><span class="lineno"> 314</span>&#160; weights_addr += weights_stride_z;</div><div class="line"><a name="l00315"></a><span class="lineno"> 315</span>&#160; }</div><div class="line"><a name="l00316"></a><span class="lineno"> 316</span>&#160;</div><div class="line"><a name="l00317"></a><span class="lineno"> 317</span>&#160;<span class="preprocessor">#ifdef HAS_BIAS</span></div><div class="line"><a name="l00318"></a><span class="lineno"> 318</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#a64d779f80eeb923e0ab2313433f7b40b">CONVERT_TO_VECTOR_STRUCT_NO_STEP</a>(biases);</div><div class="line"><a name="l00319"></a><span class="lineno"> 319</span>&#160;</div><div class="line"><a name="l00320"></a><span class="lineno"> 320</span>&#160; values0 = <a class="code" href="direct__convolution3x3_8cl.xhtml#aebbeb1f22eca3a3f4c3e019e8f419f39">ADD_OP</a>(values0, (<a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(DATA_TYPE_PROMOTED, 8)) * ((__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)(<a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a7e4940407322d6f0ccb8b6b86b856019">vector_offset</a>(&amp;biases, kernel_index))));</div><div class="line"><a name="l00321"></a><span class="lineno"> 321</span>&#160;<span class="preprocessor">#endif </span><span class="comment">/* defined(HAS_BIAS) */</span><span class="preprocessor"></span></div><div class="line"><a name="l00322"></a><span class="lineno"> 322</span>&#160;</div><div class="line"><a name="l00323"></a><span class="lineno"> 323</span>&#160; vstore8(<a class="code" href="direct__convolution3x3_8cl.xhtml#a1f15728672380ade7a238f5e783d54d2">CONVERT_SAT</a>(values0, <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>, 8)), 0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)<a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#adbf67dcee294e673cf796f1ed8aeb6a4">dst</a>.ptr);</div><div class="line"><a name="l00324"></a><span class="lineno"> 324</span>&#160;}</div><div class="line"><a name="l00325"></a><span class="lineno"> 325</span>&#160;<span class="preprocessor">#endif //defined(DATA_TYPE) &amp;&amp; defined(STRIDE_X) &amp;&amp; defined(WEIGHTS_DEPTH)</span></div><div class="line"><a name="l00326"></a><span class="lineno"> 326</span>&#160;</div><div class="line"><a name="l00327"></a><span class="lineno"> 327</span>&#160;<span class="preprocessor">#if defined(WEIGHTS_DEPTH)</span></div><div class="line"><a name="l00328"></a><span class="lineno"> 328</span>&#160;</div><div class="line"><a name="l00329"></a><span class="lineno"> 329</span>&#160;<span class="preprocessor">#define CONVOLUTION1x3_BIFROST(acc, src0, src1, weights_row0) \</span></div><div class="line"><a name="l00330"></a><span class="lineno"> 330</span>&#160;<span class="preprocessor"> ({ \</span></div><div class="line"><a name="l00331"></a><span class="lineno"> 331</span>&#160;<span class="preprocessor"> acc.s0 = mad(src0.s0, weights_row0.s0, acc.s0); \</span></div><div class="line"><a name="l00332"></a><span class="lineno"> 332</span>&#160;<span class="preprocessor"> acc.s1 = mad(src0.s1, weights_row0.s0, acc.s1); \</span></div><div class="line"><a name="l00333"></a><span class="lineno"> 333</span>&#160;<span class="preprocessor"> acc.s2 = mad(src0.s2, weights_row0.s0, acc.s2); \</span></div><div class="line"><a name="l00334"></a><span class="lineno"> 334</span>&#160;<span class="preprocessor"> acc.s3 = mad(src0.s3, weights_row0.s0, acc.s3); \</span></div><div class="line"><a name="l00335"></a><span class="lineno"> 335</span>&#160;<span class="preprocessor"> acc.s0 = mad(src0.s1, weights_row0.s1, acc.s0); \</span></div><div class="line"><a name="l00336"></a><span class="lineno"> 336</span>&#160;<span class="preprocessor"> acc.s1 = mad(src0.s2, weights_row0.s1, acc.s1); \</span></div><div class="line"><a name="l00337"></a><span class="lineno"> 337</span>&#160;<span class="preprocessor"> acc.s2 = mad(src0.s3, weights_row0.s1, acc.s2); \</span></div><div class="line"><a name="l00338"></a><span class="lineno"> 338</span>&#160;<span class="preprocessor"> acc.s3 = mad(src1.s0, weights_row0.s1, acc.s3); \</span></div><div class="line"><a name="l00339"></a><span class="lineno"> 339</span>&#160;<span class="preprocessor"> acc.s0 = mad(src0.s2, weights_row0.s2, acc.s0); \</span></div><div class="line"><a name="l00340"></a><span class="lineno"> 340</span>&#160;<span class="preprocessor"> acc.s1 = mad(src0.s3, weights_row0.s2, acc.s1); \</span></div><div class="line"><a name="l00341"></a><span class="lineno"> 341</span>&#160;<span class="preprocessor"> acc.s2 = mad(src1.s0, weights_row0.s2, acc.s2); \</span></div><div class="line"><a name="l00342"></a><span class="lineno"> 342</span>&#160;<span class="preprocessor"> acc.s3 = mad(src1.s1, weights_row0.s2, acc.s3); \</span></div><div class="line"><a name="l00343"></a><span class="lineno"> 343</span>&#160;<span class="preprocessor"> })</span></div><div class="line"><a name="l00344"></a><span class="lineno"> 344</span>&#160;<span class="comment"></span></div><div class="line"><a name="l00345"></a><span class="lineno"> 345</span>&#160;<span class="comment">/** An optimized direct convolution 3x3 OpenCL kernel for Bifrost architectures when the data type is F32</span></div><div class="line"><a name="l00346"></a><span class="lineno"> 346</span>&#160;<span class="comment"> *</span></div><div class="line"><a name="l00347"></a><span class="lineno"> 347</span>&#160;<span class="comment"> * @note This OpenCL kernel works only with stride_x and stride_y equal to 1</span></div><div class="line"><a name="l00348"></a><span class="lineno"> 348</span>&#160;<span class="comment"> * @note The third dimensions of the weights tensors must be passed at compile time using -DWEIGHTS_DEPTH</span></div><div class="line"><a name="l00349"></a><span class="lineno"> 349</span>&#160;<span class="comment"> * @note In case biases, -DHAS_BIAS must to be passed at compile</span></div><div class="line"><a name="l00350"></a><span class="lineno"> 350</span>&#160;<span class="comment"> *</span></div><div class="line"><a name="l00351"></a><span class="lineno"> 351</span>&#160;<span class="comment"> * @param[in] src_ptr Pointer to the source tensor. Supported data types: F32</span></div><div class="line"><a name="l00352"></a><span class="lineno"> 352</span>&#160;<span class="comment"> * @param[in] src_stride_x Stride of the source tensor in X dimension (in bytes)</span></div><div class="line"><a name="l00353"></a><span class="lineno"> 353</span>&#160;<span class="comment"> * @param[in] src_step_x src_stride_x * number of elements along X processed per workitem(in bytes)</span></div><div class="line"><a name="l00354"></a><span class="lineno"> 354</span>&#160;<span class="comment"> * @param[in] src_stride_y Stride of the source tensor in Y dimension (in bytes)</span></div><div class="line"><a name="l00355"></a><span class="lineno"> 355</span>&#160;<span class="comment"> * @param[in] src_step_y src_stride_y * number of elements along Y processed per workitem(in bytes)</span></div><div class="line"><a name="l00356"></a><span class="lineno"> 356</span>&#160;<span class="comment"> * @param[in] src_stride_z Stride of the source tensor in Z dimension (in bytes)</span></div><div class="line"><a name="l00357"></a><span class="lineno"> 357</span>&#160;<span class="comment"> * @param[in] src_step_z src_stride_z * number of elements along Z processed per workitem(in bytes)</span></div><div class="line"><a name="l00358"></a><span class="lineno"> 358</span>&#160;<span class="comment"> * @param[in] src_offset_first_element_in_bytes The offset of the first element in the source tensor</span></div><div class="line"><a name="l00359"></a><span class="lineno"> 359</span>&#160;<span class="comment"> * @param[out] dst_ptr Pointer to the destination tensor. Supported data types: same as @p src_ptr</span></div><div class="line"><a name="l00360"></a><span class="lineno"> 360</span>&#160;<span class="comment"> * @param[in] dst_stride_x Stride of the destination tensor in X dimension (in bytes)</span></div><div class="line"><a name="l00361"></a><span class="lineno"> 361</span>&#160;<span class="comment"> * @param[in] dst_step_x dst_stride_x * number of elements along X processed per workitem(in bytes)</span></div><div class="line"><a name="l00362"></a><span class="lineno"> 362</span>&#160;<span class="comment"> * @param[in] dst_stride_y Stride of the destination tensor in Y dimension (in bytes)</span></div><div class="line"><a name="l00363"></a><span class="lineno"> 363</span>&#160;<span class="comment"> * @param[in] dst_step_y dst_stride_y * number of elements along Z processed per workitem(in bytes)</span></div><div class="line"><a name="l00364"></a><span class="lineno"> 364</span>&#160;<span class="comment"> * @param[in] dst_stride_z Stride of the destination tensor in Z dimension (in bytes)</span></div><div class="line"><a name="l00365"></a><span class="lineno"> 365</span>&#160;<span class="comment"> * @param[in] dst_step_z dst_stride_z * number of elements along Z processed per workitem(in bytes)</span></div><div class="line"><a name="l00366"></a><span class="lineno"> 366</span>&#160;<span class="comment"> * @param[in] dst_offset_first_element_in_bytes The offset of the first element in the destination tensor</span></div><div class="line"><a name="l00367"></a><span class="lineno"> 367</span>&#160;<span class="comment"> * @param[in] weights_ptr Pointer to the weights tensor. Supported data types: same as @p src_ptr</span></div><div class="line"><a name="l00368"></a><span class="lineno"> 368</span>&#160;<span class="comment"> * @param[in] weights_stride_x Stride of the weights tensor in X dimension (in bytes)</span></div><div class="line"><a name="l00369"></a><span class="lineno"> 369</span>&#160;<span class="comment"> * @param[in] weights_step_x weights_stride_x * number of elements along X processed per workitem(in bytes)</span></div><div class="line"><a name="l00370"></a><span class="lineno"> 370</span>&#160;<span class="comment"> * @param[in] weights_stride_y Stride of the weights tensor in Y dimension (in bytes)</span></div><div class="line"><a name="l00371"></a><span class="lineno"> 371</span>&#160;<span class="comment"> * @param[in] weights_step_y weights_stride_y * number of elements along y processed per workitem(in bytes)</span></div><div class="line"><a name="l00372"></a><span class="lineno"> 372</span>&#160;<span class="comment"> * @param[in] weights_stride_z Stride of the weights tensor in Z dimension (in bytes)</span></div><div class="line"><a name="l00373"></a><span class="lineno"> 373</span>&#160;<span class="comment"> * @param[in] weights_step_z weights_stride_z * number of elements along Z processed per workitem(in bytes)</span></div><div class="line"><a name="l00374"></a><span class="lineno"> 374</span>&#160;<span class="comment"> * @param[in] weights_offset_first_element_in_bytes The offset of the first element in the weights tensor</span></div><div class="line"><a name="l00375"></a><span class="lineno"> 375</span>&#160;<span class="comment"> * @param[in] biases_ptr Pointer to the biases tensor. Same as @p src_ptr</span></div><div class="line"><a name="l00376"></a><span class="lineno"> 376</span>&#160;<span class="comment"> * @param[in] biases_stride_x Stride of the biases tensor in X dimension (in bytes)</span></div><div class="line"><a name="l00377"></a><span class="lineno"> 377</span>&#160;<span class="comment"> * @param[in] biases_step_x biases_stride_x * number of elements along X processed per workitem(in bytes)</span></div><div class="line"><a name="l00378"></a><span class="lineno"> 378</span>&#160;<span class="comment"> * @param[in] biases_offset_first_element_in_bytes The offset of the first element in the biases tensor</span></div><div class="line"><a name="l00379"></a><span class="lineno"> 379</span>&#160;<span class="comment"> * @param[in] weights_stride_w Stride of the weights tensor in the 4th dimension</span></div><div class="line"><a name="l00380"></a><span class="lineno"> 380</span>&#160;<span class="comment"> */</span></div><div class="line"><a name="l00381"></a><span class="lineno"> 381</span>&#160;__kernel <span class="keywordtype">void</span> direct_convolution3x3_f32_bifrost(</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#a989ab3e96426615bb98e04e0235088ca">src</a>),</div><div class="line"><a name="l00383"></a><span class="lineno"> 383</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="l00384"></a><span class="lineno"> 384</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#a64a08a9fec5aeee8650e7182b6d171d0">weights</a>),</div><div class="line"><a name="l00385"></a><span class="lineno"> 385</span>&#160;#ifdef HAS_BIAS</div><div class="line"><a name="l00386"></a><span class="lineno"> 386</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="l00387"></a><span class="lineno"> 387</span>&#160;#endif <span class="comment">/* defined(HAS_BIAS) */</span></div><div class="line"><a name="l00388"></a><span class="lineno"> 388</span>&#160; <span class="keywordtype">unsigned</span> <span class="keywordtype">int</span> weights_stride_w)</div><div class="line"><a name="l00389"></a><span class="lineno"> 389</span>&#160;{</div><div class="line"><a name="l00390"></a><span class="lineno"> 390</span>&#160; <span class="comment">// Get the kernel index</span></div><div class="line"><a name="l00391"></a><span class="lineno"> 391</span>&#160; <span class="keyword">const</span> <span class="keywordtype">int</span> kernel_index = get_global_id(2);</div><div class="line"><a name="l00392"></a><span class="lineno"> 392</span>&#160;</div><div class="line"><a name="l00393"></a><span class="lineno"> 393</span>&#160; <a class="code" href="struct_image.xhtml">Image</a> <a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#a989ab3e96426615bb98e04e0235088ca">src</a> = <a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#aebe814363556c244be043b13e7969197">CONVERT_TO_IMAGE_STRUCT</a>(<a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#a989ab3e96426615bb98e04e0235088ca">src</a>);</div><div class="line"><a name="l00394"></a><span class="lineno"> 394</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="l00395"></a><span class="lineno"> 395</span>&#160;</div><div class="line"><a name="l00396"></a><span class="lineno"> 396</span>&#160; float4 values0 = 0;</div><div class="line"><a name="l00397"></a><span class="lineno"> 397</span>&#160; float4 values1 = 0;</div><div class="line"><a name="l00398"></a><span class="lineno"> 398</span>&#160; float4 values2 = 0;</div><div class="line"><a name="l00399"></a><span class="lineno"> 399</span>&#160;</div><div class="line"><a name="l00400"></a><span class="lineno"> 400</span>&#160; __global uchar *weights_addr = (__global uchar *)(weights_ptr + weights_offset_first_element_in_bytes + kernel_index * weights_stride_w);</div><div class="line"><a name="l00401"></a><span class="lineno"> 401</span>&#160; __global uchar *src_addr = (__global uchar *)<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#a989ab3e96426615bb98e04e0235088ca">src</a>, 0, 0);</div><div class="line"><a name="l00402"></a><span class="lineno"> 402</span>&#160;</div><div class="line"><a name="l00403"></a><span class="lineno"> 403</span>&#160; <span class="comment">// Note: Since each work-item computes 4x3 elements, we need to load 5 rows from the input tensor</span></div><div class="line"><a name="l00404"></a><span class="lineno"> 404</span>&#160;</div><div class="line"><a name="l00405"></a><span class="lineno"> 405</span>&#160; <span class="keywordflow">for</span>(ushort d = 0; d &lt; (ushort)WEIGHTS_DEPTH; ++d)</div><div class="line"><a name="l00406"></a><span class="lineno"> 406</span>&#160; {</div><div class="line"><a name="l00407"></a><span class="lineno"> 407</span>&#160; <span class="comment">// Load the weights</span></div><div class="line"><a name="l00408"></a><span class="lineno"> 408</span>&#160; float3 weights_row0 = vload3(0, (__global <span class="keywordtype">float</span> *)(weights_addr + 0 * weights_stride_y));</div><div class="line"><a name="l00409"></a><span class="lineno"> 409</span>&#160; float3 weights_row1 = vload3(0, (__global <span class="keywordtype">float</span> *)(weights_addr + 1 * weights_stride_y));</div><div class="line"><a name="l00410"></a><span class="lineno"> 410</span>&#160; float3 weights_row2 = vload3(0, (__global <span class="keywordtype">float</span> *)(weights_addr + 2 * weights_stride_y));</div><div class="line"><a name="l00411"></a><span class="lineno"> 411</span>&#160; float4 src0;</div><div class="line"><a name="l00412"></a><span class="lineno"> 412</span>&#160; float2 src1;</div><div class="line"><a name="l00413"></a><span class="lineno"> 413</span>&#160;</div><div class="line"><a name="l00414"></a><span class="lineno"> 414</span>&#160; <span class="comment">// Load values from row0 of input tensor</span></div><div class="line"><a name="l00415"></a><span class="lineno"> 415</span>&#160; src0 = vload4(0, (__global <span class="keywordtype">float</span> *)(src_addr + 0 * src_stride_y));</div><div class="line"><a name="l00416"></a><span class="lineno"> 416</span>&#160; src1 = vload2(0, (__global <span class="keywordtype">float</span> *)(src_addr + 0 * src_stride_y) + 4);</div><div class="line"><a name="l00417"></a><span class="lineno"> 417</span>&#160;</div><div class="line"><a name="l00418"></a><span class="lineno"> 418</span>&#160; CONVOLUTION1x3_BIFROST(values0, src0, src1, weights_row0);</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">// Load values from row1 of input tensor</span></div><div class="line"><a name="l00421"></a><span class="lineno"> 421</span>&#160; src0 = vload4(0, (__global <span class="keywordtype">float</span> *)(src_addr + 1 * src_stride_y));</div><div class="line"><a name="l00422"></a><span class="lineno"> 422</span>&#160; src1 = vload2(0, (__global <span class="keywordtype">float</span> *)(src_addr + 1 * src_stride_y) + 4);</div><div class="line"><a name="l00423"></a><span class="lineno"> 423</span>&#160;</div><div class="line"><a name="l00424"></a><span class="lineno"> 424</span>&#160; <span class="comment">// Accumulate</span></div><div class="line"><a name="l00425"></a><span class="lineno"> 425</span>&#160; CONVOLUTION1x3_BIFROST(values0, src0, src1, weights_row1);</div><div class="line"><a name="l00426"></a><span class="lineno"> 426</span>&#160; CONVOLUTION1x3_BIFROST(values1, src0, src1, weights_row0);</div><div class="line"><a name="l00427"></a><span class="lineno"> 427</span>&#160;</div><div class="line"><a name="l00428"></a><span class="lineno"> 428</span>&#160; <span class="comment">// Load values from row2 of input tensor</span></div><div class="line"><a name="l00429"></a><span class="lineno"> 429</span>&#160; src0 = vload4(0, (__global <span class="keywordtype">float</span> *)(src_addr + 2 * src_stride_y));</div><div class="line"><a name="l00430"></a><span class="lineno"> 430</span>&#160; src1 = vload2(0, (__global <span class="keywordtype">float</span> *)(src_addr + 2 * src_stride_y) + 4);</div><div class="line"><a name="l00431"></a><span class="lineno"> 431</span>&#160;</div><div class="line"><a name="l00432"></a><span class="lineno"> 432</span>&#160; <span class="comment">// Accumulate</span></div><div class="line"><a name="l00433"></a><span class="lineno"> 433</span>&#160; CONVOLUTION1x3_BIFROST(values0, src0, src1, weights_row2);</div><div class="line"><a name="l00434"></a><span class="lineno"> 434</span>&#160; CONVOLUTION1x3_BIFROST(values1, src0, src1, weights_row1);</div><div class="line"><a name="l00435"></a><span class="lineno"> 435</span>&#160; CONVOLUTION1x3_BIFROST(values2, src0, src1, weights_row0);</div><div class="line"><a name="l00436"></a><span class="lineno"> 436</span>&#160;</div><div class="line"><a name="l00437"></a><span class="lineno"> 437</span>&#160; <span class="comment">// Load values from row3 of input tensor</span></div><div class="line"><a name="l00438"></a><span class="lineno"> 438</span>&#160; src0 = vload4(0, (__global <span class="keywordtype">float</span> *)(src_addr + 3 * src_stride_y));</div><div class="line"><a name="l00439"></a><span class="lineno"> 439</span>&#160; src1 = vload2(0, (__global <span class="keywordtype">float</span> *)(src_addr + 3 * src_stride_y) + 4);</div><div class="line"><a name="l00440"></a><span class="lineno"> 440</span>&#160;</div><div class="line"><a name="l00441"></a><span class="lineno"> 441</span>&#160; <span class="comment">// Accumulate</span></div><div class="line"><a name="l00442"></a><span class="lineno"> 442</span>&#160; CONVOLUTION1x3_BIFROST(values1, src0, src1, weights_row2);</div><div class="line"><a name="l00443"></a><span class="lineno"> 443</span>&#160; CONVOLUTION1x3_BIFROST(values2, src0, src1, weights_row1);</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">// Row4</span></div><div class="line"><a name="l00446"></a><span class="lineno"> 446</span>&#160; src0 = vload4(0, (__global <span class="keywordtype">float</span> *)(src_addr + 4 * src_stride_y));</div><div class="line"><a name="l00447"></a><span class="lineno"> 447</span>&#160; src1 = vload2(0, (__global <span class="keywordtype">float</span> *)(src_addr + 4 * src_stride_y) + 4);</div><div class="line"><a name="l00448"></a><span class="lineno"> 448</span>&#160;</div><div class="line"><a name="l00449"></a><span class="lineno"> 449</span>&#160; <span class="comment">// Accumulate</span></div><div class="line"><a name="l00450"></a><span class="lineno"> 450</span>&#160; CONVOLUTION1x3_BIFROST(values2, src0, src1, weights_row2);</div><div class="line"><a name="l00451"></a><span class="lineno"> 451</span>&#160;</div><div class="line"><a name="l00452"></a><span class="lineno"> 452</span>&#160; src_addr += src_stride_z;</div><div class="line"><a name="l00453"></a><span class="lineno"> 453</span>&#160; weights_addr += weights_stride_z;</div><div class="line"><a name="l00454"></a><span class="lineno"> 454</span>&#160; }</div><div class="line"><a name="l00455"></a><span class="lineno"> 455</span>&#160;</div><div class="line"><a name="l00456"></a><span class="lineno"> 456</span>&#160;<span class="preprocessor">#ifdef HAS_BIAS</span></div><div class="line"><a name="l00457"></a><span class="lineno"> 457</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#a64d779f80eeb923e0ab2313433f7b40b">CONVERT_TO_VECTOR_STRUCT_NO_STEP</a>(biases);</div><div class="line"><a name="l00458"></a><span class="lineno"> 458</span>&#160;</div><div class="line"><a name="l00459"></a><span class="lineno"> 459</span>&#160; <span class="keywordtype">float</span> <a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#a3a77be8aebd8e00522b32061d46ccdbd">bias</a> = (float) * ((__global <span class="keywordtype">float</span> *)(<a class="code" href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a7e4940407322d6f0ccb8b6b86b856019">vector_offset</a>(&amp;biases, kernel_index)));</div><div class="line"><a name="l00460"></a><span class="lineno"> 460</span>&#160;</div><div class="line"><a name="l00461"></a><span class="lineno"> 461</span>&#160; values0 += (float4)<a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#a3a77be8aebd8e00522b32061d46ccdbd">bias</a>;</div><div class="line"><a name="l00462"></a><span class="lineno"> 462</span>&#160; values1 += (float4)<a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#a3a77be8aebd8e00522b32061d46ccdbd">bias</a>;</div><div class="line"><a name="l00463"></a><span class="lineno"> 463</span>&#160; values2 += (float4)<a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#a3a77be8aebd8e00522b32061d46ccdbd">bias</a>;</div><div class="line"><a name="l00464"></a><span class="lineno"> 464</span>&#160;<span class="preprocessor">#endif </span><span class="comment">/* defined(HAS_BIAS) */</span><span class="preprocessor"></span></div><div class="line"><a name="l00465"></a><span class="lineno"> 465</span>&#160;</div><div class="line"><a name="l00466"></a><span class="lineno"> 466</span>&#160; vstore4(values0, 0, (__global <span class="keywordtype">float</span> *)(<a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#adbf67dcee294e673cf796f1ed8aeb6a4">dst</a>.ptr + 0 * dst_stride_y));</div><div class="line"><a name="l00467"></a><span class="lineno"> 467</span>&#160; vstore4(values1, 0, (__global <span class="keywordtype">float</span> *)(<a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#adbf67dcee294e673cf796f1ed8aeb6a4">dst</a>.ptr + 1 * dst_stride_y));</div><div class="line"><a name="l00468"></a><span class="lineno"> 468</span>&#160; vstore4(values2, 0, (__global <span class="keywordtype">float</span> *)(<a class="code" href="namespacearm__compute_1_1test_1_1validation.xhtml#adbf67dcee294e673cf796f1ed8aeb6a4">dst</a>.ptr + 2 * dst_stride_y));</div><div class="line"><a name="l00469"></a><span class="lineno"> 469</span>&#160;}</div><div class="line"><a name="l00470"></a><span class="lineno"> 470</span>&#160;<span class="preprocessor">#endif // defined(WEIGHTS_DEPTH)</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#l00341">helpers.h:341</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#l00510">helpers.h:510</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#l00311">helpers.h:311</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_1_1test_1_1validation_xhtml_a0886942393a3ba0dfefaa7516b159784"><div class="ttname"><a href="namespacearm__compute_1_1test_1_1validation.xhtml#a0886942393a3ba0dfefaa7516b159784">arm_compute::test::validation::for</a></div><div class="ttdeci">for(size_t k=0;k&lt; _target.size();++k)</div><div class="ttdef"><b>Definition:</b> <a href="_c_l_2_unstack_8cpp_source.xhtml#l00091">Unstack.cpp:91</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#l00358">helpers.h:358</a></div></div>
<div class="ttc" id="src_2core_2_c_l_2cl__kernels_2_helpers_8h_xhtml_a79e8e562daa6599317d2d1cd86ef1bf2"><div class="ttname"><a href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a79e8e562daa6599317d2d1cd86ef1bf2">CONVERT_TO_TENSOR3D_STRUCT_NO_STEP</a></div><div class="ttdeci">#define CONVERT_TO_TENSOR3D_STRUCT_NO_STEP(name)</div><div class="ttdef"><b>Definition:</b> <a href="src_2core_2_c_l_2cl__kernels_2_helpers_8h_source.xhtml#l00330">helpers.h:330</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#l00269">helpers.h:269</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_a3a77be8aebd8e00522b32061d46ccdbd"><div class="ttname"><a href="namespacearm__compute_1_1test_1_1validation.xhtml#a3a77be8aebd8e00522b32061d46ccdbd">arm_compute::test::validation::bias</a></div><div class="ttdeci">CLTensor bias</div><div class="ttdef"><b>Definition:</b> <a href="_c_l_2_convolution_layer_8cpp_source.xhtml#l00189">ConvolutionLayer.cpp:189</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#l00349">helpers.h:349</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#l00326">helpers.h:326</a></div></div>
<div class="ttc" id="src_2core_2_c_l_2cl__kernels_2_helpers_8h_xhtml_a7e4940407322d6f0ccb8b6b86b856019"><div class="ttname"><a href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a7e4940407322d6f0ccb8b6b86b856019">vector_offset</a></div><div class="ttdeci">__global const uchar * vector_offset(const Vector *vec, int x)</div><div class="ttdoc">Get the pointer position of a Vector.</div><div class="ttdef"><b>Definition:</b> <a href="src_2core_2_c_l_2cl__kernels_2_helpers_8h_source.xhtml#l00499">helpers.h:499</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="namespacearm__compute_1_1test_1_1validation_xhtml_a64a08a9fec5aeee8650e7182b6d171d0"><div class="ttname"><a href="namespacearm__compute_1_1test_1_1validation.xhtml#a64a08a9fec5aeee8650e7182b6d171d0">arm_compute::test::validation::weights</a></div><div class="ttdeci">CLTensor weights</div><div class="ttdef"><b>Definition:</b> <a href="_c_l_2_convolution_layer_8cpp_source.xhtml#l00188">ConvolutionLayer.cpp:188</a></div></div>
<div class="ttc" id="direct__convolution3x3_8cl_xhtml_a1f15728672380ade7a238f5e783d54d2"><div class="ttname"><a href="direct__convolution3x3_8cl.xhtml#a1f15728672380ade7a238f5e783d54d2">CONVERT_SAT</a></div><div class="ttdeci">#define CONVERT_SAT(a, b)</div><div class="ttdef"><b>Definition:</b> <a href="direct__convolution3x3_8cl_source.xhtml#l00030">direct_convolution3x3.cl:30</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#l00283">helpers.h:283</a></div></div>
<div class="ttc" id="src_2core_2_c_l_2cl__kernels_2_helpers_8h_xhtml_a64d779f80eeb923e0ab2313433f7b40b"><div class="ttname"><a href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a64d779f80eeb923e0ab2313433f7b40b">CONVERT_TO_VECTOR_STRUCT_NO_STEP</a></div><div class="ttdeci">#define CONVERT_TO_VECTOR_STRUCT_NO_STEP(name)</div><div class="ttdef"><b>Definition:</b> <a href="src_2core_2_c_l_2cl__kernels_2_helpers_8h_source.xhtml#l00308">helpers.h:308</a></div></div>
<div class="ttc" id="direct__convolution3x3_8cl_xhtml_aebbeb1f22eca3a3f4c3e019e8f419f39"><div class="ttname"><a href="direct__convolution3x3_8cl.xhtml#aebbeb1f22eca3a3f4c3e019e8f419f39">ADD_OP</a></div><div class="ttdeci">#define ADD_OP(a, b)</div><div class="ttdef"><b>Definition:</b> <a href="direct__convolution3x3_8cl_source.xhtml#l00028">direct_convolution3x3.cl:28</a></div></div>
<div class="ttc" id="src_2core_2_c_l_2cl__kernels_2_helpers_8h_xhtml_a2101b2fe0193ce227ae4e0945e321d85"><div class="ttname"><a href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a2101b2fe0193ce227ae4e0945e321d85">tensor3D_offset</a></div><div class="ttdeci">__global const uchar * tensor3D_offset(const Tensor3D *tensor, int x, int y, int z)</div><div class="ttdoc">Get the pointer position of a Tensor3D.</div><div class="ttdef"><b>Definition:</b> <a href="src_2core_2_c_l_2cl__kernels_2_helpers_8h_source.xhtml#l00522">helpers.h:522</a></div></div>
<div class="ttc" id="namespacearm__compute_1_1test_1_1validation_xhtml_a989ab3e96426615bb98e04e0235088ca"><div class="ttname"><a href="namespacearm__compute_1_1test_1_1validation.xhtml#a989ab3e96426615bb98e04e0235088ca">arm_compute::test::validation::src</a></div><div class="ttdeci">cast configure &amp; src</div><div class="ttdef"><b>Definition:</b> <a href="_c_l_2_cast_8cpp_source.xhtml#l00169">Cast.cpp:169</a></div></div>
<div class="ttc" id="src_2core_2_c_l_2cl__kernels_2_helpers_8h_xhtml_a36f754c05b6fddf6df0d8d0a74f8159f"><div class="ttname"><a href="src_2core_2_c_l_2cl__kernels_2_helpers_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a></div><div class="ttdeci">#define VEC_DATA_TYPE(type, size)</div><div class="ttdef"><b>Definition:</b> <a href="src_2core_2_c_l_2cl__kernels_2_helpers_8h_source.xhtml#l00255">helpers.h:255</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="direct__convolution3x3_8cl.xhtml">direct_convolution3x3.cl</a></li>
<li class="footer">Generated on Thu Mar 5 2020 16:06:57 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>